JP6545697B2 - Monitoring structural features of cerebral blood flow velocity for the diagnosis of neurological conditions - Google Patents

Description

Description of copyright

  [0001] A portion of the disclosure of this patent application document includes material that is subject to copyright protection, including the drawings. The copyright holder does not dispute the patent documents or patent disclosures appearing in the Patent and Trademark Office files or records, but otherwise reserves all copyrights.

  [0002] The present disclosure relates to the field of physiological monitoring, in particular to monitor physiological functions of the brain, including intracranial pressure, cerebral blood flow velocity, cerebral blood flow, and cerebral circulation reserve. Involved. Acquisition of physiological signals is performed by automated ultrasound devices to increase accuracy and reliability.

  [0003] Including mild and severe traumatic brain injury (traumatic brain injury: TBI), stroke or subarachnoid hemorrhage (SAH), cerebral malaria (CM), pseudo brain tumor, and brain tumor. Neurological conditions affect millions of people worldwide each year. One specific physiological parameter of interest is intracranial pressure (ICP), usually defined as the pressure in cerebrospinal fluid (CSF) in the cerebral ventricles of the brain, which is a rise in ICP. Because timely detection is required for treatment to avoid severe deterioration including cerebral ischemia and herniation, it is an important parameter for managing patients with brain injury. Unfortunately, currently available techniques for monitoring ICP and managing patients at risk for acute ICP elevation are invasive. For example, one way to monitor intracranial pressure in the skull is to use an intracerebroventricular catheter, which is inserted through a hole drilled through the skull and inserted into the lateral ventricle. Another invasive technique is to use a hollow subdural screw, again inserted through the hole in the skull and placed through a membrane (dura) that protects the brain and brain. is there. Finally, a third invasive method is to insert an epidural sensor between the skull and the dural tissue.

  [0004] The invasive nature of ICP measurement eliminates its use in many clinical situations where ICP measurement is of important diagnostic and precursor utility because of the great risk of infection and secondary bleeding. One example is the management of patients with acute liver failure. As coagulation abnormalities (bleeding disorders) are customary among patients with acute liver failure, the invasive ICP, despite the important potential consequences of foresight based on measurement of elevated ICP, The risks associated with monitoring prevent its use. Another example is the diagnosis of idiopathic intracranial hypertension (IIH), also called a pseudobrain tumor, which would benefit from direct ICP measurements. Still, these measurements are rarely done due to the associated risks and complications of invasive ICP. Finally, CM will benefit from ICP monitoring, but is an area of research limitation where malaria is normal, so it is another example of a condition that is technically unworkable.

  [0005] While several attempts have been made to identify reliable non-invasive ICP monitoring techniques to meet these important unmet needs, none of these attempts have It showed no significant clinical applicability. In addition, several groups have been proposed for non-invasive evaluation of ICP, such as systolic velocity, diastolic velocity, mean flow velocity, pulsatility index (PI), blood flow velocity of cerebral resistance such as the coefficient of vascular resistance (cerebral A few simple indicators of blood flow velocity (CBFV) were proposed. However, it remains controversial whether these simple indicators can give reliable and accurate information about ICP.

  [0006] In recognition of the limitations of current non-invasive ICP assessment techniques, systems and methods for increased ICP or intracranial hypertension (IH) detection may provide important benefits for patients and clinicians .

  [0007] IIH is characterized by increased ICP of unknown cause and is relatively common to obese women. Management of IIH patients in the United States is estimated to cost $ 444 million annually. Currently, IIH patients are treated with weight loss, medical therapy and surgery. Treatment decisions are often based on subjective symptoms, the presence and severity of papillary edema, and invasive tests such as lumbar puncture. Given the possibility of subjective symptom change and, if optic atrophy has begun, papillary edema may appear to be improved despite the aggravating disease, non-invasive IH diagnostic tools are: By allowing real-time measurement of ICP and clinical correlation of symptoms and changes in symptoms, treatment decisions can be simplified. By enabling earlier detection of changes in ICP, the therapeutic effect of the patient can be improved, leading to more effective interventions to save the vision against the deteriorating disease.

  [0008] A related but distinct other physiological disadvantage is the defect caused by mild TBI with no apparent increase in ICP but with remaining changes in hidden physiology (demerits in cerebral circulation reserve) It is. Historically, many of the studies on mild TBI have focused on the neurologic and neuropsychological therapeutic effects of injury. Current diagnostic and return-to-play guidelines include Post-Con Symptom Scale (PCSS), Graded Symptom Checklist (GSC), and Standardized Concussion The results are based largely on the results of neuropsychological tests based on the patient's symptoms such as Standardized Assessment of Concussion (SAC) and Immediate Post-Concussion Assessment and Cognitive Testing (ImPACT). It is based. However, there is an unquestionable need to compensate for these neurological tests in a way that takes into account the pathophysiology of mild TBI. Recent reviews include structural imaging (MRI, CT), diffusion tensor imaging, single photon emission tomography, positron emission tomography, functional MRI, near infrared spectroscopy, electroencephalography, MEG, heart rate variability, and An overview of some pathophysiology-based methods for monitoring mTBI, such as blood markers, is given. However, the review emphasizes that most of these methods are in the early stages of the study and that none have received clinical approval.

  [0009] As mentioned above, the pathological increase in ICP does not appear in mild TBI, thus requiring additional physiological parameters to be evaluated. Several studies have identified possible underlying causes of possible loss in CBF, some of which are physiological defects, and to identify changes in brain hemodynamics following mild TBI. One of the relevant aspects of CBF is the description of cerebral circulation reserve, the extent of cerebral perfusion change from baseline. Changes in this range of irritated brain perfusion help diagnose for several different conditions, including severe TBI, migraine headaches, long-term space flight, blows, and carotid artery stenosis / It can be a omen. Cerebral circulation reserve can be assessed using non-invasive techniques including transcranial doppler, and thus benefit from the advanced framework intended in this study.

  [0010] So far, traditional analysis of CBFV obtained using transcranial doppler (TCD) has proved to be unsuitable in the diagnosis of neurological conditions such as TBI and SAH. In recognition of the limitations of current approaches to diagnosing TBI, therefore, improved systems and methods for the diagnosis of TBI and other neurological conditions are desired.

  [0011] The systems and techniques described herein include transcranial doppler (TCD), intracranial hypertension (IH), and various neurological conditions including mild / moderate TBI. Using a system for combining and extracting structural features, partially using a database of CBFV pulses previously validated for classification, using one or more blood vessels feeding the brain Including the collection of CBFV data.

  [0012] The systems and methods described herein include non-invasive diagnostic tools for IH based on structural analysis of CBFV waveforms measured via TCD. The performance of these systems and methods is validated by two types of classification methods: supervised learning, which takes the traditional 'pre-given data' as an example and learns it as a guide Based on the "learning" approach, the other is the "semisupervised learning" approach that learns from mixed data with and without labels. Our simulation results show that the prediction accuracy of the semi-supervised learning IH detection (area under the curve) can be as high as 92%, while the prediction accuracy of the supervised learning IH detection method Is only about 82%. It should be noted that the prediction accuracy based on the traditional TCD embodiment (pulsatility index (PI)) based IH detection method is as low as 59%.

  [0013] TCD measurements may include CBFV from one or more blood vessels in the head and neck. For example, measurement. It may be obtained from the middle cerebral artery (MCA), the internal carotid artery (ICA), and / or the basilar artery (BA), or a combination thereof.

  [0014] In addition to the lack of accuracy of TCD caused by the limited set of features, inter- and intra-observer variation is plaguing TCD adoption. To improve the reliability of our morphological framework, we have also introduced a fully automated headset for TCD signal acquisition.

  [0015] In an embodiment, systems, devices, and methods include methods for non-invasively detecting IH. In one approach, the method detects at least one valid pulse by detecting individual CBFV waveform pulses from consecutive CBFV segments, grouping detected pulses, and utilizing a CBFV pulse library. Recognize and Including constructing a representative pulse from the group, extracting over 100 structural features from the representative pulse, and using the classification framework for ICP.

  [0016] In one approach, CBFV waveform segments are associated with ECG segments that are recorded simultaneously. The method may further include identifying structural features that include sub-peaks of the constructed representative pulse. The method may include calculating a representative index of the constructed representative pulse. For example, the size of the subpeak may be used to characterize the ICP as normal or IH.

  [0017] In an embodiment, the systems and methods described herein include utilizing spectral regression to cluster detected CBFV pulses. The method may include constructing a graph by defining correct node connections. In one approach, the composition of the graph is weighted. In one embodiment, the method comprises decomposing eigenvectors. In one approach, normalized least squares are solved for at least one eigenvector. In one embodiment, spectral regression comprises kernel discriminant analysis. The systems and methods described herein provide for making curve analysis decisions by quantifying the accuracy of predictions that make use of area features under the curve. In one approach, intracranial pressure pulses are divided into three groups: normal (<14 mm Hg), gray zone (15-30 mm Hg), and IH (> 30 mm Hg).

  [0018] In an embodiment, the systems and methods described can be used for mild and moderate diagnostics without an increase in ICP. Our framework extends the CBFV analysis from this basic method to the 100 distinct structural features present in the waveform, thereby quantifying minor changes in the waveform and providing greater diagnostic and predictive accuracy give. A distinct advantage of our approach is that TCD-based devices are low cost, secure and portable, and they have been shown to be effective prior to hospital arrival.

  [0019] These and other embodiments are described in more detail herein. Changes and modifications to these embodiments will occur to those skilled in the art after reviewing this disclosure. The features and aspects described above may be implemented in any and all combinations (including multiple combinations and subcombinations) of one or more of the features described herein. The various features described or illustrated above, including any of their components, may be combined or integrated with other systems. Furthermore, certain structural features may be omitted or not implemented.

  The above and other issues and advantages will become apparent upon consideration of the following detailed description taken in conjunction with the accompanying drawings, in which like reference numerals refer to like parts in the drawings Point to.

[0021] A cerebral blood flow velocity (CBFV) obtained from a TCD unit, and the maximum velocity envelope is shown in white. [0022] FIG. 1 is a flowchart of the entire algorithm using multiple blood vessels collected using TCD from head and neck. [0023] FIG. 7 is a block diagram of a structural feature extraction process showing a continuous CBFV input waveform converted into one representative output CBFV pulse with three sub-peaks. [0024] Kim, S. In a plot taken from "Noninvasive intracranial hypertension detection utilizing semisupervised learning. IEEE Trans Biomed Eng, 2013. 60 (4): p. 1126-33)" by C. et al., CBFV waveform associated with various average ICP values An example is shown: upper row (normal) and lower row (hypertension), black dots represent three sub-peaks, CBFV waveform associated with lower average ICP value (mICP in mmHG) is higher average ICP pulse The waveform y associated with Tc also tends to have more distinct sub-peaks, and the difference between the magnitudes of the second and third sub-peaks is more normal in the CBFV waveform associated with higher average ICP pulses. Larger than in the waveform associated with the average ICP pulse. [0025] A plot taken from the literature by Kim, S et al., Showing AUC vs. close neighbors (k), each line and gray area represents average AUC and multiple (= 100) 10-fold cross validation It represents one standard deviation displacement above. [0026] Kim, S. A plot taken from M. et al.'S literature, showing a threshold p _t of overall net benefit versus disease potential, black solid line for the Treat-All approach, black The dotted line relates to an approach of treating no one (Treat-None). [0027] Kim, S. Et al., Plots of continuous scale label estimation of gray zone samples versus corresponding ICP values as a result of the second cross validation test, the correlation coefficient between them is 2e-4. The P value was 0.55. [0028] Kim, S. Et al., A plot taken from the literature and shows the ROC curve of a semi-supervised learning ²⁰⁰ IH detection method with three different operating points, where the red point is the Youden index with P _a = 0.12 relates to optimal accuracy of the operating point based on, the green dot relates to optimal net for P _{t =} 0.2, the point of blue according to the optimal net for P _{t =} 0.4. [0029] An example of a main artery in cerebral blood circulation and a cerebral artery ring are shown. [0030] FIG. 1 is a front view of a portable transcranial doppler device, where the portable device operates with either hand, the screen is aligned in a predetermined direction, and the ultrasound probe is magnetically stored on the back. [0031] Fig. 1 1 is a rear view of a portable transcranial doppler (TCD) device, wherein an ultrasound probe is shown in the left housing. [0032] Shown is an automated TCD headset design, the display is shown on the front of the device, two ultrasound probes are included in the side unit of the device, through the temporal window to the known database of vessel locations Automatic positioning of MCA, ACA and PCA based on the compensated robot system. 6 is an example of an exemplary TCD head debt on a patient's skull. 6 is an example of an exemplary TCD head debt on a patient's skull. [0033] FIG. 6A is a side view of another exemplary TCD headset worn by a patient, including a scanner that has a strap around the head and that reciprocates. [0034] The outer housing is spoofed to visualize the internal components of the headset, and is anchored to the side of the patient's head to indicate that it can be adjusted to different sizes of the patient FIG. 1 is a perspective view of a TCD headset. The outer housing to the contour of the wearer's head is shown. [0035] FIG. 15 is a side view of the TCD headset of FIGS. 14A and 14B. The headset is shown relative to the wearer's head profile to visualize the components of the headset.

Detailed description

  [0036] In order to provide a thorough understanding of the systems, devices, and methods described herein, certain example embodiments are described. The embodiments and features described herein are specifically described for use in connection with monitoring intracranial pressure using a transcranial Doppler (TCD) system, but The outlined method, components, mechanism, adjustable system, manufacturing method, and all other features may be combined with one another in any suitable manner, and other types of non-types including MRI and CT. It is understood that it can be adapted and adapted to other physiological and non-physiological monitoring including mild and moderate TBI using physiological monitoring of insult.

  [0037] The term "non-invasive" relates to methods of surgery or physiological monitoring that do not require any type of stab wound. As mentioned above, in addition to transcranial doppler (TCD) systems, MRI systems, CT scanners, pressure transducers, optical imaging devices, near infrared imaging devices, and other such devices are possible sources of raw data. It should be considered that the application is limited only by the appended claims.

  [0038] The present application describes systems and methods for non-invasive collection of raw brain blood flow velocity (CBFV) from one or more blood vessels that feed the brain, as well as structural features in the CBFV waveform. Techniques to identify and extract those features for analysis are described. In this sense, "structural features" refers to identifiable features of the measured CBFV waveform (e.g., subpeaks, subtroughs, landmarks). As explained, these structural features can then be compared to previously identified reference data to classify structural features and to recommend diagnostic results.

  [0039] The systems and methods described herein include in the head and neck including the middle cerebral artery forming the cerebral artery ring, the internal carotid artery, the basilar artery, the vertebral artery, the anterior cerebral artery, and other blood vessels. Provided is a non-invasive IH detection method based on TCD measurement of CBFV in one or more blood vessels. These systems and methods are further enabled and clarified through examples using various learning / classification algorithms.

[0040] For convenience, the following abbreviations are used throughout the text and description contained herein.
aSAH-Aneurysmal subarachnoid hemorrhage
ACA-anterior cerebral artery (anterior cerebral artery)
ACU-area under the curve
BA-basilar artery (basilar artery)
CBFV-cerebral blood flow velocity
ECG-electrocardiogram
ICA-Internal Carotid Artery
ICP-intracranial pressure
IH-intracranial hypertension
IIH-Idiopathic intracranial hypertension
MCA-middle cerebral artery (middle cerebral artery)
mTBI-mild traumatic brain injury
NPH-normal pressure hydrocephalus
PI-Pulsatility index
ROC-receiver operating characteristic
SRKDA-spectral regression kernel discriminant analysis (spectral regression kernel discriminant analysis)
TBI-traumatic brain injury
TCD-transcranial doppler
[0041] The systems and methods described herein are an advanced, understandable structure of CBFV waveforms to establish an alternative diagnostic method for non-invasive ICP assessment and mild / moderate TBI. Dynamic feature analysis.

  [0042] IH detection is a classification problem that distinguishes patients with high ICP from those with normal (non-pathetic) ICP. The conventional approach to such classification problems is to use only labeled samples to train a given classification indicator, called supervised learning. The main problem with this approach is that unlabeled samples are not available even when useful information learned from unlabeled samples results in an improvement in classification accuracy. Samples that are not labeled have the high cost or labor intensity of labeling all samples, or the ambiguity of giving a binary label as in IH detection or mild TBI / shake diagnosis It can exist for various reasons. For example, a simple approach is to label the CBFV waveform as IH when the corresponding ICP is above the widely accepted threshold of 20 mm Hg to regard the ICP as high, and build a classification index Use supervised learning algorithms to do this. This simple paradigm uses 20 mm Hg as the threshold, since 20 mm Hg ICP levels do not represent high levels (false positives) for certain patient classifications, and 20 mm Hg thresholds deviate from IH diagnosis for other patients. It is too inflexible detection of the true IH state that relies strictly on the validity of the. However, choosing different thresholds is not an easy task. If the threshold is too high or too low, there is a risk of failing the IH diagnosis or producing too many false positives.

  [0043] To address this ambiguity in labeling samples, the systems and methods described herein utilize a semi-supervised learning classification approach. The semi-supervised learning approach does not require labeling of all samples, as classification indices can be trained with labeled and unlabeled samples. In one approach, semi-supervised learning techniques in the systems and methods described herein are generated by generative models, self-training, co-training, and transductive support vector machines. machines), including graph based methods. In one approach, conventional regression techniques overcome some of the problems of conventional graph-based semi-supervised learning techniques.

  [0044] In one approach, the systems and methods described herein are implemented using processing circuitry. As described herein, the processing circuitry may be microcontroller, integrated circuit, application specific integrated circuit (ASIC), programmable logic device, field programmable gate array (FPGA), digital signal processor, application specific instructions It is understood to mean a circuit comprising one or more of a set processor (ASIP) or any other suitable digital or analog processor. This processing circuit may be utilized as part of a computer, mobile device, television, tablet, TCD monitoring system, ECG monitoring system, wearable, or any other user system including any suitable device, without limitation. Processing circuitry may be used to execute data and signal processing algorithms as described herein. Processing circuitry may be used to send and receive data, commands, user input, or to other network devices, including networked systems and devices, or from devices.

  The processing circuitry may be connected to electronic storage or memory. Electronic storage, as used herein, includes but is not limited to RAM, ROM, EPROM, EEPROM (registered trademark), flash memory or other semiconductor memory technology, CD-ROM, DVD, or Any suitable, including other optical storage, magnetic storage devices, or any other physical or physical medium for storing desired information, data, instructions, software, firmware, drivers or codes. May include readable memory media. For example, the storage may be software instructions or machine code for controlling processing circuit inputs, outputs, or other processes such as performing the methods and system algorithms and other processing steps described therein. May be included.

  [0046] Processing circuitry may be part of a system that includes devices for interfacing with a user, such as a display or user interface. For example, the display may be any suitable display interface, including, but not limited to, a monitor, a television, an LED display, an LCD display, a projection, a mobile device, a handset, or any other suitable display system. . The user input interface may be a keyboard, touch screen, mouse, microphone, stylus, voice input, or any other suitable user input interface. The display and user input interface allow the processing circuitry to provide information to the user and receive user generated commands, responses, and data. In one approach, the systems and methods described herein include actuators, sensors, and / or transducers. For example, biomedical electrodes and Doppler transducers may be included.

  [0047] In one aspect, over 100 structural features of the CBFV waveform are extracted from the raw CBFV signal collected by TCD. In one approach, these structural feature algorithms are implemented by processing circuitry. The systems and methods described herein further develop and apply these techniques specifically for non-invasive ICP evaluation from TCD-based CBFV and / or ECG waveforms for detection of IH.

  [0048] FIG. 2 shows a block diagram of a structural feature algorithm. After obtaining the raw data, there is a three step process: structural feature extraction, classification and results / diagnosis. The input to the system varies based on the number of blood vessels, but at least one intracranial blood vessel is required. Also, the ground truth for classification (reference data) is determined by neurological status (mild TBI, severe TBI, stroke, etc.).

  First, individual CBFV pulses are extracted from consecutive CBFV segments in association with simultaneously recorded ECG segments. FIG. 3 is a block diagram of a structural feature extraction process showing a continuous CBFV input waveform that is converted into one representative output CBFV pulse with three sub-peaks. The inset to the right shows a schematic representative pulse from a CBFV waveform with six landmarks (three peaks and three valley points). The maximum velocity envelope shown in FIG. 1 is the input to the block diagram. Identification of six landmarks is essential for structural feature extraction.

  [0050] In one approach, a series of individual CBFV pulses are grouped into groups based on correlation coefficients. In one approach, groups of pulses are identified via principal component analysis, correspondence analysis, matrix decomposition, spectral analysis, independent component analysis, or other waveform signal processing methods. The group representative pulse is the average of the largest subgroup identified by the number of pulses in the cluster or group. Representative pulses may be identified through the average of the pulses for the largest subgroup. After constructing the representative pulse, the pulse is verified against the previously validated set of CBFV pulses. The CBFV pulse library may include data sets from many patients / subjects and representative pulses. In one embodiment, the pulse library comprises at least 100 CBFV pulses. In one embodiment, the pulse library comprises at least 10000 or more CBFV pulses.

  [0051] Next, the representative pulse is used for further quantification and diagnosis. In one embodiment, three subpeaks of the representative pulse are designated among the several peak candidates. The inset in FIG. 3 illustrates a typical representative pulse with six landmarks {P1, P2, P3, V1, V2, V3} including three sub-peaks and three sub-trophs. In one embodiment, the position of the peak is found using the recess of the pulse curve according to the four possible definitions in the illustrated embodiment. In the first definition, when the first derivative of the recess is larger than zero, the intersection with the convex region of the recess is treated as a peak, otherwise the intersection with the lateral region of the convex region is the peak. The second definition is based on the curvature of the signal so that the peak is the position with maximum absolute curvature within each recess area, and the third and fourth definitions both refer to the two endpoints of the recess area Includes a straight line to link. According to the third and fourth definitions, a peak can be found at the position where the perpendicular distance or vertical distance from the CBFV to this line is respectively a maximum. Typically, the peak corresponds to a cross to the recess area of the protrusion on the rising edge of the CBFV, or a cross to the protrusion area of the recess on the falling edge of the pulse. This detection process generates a pool of N peak candidates (a1, a2,..., AN). Additionally or alternatively, peak detection and assignment may be assigned using a regression analysis such as spectral regression analysis or multiple linear regression analysis.

  [0052] In one embodiment, structural features (ie, subpeaks, subtrophs, landmarks) may further include ICP status or other neurological status or neurological indicator (Cerebrovascular reactivity, autoregulation, and It is characterized via metrics used to identify the temporal and spatial relationship between neural activity and blood circulation. In one approach, total structural metrics greater than 100 can be extracted from representative pulses in relation to subpeaks and other structural features. These metrics include delay time, magnitude, curvature, slope, and the ratio between subpeaks. In one embodiment, metrics between about 1 and about 10 are extracted. In one approach, at least 10 metrics are extracted. In one approach, between about 10 and about 50 metrics are extracted. In one approach, between about 50 and about 100 metrics are extracted. In one approach, at least 100 metrics are extracted. In one approach, structural metrics greater than 100 are extracted.

  [0053] Typical TCD-based CBFV waveforms are mainly triphasic and were not previously known. The plots in FIG. 4 show typical CBFV waveforms associated with different mean ICP values (mICP, 5-33 mm Hg): upper row (normal) and lower row (hypertension). CBFV representative waveforms associated with low average ICP values tend to have more distinct sub-peaks than those associated with high average ICP pulses. This is one of the main advantages of this framework compared to others, as our approach places particular emphasis on waveform subpeaks. The difference between the magnitudes of the second and third subpeaks is greater in the CBFV representative waveform associated with the high average ICP pulse than in the normal average ICP pulse. In one approach, the difference between the size of the subpeaks and / or the size of the subpeaks is used to characterize the ICP as normal or IH.

  [0054] The method extracts various structural features from the TCD-based CBFV waveform. In one approach, the method is performed by a processing circuit. And the next step is to associate the association rules (or functions) between their CBFV structural features and the corresponding labels (eg +1 for high blood pressure samples, -1 for normal samples) It is to learn. It can simply be expressed as:

Where X is an n by 100 matrix of structural features, Y is an n by 1 vector of corresponding labels, n is the number of samples, and f is to be learned or to be trained It is a related function or classification indicator. In one embodiment, the quality of the trained classification indicator is measured by its predictive accuracy. In other words, a good classification index is an index that can assign new features to the correct classification that were not found during training.

  [0055] In one approach, the learning algorithm includes a graph based semi-supervised learning classification index technique called spectral regression. This approach combines conventional regression techniques with spectral graph analysis and can be used as clustering and dimension reduction techniques. In contrast to many conventional graph-based algorithms, which are inherently transductive, spectral regression techniques are natural out-of-sample in both the linear and kernel cases. give an extension).

[0056] The first step of spectral regression is to calculate the set of responses y _i for each sample x _i by applying spectral techniques to the graph matrix. Once those responses are obtained, the ridge regression technique finds the regression function. The algorithmic procedure of spectral regression can be outlined as follows.

1) Adjacent graph structure: Let G show a graph with n nodes. Here, the ith node represents the _ith sample x _i . The graph G is constructed by the following three steps.
a) Connect nodes i and j if nodes i and j are in the k closest neighbors to each other.
b) Connect nodes i and j if nodes i and j belong to the same class.
c) If i and j belong to different classes, delete the connection between i and j.

[0058] 2) Construction of weight matrix: let W indicate an n × n sparse matrix, and elements of the sparse matrix W _ij may be assigned as follows:

Otherwise, l _q is the number of samples belonging to the q th class, and S (i, j) is a similarity function between x _i and x _j . The choice of our similarity function is the heat kernel:

  3) Eigen decomposition: Find the largest eigenvector of the lower eigenproblem.

Here, D is a diagonal matrix, and its element D _ij is equal to the sum of the ith row of W.

4) Regularized Least Squares: Solve a regularized least squares problem for the p-th largest eigenvector y ^p as follows.

Here, a is a regression coefficient vector, l is the number of labeled samples, γ is a parameter for adjusting the weight of unlabeled samples, and α is a regularization parameter. It is important to note that while y _i is a scalar response, x _i is a sample vector. By setting γ = 1, the closed-form solution of a ^p is expressed as follows.

  [0061] One of the many benefits of spectral regression is that it provides a uniform learning approach. When all the samples are labeled, spectral regression is essentially the same as regularized discriminant analysis. In this case, the sparse matrix W is block diagonal and the response y in (3) is equal to:

Here, l ^p is the number of samples belonging to the p-th class, and c is the total number of classes. On the other hand, if all the samples are not labeled, then the spectral regression is a spectral clustering technique with natural out-of-sample expansion capabilities, whose objective function is

Equation (7) states that responses y _i and y _j should be close to one another if the ith sample and the jth sample are similar. The eigenvectors of the problem in (3) yield the optimal solution of the problem in (7). In the case of semi-supervised learning, the responses y _i and y _j as solutions of the eigenproblem in (3) can be as close as possible if the ith and jth samples belong to the same class. Such a property is essential for semi-supervised learning as the same labeled sample is expected to have the same or similar response.

[0063] Another important advantage of spectral regression is that it can be easily extended to non-linear discriminant analysis by mapping all the samples to the reconstructed nuclear Hilbert space. And we can perform spectral analysis in high dimensional feature space, which is called spectral regression kernel discriminant analysis (SRKDA). In this case, the closed-form solution of a ^p in (5) is as follows.

Here, K is an n × n matrix, its element K _{i, j} is K (x _i , x _j ), and K (·, ·) is a kernel function. In one approach, a Gaussian kernel is selected and used. SRKDA has been utilized in certain clinical and experimental approaches, as described in further detail below.

  [0065] There are two important parameters to be optimized in the SRKDA algorithm: standard deviation σ of heat kernel in (2) and standard of nonlinear (ie Gaussian) kernel function K (···) deviation. The standard deviation σ of the heat kernel is estimated as follows.

Here, n is the total number of training samples. In one approach, the parameter σ may be optimized by performing individual cross validation within a given training data set. However, there is a risk of overtuning σ for a given training data set and compromising the generalizability of the model. In contrast, the estimation of σ in (9) is simple to obtain, and its value is similar to that which would have been obtained by taking a cross validation approach. Thus, in one embodiment, the standard deviation of the Gaussian kernel function K (·, ·) is estimated as in (9).

[Clinical example]
[0066] In order to verify the systems and methods described herein, datasets comprising ICP, CBFV, and ECG data were recorded between UCLA Medica between July 15, 2008 and November 16, 2011. The data were collected from 90 patients (age: 18-92 [median: 47], sex: male 47 / female 43) housed in the nervous system ICU of the center and the ward unit. Among them, 44 patients suffered from TBI, 36 had SAH and the rest were diagnosed as suspected NPH. Table 1 summarizes patient diagnosis and demographic information.

  [0067] ICP was measured invasively via continuous ICP monitoring for clinical purposes, using an intracerebroventricular catheter for brain injury or an intraparenchymal microsensor for NPH. Simultaneous cardiovascular monitoring was also performed using a bedside GE monitor. CBFV signals were obtained at the MCA ipsilateral to the ICP measurement site, while experts at the Cerebral Blood Flow (CBF) laboratory of the neurosurgical department of UCLA used TCD. Daily clinical assessment of the patient's cerebral hemodynamics was performed. The duration of the signal collected varies depending on how long MCD TCD monitoring can be done. Typically, TCD monitoring lasted only 3-5 minutes as the probe had to be handheld. This study was approved by the Research Ethics Review Board without any personal health information intervention.

  [0068] All signals are recorded via a mobile cart equipped with a PowerLab data acquisition system (ADInstruments, Inc., Colorado Springs, Colo.), Which samples analog signals from the bedside monitor at 400 Hz. The recorded signal was then stored in chart binary file format for further analysis.

  [0069] The range of ICP was divided into three groups: normal (<15 mmHg), gray zone (15-30 mmHg), and IH (> 30 mmHg). ICP below 15 mm Hg is determined to be a measure of normality. In contrast, it is determined that the patient's condition is at greater risk if the ICP exceeds 30 mm Hg.

  [0070] The ICP and CBFV segments 3-5 minutes long, recorded simultaneously during each session of daily cerebral hemodynamic assessment, were resolved into 1-minute segments. Each of these one-minute segments was used to provide one sample, a set of CBFV structural features. From 90 patients, 563 samples were obtained over 131 sessions. The samples were assigned labels by applying the labeling criteria described above at the session level rather than at the sample level. In other words, if any of the samples belonging to a given session meet the criteria of IH, then all samples of that session are labeled as IH. The rationale behind this labeling scheme is that what the carer is most interested in is whether the patient experiences any IH during a given session. Which one-minute segment during a session is associated with IH is typically of less interest. However, one approach is to identify the specific time of occurrence of IH or multiple occurrences of IH. In contrast, a given session is labeled as normal only if all the samples in that session meet the normal (ie, <15 mm Hg) criteria. Any session not labeled as IH or normal is labeled as a gray zone. Table 2 outlines the results of our labeling scheme. It is important to note that although all samples from 46 normal sessions correspond to ICP below 15 mmHg, only some of the 48 samples from 8 IH sessions correspond to ICP above 30 mmHg It is.

[0071] With the labeling scheme described above, we performed two separate cross-validation experiments. The purpose of the first cross validation experiment was to quantify the ability of SRKDA to distinguish IH samples from normal samples. In the first cross validation experiment, 10-fold cross validation was performed only on IH and normal samples, and gray zone samples were used only for training purposes. We them gray zone sample, used in three different ways: supervised ^1, supervised ^2, and semi-supervised. In the supervised ¹ setting, gray zone samples are labeled as IH or normal based on a conventional IH threshold of 20 mm Hg, and used as a "labeled" sample for training purposes. In the supervised ² setting, they are considered "noisy" samples and are completely discarded. Finally, in a semi-supervised setting, it is only used as an "unlabeled" sample for training purposes. We also considered PI-based IH detection as our baseline classification index and compared its performance against our proposed method.

  [0072] The purpose of the second cross validation was to test whether SRKDA could group gray zone samples according to their corresponding ICP values. In this experiment, 10-fold cross validation is performed only on gray zone samples in a semi-supervised learning mode, all IH and normal samples are used only for training purposes. The label of the hypertensive sample is +1 and the label of the normal sample is -1, but the direct output of SRKDA is a continuous scaled estimate of the label. We were mainly interested in whether these continuous scale estimates of gray zone samples were strongly correlated with their corresponding ICP values.

  [0073] It is important to note that all cross validation in our study was done in a patient-free manner. Several samples from one patient are used for training purposes and none of the samples from the same patient can be used for testing purposes. The performance of IH detection is calculated on the session level, not on the sample level. As mentioned above, there is great interest in knowing whether an individual session is associated with an IH. Since the direct output of the SRKDA is a continuous scale label estimate of the individual samples, we collected all samples belonging to a given session and selected the estimate with the largest value of the label as the session label.

  [0074] The following section describes two separate performance measurements: area under the curve (AUC) used in our study and decision curve analysis.

  Area Under Curve: The prediction accuracy is measured by the area under the receiver operating characteristic (ROC) curve. The area under the ROC curve can be considered as the probability that the rank of randomly selected positive samples is greater than the rank of randomly selected negative samples. By plotting the AUC of the semi-supervised SRKDA against the number k of nearby neighbors, we examined the effect of k on the performance of the semi-supervised classification index.

[0076] Decision curve analysis: AUC as a measure of predictive accuracy does not emphasize the clinical consequences of false-positive and false-negative results. In other words, we can not at all tell us that it is clinically useful to use a given diagnostic method. For example, if misdiagnosis is more harmful than unnecessarily treating the disease, diagnostic method A with higher sensitivity has other diagnostic methods with higher specificity but lower sensitivity. For B, although the AUC of method A may be slightly smaller than the AUC of method B, it is a better clinical choice. We used decision curve analysis to evaluate and compare different diagnostic methods by incorporating clinical consequences. Decision curve analysis draws net given diagnosis Throughout the disease potential threshold p _t (i.e., clinical benefit). The disease probability threshold _pt for which the patient chooses treatment (ICP monitoring of insult in our case) decides to reflect the patient's weighting for needed (true positive) and unnecessary (false positive) treatment. However, when calculating profit, there is no clear reason to focus only on those individuals who choose treatment. Recently, there is a modified net benefit for all individuals with and without treatment. This overall net benefit can be expressed as:

[0077] FIG. 5 compares our IH detection method in the first cross-validation experiment, where the dashed green line is that of the PI-based IH detection method (baseline method) and the thin dashed dotted blue line The line is for supervised ¹ IH detection method, the thick dashed-dotted light blue line is for supervised ² IH detection method, and the red solid line is for semi-supervised ^k IH detection method is there. Since only the semi-supervised ^k IH detection method has to deal with the number of neighbors to search for k, the AUC of all other methods was constant throughout the range of k. Each line and gray area represents the mean AUC and one standard deviation displacement on multiple (= 100) split cross validation. There are several interesting aspects to be pointed out in FIG. First, all of the IH detection methods we proposed are sufficiently better than PI-based IH detection methods. Second, the supervised ¹ IH detection method is slightly worse than the supervised ² IH detection method. It shows that using gray zone samples as labeling data based on 20 mmHg actually degrades the prediction accuracy of the SRKDA classification indicator. Third, semi-supervised ^k IH detection methods tend to increase as k increases.

[0078] Figure 6 shows the decision curve of the IH detecting method in the first cross-validation experiments (net vs. potential threshold p _t). Net of PI-based IH detecting process (dashed green line) is only on a very narrow range of _{p t} from 0.14 to 0.27, two extreme approaches (i.e., treating to all Slightly better than the net profit of (Treat-All) and no treatment for anyone (Treat-None). In contrast, net of our proposed method based on structural features, significantly better than the net of the two extreme approaches on a wide range of p _t.

[0079] FIG. 5 also demonstrates the superior performance of the semi-supervised IH detection method over the supervised method in a qualitative sense. However, since the decision curves in FIG. 6 cross each other, it is not trivial to compare quantitative performance. Table 3 outlines the net income gains of each IH detection method as the averaged difference between the net income of each IH detection method and the net income of the two extreme approaches over the entire range of p _t . The net benefit gain is the degree of true net income that can be achieved by using a unique IH detection method beyond two extreme approaches (ie, treating for all and not treating for everyone) I plan to measure it. The net benefit gains listed in Table 3 show that the semi-supervised IH detection method is significantly better than the other methods and that the PI based IH detection approach to all treatment and no treatment for anyone It clearly proves that it does not go better.

  [0080] FIG. 7 visualizes the results of a second cross validation experiment in which the continuous scale label estimate of the gray zone sample is on the y-axis and the corresponding ICP value is on the x-axis. Continuous measure label estimates tended to increase as the corresponding ICP values increased, and the correlation coefficient between them was 0.55 with a p value of 2e-4.

  The regularization parameter α in (4) has its complexity,

In order to avoid overfitting the least squares solution ^ap by penalizing In one approach, this parameter may be optimized by performing individual cross validation in the training data set. Instead, by testing SRKDA on a preliminary data set, we learned that the regularization parameter α does not significantly affect the performance of SRKDA as long as its value is small (<0.01) . Thus, in one approach, such as the clinical data set and analysis described herein, α is set to 0.01.

  [0082] In certain approaches, such as those used for analysis of clinical data described herein, although correlations between some structural features are likely, feature selection methods are not used. Thus, in one approach, a feature selection method is implemented that exploits the correlation between features. Non-linear kernel based classification methods such as SRKDA are effective in classifying high dimensional data such that feature selection or feature weighting is unnecessary for the purpose of classification. For this data, feature selection techniques did not provide significant performance improvement for IH detection methods. However, it should be noted that the time delay between the ECG-QRS and the first trof was one of the most important features for accurate IH detection, as shown in FIG. . By simply excluding this feature from our simulation studies, the performance of IH detection degraded to -10% on average. There was no other subset of features that affected the performance of IH detection to that extent.

  [0083] Our cross-validation results in FIGS. 5 and 6 clearly show that CBFV PI does not reflect well on elevated ICP as compared to using the complete set of pulse structure metrics. There is. Changes in reported PI-ICP correlation properties can be attributed to the fact that CBFV PI is influenced by many other factors, including arterial blood pressure and age. In addition, there are three very different patient populations in this study, which further disrupts the PI-ICP relationship. The superior performance of our approach shows that the SRKDA model can implicitly select discriminable features from a given set of structural metrics that are hardly disrupted by factors not related to IPC status ing.

  [0084] As shown in FIG. 5, the performance (ie, prediction accuracy) of the semi-supervised IH detection method improves as the number of near neighbors (or samples) k increases. The qualification is that the unique data structure in the sample whose weight matrix W is more dense with larger k and is labeled as unlabeled sample improves the predictive power of SRKDA. It can be explained by pointing out the fact that it can be searched over a wider range. The decision curve analysis results in Figure 6 and Table 3 supporting the idea that the semi-supervised IH detection method works better with large k.

  [0085] The performance of the proposed IH detection method at the sample level was significantly lower than the performance at the session level. One possible explanation is that CBFV can respond to the rise to ICP in a delayed manner by CBF autoregulation. When acute ICP elevation occurs, an inherent low interest rate delay is inevitable to see CBFV pulse structure changes. The delay is usually 10-20 s for undamaged autoregulation. Thus, in one approach, IH detection is used on session level.

While decision curve analysis is devoted to the utility of the model, ROC curve analysis is focused only on the accuracy of a given prediction model. As a result, the optimal operating point based on the latter is quite different from the optimal operating point based on the former. Typically, the optimal operating point based on the ROC curve is the operating point where Youden's modulus (ie sensitivity + specificity -1) is maximized. Threshold corresponding to the optimum operating point, called the Optimal precision operating point and optimum accuracy threshold p _a. However, the net benefit of the prediction model with the optimal accuracy threshold p _a falls below the net benefits of the two extreme approaches as soon as p _t leaves the optimal accuracy site. The optimal operating point and the corresponding threshold are referred to as the optimal net profit operating point and the optimal net profit threshold. The optimal net operating point on the ROC curve is determined as the point whose slope is equal to

here,

Is a portion of all positive samples. The optimum net operating point is "ideal (optimal)" in the sense that it maximizes the net in the singular values of p _t.

[0087] FIG. 8 shows three different operating points on the ROC curve of the semi-supervised ²⁰⁰ IH detection method, where the red point is that of the optimal accuracy operating point with p _a = 0.12 and is green The point of is that of the optimum net profit operating point of p _t = 0.2, and the blue point is that of the optimum net profit operating point of p _t = 0.4. The semi-supervised ²⁰⁰ IH detection method with p _a = 0.12 can provide optimal accuracy performance. However, it is when p _t is close to 0.12 only treated for all, or not be treated to everyone, can lead to better net than approach, it is highly p _t value When selected, it is virtually useless. Figure 8 is why, predictive models of high sensitivity is to be preferred with a small value of p _t, on the other hand, the prediction model of high specificity indicates good or, is to be preferred in a large value of p _t .

  [0088] The IH diagnostic tools described herein may be used in various sets of clinical applications where the appropriate pt may vary. Thus, it is very useful to perform decision curve analysis to select different models and their operating points to suit the intended use of obtaining IH diagnostics.

  However, our results show that using a set of CBFV pulse structural metrics is more promising than using a single metric such as PI for dealing with data from disparate patient populations However, interest remains in examining whether the SRKDA model trained using data from patients with brain injury and hydrocephalus can be extrapolated to IIH patient populations.

  [0090] An ICP level of 20 mm Hg is a conventional threshold to define an illustration of IH. However, it is somewhat arbitrary and tends to be the cause of many false positive criminal laws. In one approach, the systems and methods described herein divide the ICP range into three groups: normal (<15 mm Hg), gray zones (15-30 mm Hg), and IH (> 30 mm Hg). By adopting the SRKDA algorithm, we demonstrate that semi-supervised learning approaches in which gray zone samples are treated as unlabeled data are more suitable for IH detection than traditional supervised learning approaches.

  [0091] The above steps such as described and shown in the flow diagram may be performed in any order or flow, not limited to the order or flow shown or described in the figure or row It should be understood that In one approach, steps may be excluded. In one approach, steps may be added or combined. Additionally or alternatively, some of the above steps may be performed or performed substantially simultaneously, or in parallel where appropriate, to reduce latency and processing time.

  [0092] The methodology disclosed herein preferably has a Transcranial Doppler (TCD) system designed to detect potential brain trauma by monitoring brain blood flow. This is made possible by using an ultrasonic transducer positioning mechanism. This is accomplished by positioning the ultrasound transducer on either side of the patient's head and optimally positioning the transducer to maximize the ultrasound Doppler flow signal.

  [0093] In use, the Ultrasonic Transducer Positioning Mechanism (UTPM) is placed adjacent to the temple area on both sides of the patient's head. The line of intersection above the patient's head and earlobe provides a reference landmark for the placement of the mechanism's housing. Although a hand-held probe as shown may be used, the position of the housing relative to the head is maintained as desired via attachment to a separate headgear device.

  [0094] The ultrasonic transducer positioning mechanism provides an optimal location on the patient's head to provide the best Doppler flow signal through minimal bone attenuation and a zero degree angle of ultrasound to the brain's arteries. Ask. That is, the mechanism positions the transducer under the direction of the processing unit seeking signal maximization via XYZ + XY tilt commands to the drive circuitry of the mechanism. Preferably, the mechanism is capable of free standing scanning and positioning.

  [0095] Figures 10 and 11 are front and back views of a portable transcranial Doppler device 20 for use in collecting CBFV raw data as described herein. The device 20 includes a body 22 having a size and shape very similar to a conventional smart phone, with a display screen 24 that may be an LCD with a touch sensor. An ultrasound probe 26 may be housed within the holster 28 at the back of the device and magnetically secured. Various controls may be provided as the upper panel 30 or the button 32 under the screen 24. The portable device operates with either hand, and the display screen 24 can be adjusted in a given direction. The expert withdraws the ultrasound probe 26 from the holster 28 and applies it to the area above the patient's head, typically around one of the temples. Then, measurements of CBFV raw data are taken and recorded during the time period. The same process scan is repeated at different locations and is completely non-invasive. Preferably, an ultrasound coupling gel, as typically used for fetal ultrasound probes, enhances comfort to the patient and improves the transmission of ultrasound through the epidermis and dermis. Used for

  FIG. 12 shows an automated TCD headset 40 having a display screen 42 in front of it. More specifically, the headset 40 includes two ultrasonic probes 44 on both sides, and a headband 46 extending around the front so as to connect the two probes. As seen in FIGS. 12A and 12B, the TCD headset 40 fits the patient's skull, with the probe 44 positioned in both temples. The probe 44 includes therein a TCD scanner that can automatically position the middle cerebral artery (MCA). Desirably, the headset 46 is resilient so that the headset is hooked on the front of the head of various different head sizes so that the inner surface of the probe 44 makes good contact with the temples Allows for a perfect fit. Again, preferably, a lubricating gel is used to improve sound transmission.

  [0097] FIG. 13 is a side view of another exemplary TCD headset 50 worn by a patient and having a forehead strap 52, a back strap 54, and a cranial strap 56. As shown in FIG. The straps 52, 54, 56 help secure the headset on the head and in particular ensure good contact with both temples of a pair of reciprocating TCD scanners 58. A TCD scanner 58 is provided on the interface member 60 at the intersection of the three straps 52, 54, 56 for forward and backward reciprocation, as indicated by the movement arrows. In one embodiment, the TCD scanner 58 rotates approximately 60 degrees in each direction about a Z axis perpendicular to the XY scan plane. Although not shown, a small motor in the interface member 60 allows movement of the scanner 58.

  [0098] The system of three straps 52, 54, 56 is extremely effective at keeping the headset 50 in place. The cranial straps 56 include Velcro® cuts to make them adjustable and the posterior straps 54 are preferably resilient so that a pair of clamping knobs 62 on each joint member 60 and a clamping knob in the middle of the forehead strap 52 64 allows fine adjustment of the position of the scanner 58 for XY calibration. The cranial straps 56 help to limit the movement of the headset 50 once it has been fixed due to the jaw and associated muscle movement.

  The cable 66 may be attached to the junction member 60 for connection to a control unit, such as a tablet computer, or the system may be wireless. Desirably, each scanner 58 preferably includes an injection port 68 formed by a recessed guideway into the conduit for insertion of the lubricating gel into the inner contact surface. This helps to reduce the spread of the gel. In a preferred embodiment, the TCD sensor inside each scanner 58 may be offset in the Z-direction or towards and away from the temples to optimize acoustical contact.

  [00100] FIG. 14A is a perspective view of an exemplary TCD headset 100 positioned on a soft mounting leg 102 on the side of a patient's head. Two sizes of the patient's head, small S and large L, are shown in the outline to show the adjustable range of the headset 100 for different sizes of the patient. Outer housing 104 is shown in phantom to visualize the components inside headset 100.

  [00101] FIG. 14B shows the outer housing 104 to the contour of the wearer's head for clarity and a second on the opposite side of the patient's head connected to the first set by the strap 110. The headset 100 is shown. Preferably, each headset 100 is a set of cushions against the head, and a plurality of mounting legs similar to small suction rings to provide some space between the head and the outer housing 104. It has 102. Desirably, there are three mounting legs 102 on each side. Headset 100 is secured by applying tension to strap 110. As shown, there may be one frontal strap 110 or, as mentioned above, there may be one around the back and one on the skull.

  [00102] Referring to Figures 15A and 15B, a side view of the TCD headset of Figures 14A and 14B is shown with the housing 102 removed. Inside the housing, a scanner 120 is provided on a carriage 122 that slides on a pair of angled rails 124. The carriage 122 includes a small motor that turns a drive gear that meshes with the small teeth. The motor 130 can be controlled remotely or by wiring, so the carriage 122 can be moved diagonally along the rail 124.

  [00103] The TCD scanner 120 mounted on the carriage 122 can be moved over the area of the subject's temples. The headset 100 can desirably scan an area of about 2 square inches (sq in) as indicated by the dashed square area 150. To cover the entire area 150, the upper end of the rail 124 pivotally attaches to the frame member 152 which translates laterally along a generally horizontal path. More specifically, the pivot point 154 on the frame member 152 connects to a translating rod 156 that can be moved by the cylinder 158 in a piston / cylinder relationship. Alternatively, the cylinder 158 can include a small motor that engages the end of the rod 156 opposite the pivot point 154 to translate it laterally. There are several ways to achieve this movement, but each is controlled with the movement of the carriage 122 for coordinated two-dimensional movement of the scanner 140 in the XY plane above the target area 150.

  In addition, the robotic arm surrounding the scanner 140 provided for movement on the carriage 122 has a Z-axis displacement device which is preferably actuated by the stepper motor 160. The robotic arm is further equipped with a pressure sensor (not shown) that maintains sufficient pressure of the scanner 140 on the skin for consistent signal quality. This constant pressure helps to address some of the variability problems associated with patient movement and the TCD.

  [00105] With respect to the preferred mechanism, translational linear motion along the XYZ axis + XY tilt is achieved through the use of a stepper motor driven by a local Motor Control Unit (MCU). Servo feedback is provided to ensure that the command number of steps has been achieved. The servo feedback signal is in the form of a reverse EMF or an encoder signal provided to the MCU.

  [00106] Command Set: The movement of XYZ axis + XY tilt is controlled via TPU processor. The commands for movement along any axis are in the form of a signed integer indicating the incremental number of steps to be moved along each axis. Preferably there is also control of the tilting / pivoting movement.

  [00107] Units that can fit several head sizes are important for widespread adoption. If the head mounts do not fit properly, the TCD probe can not obtain an optimal signal. The disclosed design addresses this concern by separating headset anchoring and robotics. This allows the user to fit the headset to a head of any size without impact on the ultrasound mechanism to reach the signal.

  [00108] Each of the headset embodiments can remove all ultrasound coupling gels after use. Preferably, a wipe or other similar device is provided to protect the mechanism from the accumulation of foreign matter in the mechanism. The material chosen should withstand the line with water, isopropyl alcohol and other cleaning agents routinely used in the doctor's office or clinical setting. In the preferred form, the headset should not be heavier than 10 ounces.

  [00109] The foregoing is merely illustrative of the principles of the present disclosure, and systems, devices, and methods can be practiced other than the described embodiments, which are presented for purposes of illustration and not limitation. The systems, devices, and methods described herein are indicated for non-invasive diagnosis of IH using TCD, but other diagnoses or treatment procedures, cerebral malaria, mild / moderate traumatic It may be applied to systems, devices and methods to be used in other procedures, including procedures outside of physiological applications, including brain injury and others.

  [00110] In certain embodiments, the systems and methods described can be used for the diagnosis of mild and moderate TBI without an increase in ICP. The underlying physiological functions are different but the core analysis is the same. Mild TBI followed by changes in brain hemodynamics, which are well documented by several studies. The physiological causes of these changes are local blood flow changes due to increased metabolic demand in certain areas of the brain, changes in CBF due to destruction in the cerebral vasculature or the brain itself (high intracranial pressure-ICP To a reduced local elastic modulus (e.g., decreased compliance).

  [00111] For example, a study by Jaffres et al. (Jaffres, P., et al., Transcranial Doppler to detect on admission patients at risk for neurological determination following mild and moderate brain trauma. Intensive Care Med, 2005. 31 (6): p. 785-90) investigated the use of the Pulsatility Index (PI) of CBFV in mild and moderate TBI in the emergency room for prognostic purposes, and their results It has been shown that only can be distinguished from those who do not have patients with secondary neurological deterioration (SND). Research by Bouzat et al. (Bouzat, P., et al., Transcranial Doppler to screen patients with mild to moderate traumatic brain injury. Neurosurgery, 2011. 68 (6): p. 1603-9; discussion 1609-10) Confirmed these results and committed 95% overall accuracy in identifying patients that express SND.

  [00112] In addition, a number of studies have investigated a decrease in CBF, a possible underlying cause of physiological defects in mild TBI (eg, Giza, C. and DA Hovda, The Neurometabolic Cascade of Concussion. J. Athl Train, 2001. 36 (3): p. 228-235, Grindel, SH, Epidemiology and pathophysiology of minor traumatic brain injury. Curr. See Sports Med Rep, 2003. 2 (1): p. 1 8-23. ).

  [00113] An important study by Maugans et al. Using phase contrast angiography in children with sports-related concussion (Maugans, TA, et al., Pediatric sports-related concussion produces cerebral blood flow alterations. Pediatrics, 2012. 129 (1): p. 28-37) report two main results. First, a marked reduction in CBF in children aged 11-15 years within 72 hours of mild TBI. Second, only 14% and 64% of patients returned to the normal CBF range 14 and 30 days after injury and 14 days after asymptomatic, respectively. Furthermore, a related study by Gall et al. (Gall, B., WS Parkhouse, and D. Goodman, Exercise following a sport induced concussion. Br J Sports Med, 2004. 38 (6): p. 773-7) After concussion, Hockei reported that he had a unique heart rate response when stressed by exercise, despite the absence of post-concussion symptoms. Both studies revealed that despite athletes being asymptomatic, there remains a physiological deficit that can be harmful if further injury or activity is sustained. Finally, in the study by Len et al. (Len, TK, et al., Cerebrovascular reactivity impairment after sport-induced concussion. Med Sci Sports Exerc, 2011. 43 (12): p. 2241-8) It has been shown to negatively impact cerebrovascular reactivity (CVR) as compared to the control group. The results showed that the CVR test identified athletes who had a concussion and athletes who did not have a concussion. These results resonated with those of Gall et al. Who showed that asymptomatic individuals under stress show physiological changes.

  [00114] One approach to examining the underlying physiological functions of mild TBI is to stimulate to aggravate changes in the cerebrovascular system and to describe the changes more accurately in our described framework To use. The stimulation may be a change in arterial blood pressure (exercise, leg cuffs, medications, etc.), a change in carbon-dioxide (CO2) concentration in the arterial blood supply, or local by changing metabolism in specific areas of the brain (ie A flash of light can be given in a variety of different ways, including to stimulate the occipital lobe.

[00115] In one design, cerebrovascular bed, arterial concentration of _{CO 2 (arterial blood concentrations of CO} 2: PaCO 2) is very sensitive to changes in. Increased arterial CO ₂ levels (such as from suppressing human respiration) cause arteriolar dilation resulting in increased velocity in the large cerebral artery upstream by increased cerebral blood flow. Conversely, decreased CO2 (via hyperpnea) results in decreased CBFV due to arteriolar contraction causing a decrease in CBF.

[00116] cerebrovascular reactivity (Cerebrovascular reactivity: CVR) describes the change in CBFV due to changes in PaCO _2. The goal of the CVR test is to assess the vasodilator or vasoconstrictor capacity of the resistive arterioles of the brain and to be harmed after severe TBI, migraine, long-term space flight, stroke, and carotid artery stenosis It is shown. More recently, CVR has been shown by Len et al. (Later) to have potential as a marker of physiological dysfunction in mild TBI. In their study, both concussion and control subjects were studied with respiratory depression and hyperpnea to examine CVR. Similar to Gall et al.'S study using exercise as a physiological stress to clarify changes in concussion patients, Len et al showed changes in mean CBFV dynamics from repeated breath hold and hyperpnea . However, CBFV data were sampled at 1 Hz, excluding all morphological information from the analysis. In this application, the CVR test utilized by Len et al. Is extended to take into account not only the mean velocity but also the overall shape of the CBFV waveform. Patients are asked to press his or her breathing to increase the CO ₂ level, CBFv is monitored. Conversely, the patient is asked to hyperbreathe to lower the CO ₂ levels and CBFV is monitored. Observing CVR using average velocity "only" as in Len et al. Gives an incomplete depiction.

[00117] While some embodiments have been described as being examples of the present systems and methods, those skilled in the art will appreciate that the intent and scope of the systems and methods described herein are within the purview Will recognize an embodiment of Modifications and variations will be made to the disclosed embodiments without departing from the scope of the present disclosure. Those skilled in the art will understand that the application of the embodiments disclosed herein will vary. Thus, additions and modifications can be made without departing from the essence of the present disclosure. In this regard, such modifications are still intended to be within the scope of the present disclosure. Changes and modifications will occur to those skilled in the art after reviewing this disclosure. The disclosed features may be implemented in combination or subcombination (including combinations of multiple sub-subsets or subcombinations) with one or more other features disclosed herein. The various features described or illustrated above, including their optional components, may be combined or integrated into other systems. Furthermore, certain features may be excluded or not implemented. Thus, the disclosure is not intended to be limited to particular embodiments, but is intended to cover modifications within the spirit and scope of the present disclosure.
In the following, the invention described in the claims attached at the time of filing of the present application is appended.
[C1]
A non-invasive method for diagnosing pathological intracranial pressure status in a patient, comprising:
Non-invasively collecting raw data indicative of cerebral blood flow velocity from at least one blood vessel disposed in the cranial region of said patient;
The method is
Converting the raw data into structural features using a previously validated database of cerebral blood flow velocity waveforms;
Classifying the structural features using previously identified pathological intracranial pressure conditions;
Recommending a diagnosis based on the step of classification;
Characterized by
[C2]
The pathological intracranial pressure status is described in [C1], which is selected from the group consisting of moderate traumatic brain injury (TBI), severe TBI, stroke, brain edema, spontaneous elevated intracranial pressure, cerebral malaria the method of.
[C3]
The method according to [C1], further comprising the step of updating the database of mild traumatic brain injury (mTBI) status identified previously with said diagnosis.
[C4]
The method according to [C1], wherein said step of non-invasively collecting raw data comprises using a transcranial doppler.
[C5]
Said step of non-invasively collecting raw data comprises processing the ultrasound waves generated and reflected by the ultrasound transceiver,
The reflected ultrasound is processed using Doppler waveform analysis,
The method described in [C1].
[C6]
The step of non-invasively acquiring raw data includes using one of an MRI system, a CT scanner, a pressure transducer, an optical imaging device, and a near infrared imaging device, in [C1] Method described.
[C7]
The method according to [C1], wherein said step of recommending a diagnosis comprises recommending at least one of normal, suspected injury and definitive injury.
[C8]
The method of [C1], wherein the structural feature comprises at least one peak.
[C9 ]
The method of [C8], wherein the structural feature further comprises at least one subpeak.
[C10]
Furthermore, further comprising providing stimulation to the patient,
The method of [C1], wherein the structural feature is an indicator of a subject's response rate to the stimulus.
[C11]
The method according to [C10], wherein the stimulation comprises changing the carbon dioxide (CO ₂ ) concentration of the artery of the patient .
[C12]
The method of [C10], wherein the stimulation comprises changing the level of light stimulation to the eye.
[C13]
The method of [C10], wherein the stimulation comprises changing arterial blood pressure levels in the subject.
[C14]
The method according to [C1], wherein said step of non-invasively collecting raw data comprises using at least one blood vessel and using a headset device to measure said raw data.
[C15]
The method according to [C14], wherein the headset device includes a pressure sensor and communicates with a processor to control the pressure exerted by the scanner probe on the patient's head.

Claims (16)

A headset for detecting a neurological condition of a user,
A probe configured to be positioned relative to the face of the user's head when the headset is placed on the head of the user, the headset generating energy generated by the headset A probe configured to be provided on the head of the user such that the probe is focused on the head of the user;
A plurality of rails configured to be positioned across the area of the temples of the face of the user's head when the headset is mounted on the user's head;
A carriage coupled to the plurality of rails ;
The carriage comprises a motor
The probe is provided on the carriage,
The motor, by moving the carriage, the carriage, while carrying the probe, laterally along the upper the surface of the head of the user to the plurality of rails, and the surface of the head of the user Configured to move perpendicular to
With the carriage,
A frame member attached to an end of the plurality of rails;
A drive mechanism configured to drive the frame member along a path to move the end of the plurality of rails to provide two dimensional movement of the probe over the region of the temple; a headset with The probe, when the headset is provided on the head of the user, configured to be positioned in the vicinity of the region of the temples of the user, the headset according to claim 1.   The headset according to any one of claims 1 or 2, wherein the energy generated by the headset is ultrasonic energy.   The motor according to claim 1, 2 or 3, wherein the motor is configured to position the probe on the face of the user's head such that the probe concentrates the energy towards the middle cerebral artery of the user. The headset according to any one of claims 3 to 10. The motor attached to the carriage is:
When the headset is fixed to the user's head, the probe, along the surface of the head of the user, so as to move on the area of the temple of the user laterally, to drive the carriage A first motor configured as
To displace the probe provided on the carriage along an axis perpendicular to the surface of the head of the user, and a second motor for driving the displacement device,
The headset according to any one of claims 1, 2, 3, or 4, which comprises   The motor rotates the probe in a range of 0 degrees to about 60 degrees around the axis perpendicular to the face of the user's head, with the axis perpendicular to the face of the user's head. The headset of claim 5, further comprising a third motor configured to cause   The system further comprises an additional probe configured to be positioned on a different side of the user's head and to concentrate the energy generated by the headset in the different side of the user's head. The headset according to any one of claims 1 to 6. The probe comprises an injection port configured to direct a lubricating gel to the contact surface of the probe,
The contact surface is configured to contact the surface of the user's head when the headset is provided on the user's head.
The headset according to any one of claims 1 to 7.   The headset according to any one of claims 1 to 8, wherein the neurological condition is traumatic brain injury (TBI), stroke or dementia. When the headset is fixed to the user's head, the plurality of rails, configured to be positioned diagonally from the frontal of the user toward the one ear of the temple region as the user's head of the user The headset of claim 1 . The headset of claim 10, wherein the drive mechanism is configured to drive the frame member laterally along a generally horizontal path. Further comprising a pressure sensor coupled to the probe,
The pressure sensor is configured to sense the pressure of the probe against the surface of the user's head based on the pressure exerted by the probe.
Headset according to any one of claims 1 to 11. The motor comprises a stepper motor,
The headset further comprises a motor control unit (MCU) coupled to the stepper motor.
Headset according to any one of claims 1 to 12. A cranial strap coupled to the probe;
A plurality of clamping knobs at a plurality of coaptation members along the skull strap;
And further
The tightening knob is configured to adjust the tightening of the headset on the user's head,
The headset according to any one of claims 1 to 13 . A plurality of soft mounting legs configured to be added to the user's head to stabilize the headset while provided on the user's head. The headset according to any one of 14 . A cable for connecting to a device for monitoring the signal received by the headset and / or
The headset further comprises means configured to communicate with devices wirelessly for monitoring signals received by, a headset according to any one of claims 1 to 15.

Images