JP7580448B2 - Gate area estimation program, gate area estimation method, and gate area estimation device - Google Patents

Description

The present invention relates to a gate region estimation program for estimating a gate region in flow cytometry.

Flow cytometry (FCM) is a technology that can measure multiple characteristics of individual cells. In FCM, cells suspended in a flowing liquid are made to flow in a single file. Light is shone on each individual flowing cell, and the scattering and fluorescence of the light provides indicators of the cell's size, internal complexity, constituent materials, and other properties. In medicine, flow cytometry is used, for example, in cell-mediated immune testing.

In cellular immunity testing, multiple index values obtained by flow cytometry are analyzed and returned as test results. One of the analysis techniques is gating. Gating is a technique for selecting and analyzing only specific groups from the obtained data. Traditionally, the group to be analyzed was specified by the technician drawing an ellipse or polygon (called a "gate") on a two-dimensional scatter plot. The setting of such gates depends heavily on the technician's experience and knowledge. For this reason, it is difficult for technicians with little experience or knowledge to set appropriate gates.

In response to this, technologies have been proposed to automate gate setting (Patent Documents 1 and 2, etc.). However, conventional technologies use setting methods that use cell density information or rule-based methods, and do not fully utilize the experience and knowledge accumulated by technicians.

Patent No. 6480918 Patent No. 5047803

One possible solution is to estimate the gate area using a learning model that uses deep learning to train gate setting data based on the experience and knowledge accumulated by inspectors. However, the accuracy of the gate area estimation using the learning model is insufficient.

The present invention has been made in consideration of these circumstances. Its purpose is to provide a gate area estimation program and the like that outputs more accurate estimation results when estimating a gate area using a learning model.

The gate region estimation program of the present invention is characterized in that it causes a computer to perform a process of acquiring a scatter plot group including multiple scatter plots obtained from flow cytometry measurements with different measurement items, inputting the acquired scatter plot group into each of multiple learning models that have been trained based on training data including the scatter plot group and the gate region, and outputting the estimated gate region obtained from each of the multiple learning models.

In the present invention, ensemble learning using multiple learning models makes it possible to estimate the gate region with high accuracy.

FIG. 1 is an explanatory diagram illustrating an example of the configuration of an inspection system. 3 is a block diagram showing an example of a hardware configuration of a processing unit; FIG. FIG. 11 is an explanatory diagram illustrating an example of a measurement value DB. FIG. 2 is an explanatory diagram illustrating an example of a feature information DB. FIG. 13 is an explanatory diagram showing an example of a gate DB. FIG. 11 is an explanatory diagram illustrating an example of a threshold DB. FIG. 13 is an explanatory diagram illustrating an example of a confidence level DB. FIG. 11 is an explanatory diagram regarding a generation process of a regression model. 13 is a flowchart illustrating an example of a procedure for generating a regression model. 13 is a flowchart illustrating an example of a procedure for a threshold value determination process. FIG. 11 is an explanatory diagram showing an example of an estimation result of a gate region. FIG. 11 is an explanatory diagram showing an example of an estimation result of a gate region. FIG. 11 is an explanatory diagram showing an example of dispersion degree. FIG. 11 is an explanatory diagram showing an example of dispersion degree. FIG. 11 is an explanatory diagram showing an example of dispersion degree. 13 is a flowchart showing an example of a procedure for gate region estimation processing. 13 is a flowchart illustrating an example of a procedure for determining a degree of confidence. FIG. 11 is an explanatory diagram showing an example of an estimation result of a gate region. FIG. 11 is an explanatory diagram showing an example of an estimation result of a gate region. FIG. 11 is an explanatory diagram showing an example of dispersion degree. FIG. 11 is an explanatory diagram showing an example of dispersion degree. FIG. 11 is an explanatory diagram showing an example of an estimation result of a gate region. FIG. 13 is an explanatory diagram showing an example of an estimation result display screen. FIG. 13 is an explanatory diagram showing an example of an estimation result display screen. FIG. 13 is an explanatory diagram showing an example of an ID list screen. 13 is a flowchart showing another example of the procedure of the gate region estimation process. FIG. 13 is an explanatory diagram showing an example of excluding an outlier gate region. FIG. 13 is an explanatory diagram showing an example of excluding an outlier gate region. FIG. 13 is an explanatory diagram showing an example of 10 small groups. 13 is a flowchart illustrating an example of a procedure for gate selection processing. FIG. 11 is an explanatory diagram showing an example of gate region selection. FIG. 11 is an explanatory diagram showing an example of gate region selection. FIG. 11 is an explanatory diagram showing an example of selection of a gate region based on luminance information. 13 is a flowchart showing another example of a procedure for the gate region selection process. 13 is a flowchart showing another example of a procedure for the gate region selection process.

The following embodiment will be described with reference to the drawings. In the following description, CD45 gating in Leukemia, Lymphoma Analysis (LLA) testing will be used as an example. First, the steps of the LLA testing will be described. The LLA testing roughly includes five steps: 1. Dispensing, 2. Pretreatment, 3. Measurement/Description, 4. Analysis, and 5. Reporting.

The dispensing process is a process of dividing one sample (hereinafter referred to as "ID"). In the LLA test, one ID is dispensed into a maximum of 10 samples for testing. Each dispensed sample is referred to as SEQ. The 10 dispensed samples are also referred to as SEQ1, SEQ2, ..., SEQ10. In the pretreatment process, a common process (such as adjusting the cell concentration) is performed on each SEQ, and a surface marker is individually attached. SEQ1 is the negative control. A negative control is a term used to refer to testing a subject that is already known to produce negative results under the same conditions as the subject whose effect is to be verified, or to that subject. It is also called a negative control. In the test, the results of the subject to be verified are compared with the results of the negative control, and the test results are analyzed from the relative differences.

In the measurement and depiction process, 10 SEQs are measured with a flow cytometer to obtain fluorescence values. For each cell in each SEQ, information consisting of five items, including the measured value, is obtained. The items are FSC, SSC, FL1, FL2, and FL3. FSC indicates the measured value of forward scattered light (FSC). FSC indicates the value of scattered light detected forward of the optical axis of the laser beam. FSC is approximately proportional to the surface area or size of the cell, so it is an index value that indicates the size of the cell. SSC indicates the measured value of side scattered light (SSC). Side scattered light is light detected at an angle of 90° to the optical axis of the laser beam. Most of SSC is scattered when light hits substances inside the cell. SSC is approximately proportional to the granular properties and internal structure of the cell, so it is an index value that indicates the granular properties and internal structure of the cell. FL indicates fluorescence, and here indicates multiple fluorescence detectors equipped in a flow cytometer. Numbers indicate the sequential numbers of the fluorescence detectors. FL1 indicates the first fluorescence detector, and here indicates the name of the item in which the marker information of each SEQ is set as a marker. FL2 indicates the second fluorescence detector, and here indicates the name of the item in which the marker information of each SEQ is set as a marker. FL3 indicates the third fluorescence detector, and here indicates the name of the item in which the marker information of CD45 is set.

The flow cytometer creates two scatter plots for each SEQ and displays the scatter plots on a display etc. For example, one scatter plot has one axis as SSC and the other as FL3. Another scatter plot has one axis as SSC and the other as FSC.

In the analysis process, the technician estimates the disease based on the appearance of the scatter plots and creates gates on each scatter plot that are useful for identifying the disease. Then, for each SEQ, a scatter plot of FL1-FL2 consisting of only cells present within the gate range is created and observed as a marker reaction. In the reporting process, two particularly useful gates are selected for reporting and a report is created.

(Embodiment 1)
1 is an explanatory diagram showing an example of the configuration of an inspection system. The inspection system includes a flow cytometer (gate region estimation device) 10 and a learning server 3. The flow cytometer 10 and the learning server 3 are communicatively connected via a network N. The flow cytometer 10 includes a processing unit 1 that performs various processes related to the operation of the entire device, and a measurement unit 2 that receives a sample and performs measurement by flow cytometry.

The learning server 3 is composed of a server computer, a workstation, etc. The learning server 3 is not a required component of the inspection system. The learning server 3 mainly serves to complement the flow cytometer 10, storing measurement data and learning models as backups. It may also generate learning models and re-learn the learning models on behalf of the flow cytometer 10. In this case, the learning server 3 transmits parameters that characterize the learning model, etc. to the flow cytometer 10. The functions of the learning server 3 may be provided as a cloud service.

Figure 2 is a block diagram showing an example of the hardware configuration of the processing unit. The processing unit 1 includes a control unit 11, a main memory unit 12, an auxiliary memory unit 13, an input unit 14, a display unit 15, a communication unit 16, and a reading unit 17. The control unit 11, the main memory unit 12, the auxiliary memory unit 13, the input unit 14, the display unit 15, the communication unit 16, and the reading unit 17 are connected by a bus B. The processing unit 1 may be separate from the flow cytometer 10. The processing unit 1 is constructed using a PC (Personal Computer), a notebook computer, a tablet computer, etc. The processing unit 1 may also be constructed using a multi-computer consisting of multiple computers, a virtual machine constructed virtually using software, or a quantum computer.

The control unit 11 has one or more arithmetic processing devices such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), etc. The control unit 11 reads out and executes an OS (Operating System) (not shown) and a control program 1P (gate region estimation program) stored in the auxiliary storage unit 13, thereby performing various information processing, control processing, etc. related to the flow cytometer 10. The control unit 11 also includes functional units such as an acquisition unit and an output unit.

The main memory unit 12 is a static random access memory (SRAM), a dynamic random access memory (DRAM), a flash memory, etc. The main memory unit 12 mainly temporarily stores data required for the control unit 11 to execute arithmetic processing.

The auxiliary memory unit 13 is a hard disk or SSD (Solid State Drive) or the like, and stores the control program 1P and various DBs (Databases) necessary for the control unit 11 to execute processing. The auxiliary memory unit 13 stores a measurement value DB 131, a feature information DB 132, a gate DB 133, a first regression model 1341 to a fifth regression model 1345, a threshold DB 135, and a confidence level DB 136. The auxiliary memory unit 13 may be an external storage device connected to the flow cytometer 10. The various DBs and the like stored in the auxiliary memory unit 13 may be stored in a database server or cloud storage connected via a network N.

In this embodiment, ensemble learning is performed using multiple learning models. The output results of the multiple learning models are used to calculate the confidence level for the estimation result of the gate region. In this embodiment, five learning models, the first regression model 1341 to the fifth regression model 1345, are used to perform ensemble learning, but this is not limited to this. The number of learning models may be two to four, or six or more.

The input unit 14 is a keyboard and a mouse. The display unit 15 includes a liquid crystal display panel, etc. The display unit 15 displays various information such as information for performing measurements, measurement results, and gate information. The display unit 15 may be a touch panel display integrated with the input unit 14. The information displayed on the display unit 15 may be displayed on an external display device of the flow cytometer 10.

The communication unit 16 communicates with the learning server 3 via the network N. In addition, the control unit 11 may use the communication unit 16 to download the control program 1P from another computer via the network N, etc., and store it in the auxiliary memory unit 13.

The reading unit 17 reads portable storage medium 1a including CD (Compact Disc)-ROM and DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 17 and store it in the auxiliary storage unit 13. The control unit 11 may also download the control program 1P from another computer via a network N or the like and store it in the auxiliary storage unit 13. Furthermore, the control unit 11 may read the control program 1P from the semiconductor memory 1b.

The database stored in the auxiliary memory unit 13 will be described. FIG. 3 is an explanatory diagram showing an example of a measurement value DB. The measurement value DB 131 stores the measurement values of measurements made by the flow cytometer 10. FIG. 3 shows an example of one record stored in the measurement value DB 131. Each record in the measurement value DB 131 includes a basic section 1311 and a data section 1312. The basic section 1311 includes a reception number column, a reception date column, a test number column, a test date column, a medical record number column, a name column, a gender column, an age column, and a collection date column. The reception number column stores the reception number (identification information) issued when a test request is received. The reception date column stores the date on which the test request is received. The test number column stores the test number issued when the test is performed. The test date column stores the date on which the test was performed. The medical record number column stores the number of the medical record corresponding to the test request. The name column stores the name of the subject who provided the sample. The gender column stores the gender of the subject. For example, if the subject is male, the gender column stores M. If the subject is female, the gender column stores F. The age column stores the age of the subject. The collection date column stores the date the sample was collected from the subject. In the data section 1312, each column stores the measurement value for each cell for the measurement item. Each row stores the measurement value for each measurement item for one cell.

Figure 4 is an explanatory diagram showing an example of a feature information DB. The feature information DB 132 stores information indicating features obtained from the measured values (hereinafter also referred to as "feature information"). The feature information is, for example, a scatter plot or a histogram. The feature information DB 132 includes a reception number column, an examination number column, a sequence number column, a type column, a horizontal axis column, a vertical axis column, and an image column. The reception number column stores the reception number. The examination number column stores the examination number. The sequence number column stores the sequence number of feature information within the same examination. The type column stores the type of feature information. For example, the type is a scatter plot or a histogram as described above. The horizontal axis column stores the item adopted as the horizontal axis in the scatter plot or histogram. The vertical axis column stores the item adopted as the vertical axis in the scatter plot. In the case of a histogram, the vertical axis is the number of cells, so the vertical axis column stores the number of cells. The image column stores the scatter plot or histogram as an image.

Figure 5 is an explanatory diagram showing an example of a gate DB. Gate DB 133 stores information on the gates (gate information) set for the scatter plot. The gate information is information for determining the gate area. The gate information is information on a figure showing the outline of the gate area, a value range of the measurement values included in the gate area, a set of measurement values included in the gate area, etc. It may be pixel coordinate values of a point included in the gate area on the scatter plot image. Here, the gate information is a figure showing the outline of the gate area, and the shape is an ellipse, but is not limited to this. The figure may be a polygon consisting of multiple sides or a figure connecting multiple curves. Gate DB 133 includes a reception number column, an examination number column, a horizontal axis column, a vertical axis column, a gate number column, a CX column, a CY column, a DX column, a DY column, and a θ column. The reception number column stores the reception number. The examination number column stores the examination number. The horizontal axis column stores the item adopted as the horizontal axis in the scatter plot. The vertical axis column stores the item used as the vertical axis in the scatter plot. The gate number column stores the gate sequence number. The CX column stores the x-coordinate value of the center of the ellipse. The CY column stores the y-coordinate value of the center of the ellipse. The DX column stores the value of the major axis of the ellipse. The DY column stores the value of the minor axis of the ellipse. The θ column stores the inclination angle of the ellipse. For example, the inclination angle is the angle between the horizontal axis and the major axis of the ellipse. If a polygon can be set as the gate shape, the gate DB133 stores the coordinate sequence of multiple points that form the polygon.

Figure 6 is an explanatory diagram showing an example of a threshold DB. The threshold DB 135 stores thresholds for indices (dispersion) that indicate the degree of variation in each item value indicating the gate region. The thresholds are used when determining the confidence level of the regression model. The example shown in Figure 6 is a case where the gate region is an ellipse. The threshold DB 135 includes an ID column, a horizontal axis column, a vertical axis column, a CX column, a CY column, a DX column, and a DY column. The ID column stores an ID that identifies a threshold group. The horizontal axis column stores items that are the horizontal axis in a scatter plot. The vertical axis column stores items that are the vertical axis in a scatter plot. The CX column stores a threshold for the center x coordinate value of the ellipse. The CY column stores a threshold for the center y coordinate value of the ellipse. The DX column stores a threshold for the major axis value of the ellipse. The DY column stores a threshold for the minor axis value of the ellipse. The CX column, CY column, DX column, and DY column each include an A column and a B column. Column A stores a threshold A. Column B stores threshold B. A "-" in column B indicates that no value is set. If only threshold A is set, the confidence level of the regression model will be either high or low. If threshold B is set, the confidence level is indicated by a numerical value. For example, if it is smaller than threshold A, the confidence level is 50, and if it is further smaller than threshold B, the confidence level is 70. Note that there may be three or more thresholds.

Figure 7 is an explanatory diagram showing an example of a confidence level DB. The confidence level DB 136 stores the confidence level of the estimation result of the gate area by the regression model. The confidence level DB 136 includes a reception number column, an examination number column, a gate number column, a CX column, a CY column, a DX column, a DY column, a gate total column, and a total column. The reception number column stores the reception number. The examination number column stores the examination number. The gate number column stores the gate sequence number. The reception number column, the examination number column, and the gate number column enable correspondence with the gate DB 133. The CX column stores the confidence level of the center x coordinate value of the ellipse. The CY column stores the confidence level of the center y coordinate value of the ellipse. The DX column stores the confidence level of the major axis length of the ellipse. The DY column stores the confidence level of the minor axis length of the ellipse. The gate total column stores the confidence level for each gate. The total column stores the confidence level for each examination. In the example shown in Figure 7, the confidence level value is high or low.

Next, the preparation process will be described. The preparation process is a process carried out before actual operation begins. Figure 8 is an explanatory diagram regarding the regression model generation process. Five learning models, the first regression model 1341 to the fifth regression model 1345, are generated. Figure 8 shows the process of generating the first regression model 1341 to the fifth regression model 1345 by performing machine learning. The basic process is similar for all learning models. The basic process will be described using the first regression model 1341 as a representative example.

In the flow cytometer 10 of this embodiment, the processing unit 1 performs deep learning to learn the appropriate gate feature amount for the scatter plot image created based on the measurement results obtained by the measurement unit 2, thereby generating a first regression model 1341 that inputs multiple scatter plot images (scatter plot group) and outputs gate information. The multiple scatter plot images are multiple scatter plot images with at least one axis item different. For example, two scatter plot images are a scatter plot image with a horizontal axis of SSC and a vertical axis of FL3, and a scatter plot image with a horizontal axis of SSC and a vertical axis of FSC. Three or more scatter plot images may be input. The neural network is, for example, a CNN (Convolution Neural Network). The first regression model 1341 has multiple feature extractors that respectively learn the feature amounts of each scatter plot image, a combiner that combines the feature amounts output by each feature extractor, and multiple estimators that estimate and output each item of the gate information (center X coordinate, center Y coordinate, major axis, minor axis, tilt angle) based on the combined feature amount. Note that instead of the scatter plot image, a set of measurement values on which the scatter plot is based may be input to the first regression model 1341 .

Each feature extractor includes an input layer and an intermediate layer. The input layer has a plurality of neurons that accept input of pixel values of each pixel included in the scatter plot image, and passes the input pixel values to the intermediate layer. The intermediate layer has a plurality of neurons, extracts features from the scatter plot image, and passes them to the output layer. For example, when the feature extractor is a CNN, the intermediate layer has a configuration in which a convolution layer that convolves the pixel values of each pixel input from the input layer and a pooling layer that maps the pixel values convolved in the convolution layer are alternately connected, and finally extracts image features while compressing pixel information. Instead of providing a feature extractor that inputs a scatter plot image for each image, a configuration in which multiple scatter plot images are input to one feature extractor may be used.

In this embodiment, the first regression model 1341 is described as being a CNN, but the first regression model 1341 is not limited to a CNN and may be a trained model constructed using other learning algorithms, such as a neural network other than CNN, a Bayesian network, or a decision tree.

The processing unit 1 performs learning using training data in which multiple scatter plot images are associated with correct values of gate information corresponding to the scatter plots. For example, as shown in FIG. 8, the training data is data in which gate information is labeled for multiple scatter plot images. For simplicity, two types of scatter plots are considered as one set of scatter plots. Also, although the following description assumes that one gate is provided for one set of scatter plots, multiple gates may be provided. In this case, the gate information includes a value indicating usefulness.

The processing unit 1 inputs the two scatter plot images, which are training data, into different feature extractors. The features output by each feature extractor are combined by a combiner. Combinations by the combiner can be performed in a number of ways, including simply combining the features (Concatenate), adding values indicating the features (Add), and selecting the feature with the maximum value (Maxpool).

Based on the combined features, each estimator outputs gate information as an estimation result. A combination of values output by each estimator forms a set of gate information. There may be multiple sets of gate information to be output. In this case, a number of estimators corresponding to the multiple sets are provided. For example, when outputting gate information with the first priority and gate information with the second priority, the number of estimators in Figure 8 will be 5 to 10.

The processing unit 1 compares the gate information obtained from the estimator with the information labeled for the scatter plot image in the training data, i.e., the correct value, and optimizes the parameters used in the calculation process in the feature extractor and estimator so that the output value from the estimator approaches the correct value. The parameters in question are, for example, the weights (coupling coefficients) between neurons and the coefficients of the activation functions used in each neuron. There are no particular limitations on the method of optimizing the parameters, but for example, the processing unit 1 optimizes various parameters using the backpropagation method. The processing unit 1 performs the above processing on the data for each test included in the training data, and generates the first regression model 1341.

Next, the processing performed by the control unit 11 of the processing unit 1 will be described. FIG. 9 is a flowchart showing an example of the procedure of the regression model generation processing. The control unit 11 acquires the inspection history (step S1). The inspection history is an accumulation of past inspection results, and is past measurement values stored in the measurement value DB 131. The control unit 11 selects one history to be processed (step S2). The control unit 11 acquires feature information corresponding to the selected history (step S3). The feature information is, for example, a scatter plot. The feature information is acquired from the feature information DB 132. If the feature information is not stored, it may be generated from the measurement value. The control unit 11 acquires gate information corresponding to the selected history (step S4). The gate information is acquired from the gate DB 133. The control unit 11 learns the first regression model 1341 using the acquired feature information and gate information as training data (step S5). The control unit 11 determines whether there is an unprocessed inspection history (step S6). When the control unit 11 determines that there is an unprocessed inspection history (YES in step S6), the control unit 11 returns the process to step S2 and performs a process related to the unprocessed inspection history. When the control unit 11 determines that there is no unprocessed inspection history (NO in step S6), the control unit 11 stores the first regression model 1341 (step S7) and ends the process.

The second regression model 1342, the third regression model 1343, the fourth regression model 1344, and the fifth regression model 1345 are generated by a process similar to that described using FIG. 8 and FIG. 9. However, the first regression model 1341 to the fifth regression model 1345 are regression models with different generation conditions, for example, by changing the training data, network structure, and hyperparameters. For the training data, the training data of the five learning models is made different by increasing the number of data by the restoration sampling method used in data augmentation and bootstrap method. For the network structure, the number of input layers and output layers is changed. In addition, a different learning model is generated from a certain learning model by fine-tuning. For the hyperparameters, the settings of the number of layers of the intermediate layer, the number of nodes in each layer, weights, loss function, optimization function, learning rate, batch size, etc. are made different.

After generating the first regression model 1341 to the fifth regression model 1345, a threshold value for judging the confidence level is determined. FIG. 10 is a flowchart showing an example of the procedure of the threshold determination process. The control unit 11 acquires test data (step S11). The control unit 11 inputs the acquired test data to each regression model (step S12). The control unit 11 acquires an estimated output from each regression model (step S13). The estimated output is a parameter value indicating the gate region estimated by each regression model. If the gate region is an ellipse, the parameters are the center coordinates (Cx, Cy), the length of the major axis and the minor axis (Dx, Dy), and the angle (θ) between the major axis and the x-axis. If the gate region is a polygon, the parameters are the coordinate values of each vertex. The control unit 11 calculates the degree of dispersion for each parameter from the values output from each regression model (step S14). An example of the degree of dispersion is the standard deviation. The control unit 11 determines whether there is unprocessed test data (step S15). When the control unit 11 determines that there is unprocessed test data (YES in step S15), it returns the process to step S11 and processes the unprocessed test data. When the control unit 11 determines that there is no unprocessed test data (NO in step S15), it determines a threshold value for each parameter (step S16). The threshold value means the limit of the allowable variation in the output value. The threshold value is determined by a statistical method. Alternatively, the threshold value is determined, for example, by a veteran inspector based on the value of the degree of dispersion for each test data. The control unit 11 stores the determined threshold value (step S17) and ends the threshold value determination process. The threshold value may be adjusted for each environment in which the flow cytometer 10 is operated, for example, for each testing institution. The degree of dispersion may be variance, unbiased variance, or average deviation, in addition to the standard deviation.

A specific example of the threshold determination process will be described. Figures 11A and 11B are explanatory diagrams showing an example of an estimated output of a gate area. Figure 11 is an example of a scatter plot in CD45 gating. Figure 11A shows an example of an input scatter plot. In Figure 11A, the horizontal axis is SSC (Side Scattered Light) and the vertical axis is FL3 (FL = Fluorescence: fluorescence detector, 3 means third). Figure 11B shows the results of five regression models estimating the gate area in the input scatter plot. The vertical and horizontal axes in Figure 11B are the same as in Figure 11A. Here, an example of estimating one elliptical gate area is shown. In Figure 11B, the gate area is overwritten on the input scatter plot. AI-1 indicates the estimation result of the first regression model 1341, AI-2 indicates the estimation result of the second regression model 1342, and so on, with AI-5 indicating the estimation result of the fifth regression model 1345. Below the scatter plot, the parameters of the ellipse that is the gate region are listed. From the top, they are the central X coordinate (Cx), central Y coordinate (Cy), major axis length (Dx), minor axis length (Dy), and inclination angle (θ).

Figure 12 is an explanatory diagram showing an example of dispersion. Dispersion is defined as standard deviation (SD). The table on the left of Figure 12 reproduces the values shown in Figure 11B. Note that the inclination angle is excluded from the SD calculation items because it does not affect the confidence of the estimation result even if the SD becomes large, for example, when the shape of the gate region is a perfect circle. The SD threshold is calculated for the results of multiple test data and determined based on the calculation results.

Figures 13A and 13B are explanatory diagrams showing examples of the degree of scatter. Figures 13A and 13B show the estimation results when different test data is input. In both Figures 13A and 13B, the horizontal axis is SSC and the vertical axis is FL3. In Figures 13A and 13B, the gate regions output by the five regression models are overlaid on the input scatter plot. The numbers on the right side of the scatter plot indicate the center coordinates that specify the ellipse and the SD of the major and minor axes. Figure 13A is an example where the estimation results have a small variation, and Figure 13B is an example where the estimation results have a large variation. From Figures 13A and 13B, it is considered that the thresholds should be set to Cx: 5.6 or more, Cy: between 10.9 and 36.8, Dx: 12.3 or more, and Dy: 6.4 or more. It is desirable to determine the final thresholds for each parameter taking into account not only the two test data but also the SD for other data. Once the thresholds for each parameter have been determined, the preparation process is complete. In the above-mentioned threshold determination process, it is assumed that each regression model outputs one gate as an estimation result of the gate region, but multiple gates may be output. In the case of multiple gates, a threshold is determined for each gate. In the case where a first gate, a second gate, and a third gate are output as the gate region, the degree of dispersion is calculated for the first gate of each regression model, and a threshold is determined.

Next, the operation process will be explained. In the following explanation, the degree of dispersion is the standard deviation (SD). Furthermore, if the SD is equal to or less than a threshold, the confidence level is set to high. If the SD exceeds the threshold, the confidence level is set to low. Furthermore, each regression model is assumed to output estimation results for multiple gate regions.

Figure 14 is a flowchart showing an example of the procedure for gate area estimation processing. The control unit 11 acquires a scatter diagram (step S31). The scatter diagram here is a coordinate sequence of points showing the measurement results, as well as the measurement items on the horizontal axis and the measurement items on the vertical axis. The control unit 11 inputs the acquired scatter diagram to each regression model (step S32). The control unit 11 acquires the estimated output of the gate area of each regression model (step S33). The control unit 11 calculates the degree of dispersion, here the standard deviation, for each regression model, each gate, and each parameter (step S34). The control unit 11 determines the degree of confidence (step S35). The control unit 11 stores the results (step S36). The control unit 11 stores the gate area estimation results in the gate DB 133, and stores the degree of confidence in the confidence DB 136. The control unit 11 ends the gate area determination process.

Figure 15 is a flowchart showing an example of the procedure of the confidence level determination process. The confidence level determination process corresponds to step S35 in Figure 14. The control unit 11 selects the gate region to be processed (step S51). The control unit 11 selects the parameters to be processed (variables such as Cx, Cy, Dx, Dy, etc.) (step S52). The control unit 11 judges whether the standard deviation of the parameters is equal to or less than a threshold value (step S53). If the control unit 11 judges that the standard deviation of the parameters is equal to or less than a threshold value (YES in step S53), it judges whether all parameters have been processed (step S54). If the control unit 11 judges that all parameters have not been processed (NO in step S54), it returns the process to step S52 and performs processing on the unprocessed parameters. If the control unit 11 judges that all parameters have been processed (YES in step S54), it stores in the temporary storage area that the confidence level of the gate to be processed is high (step S55). The temporary storage area is provided in the main storage unit 12 or the auxiliary storage unit 13. When the control unit 11 determines that the standard deviation of the parameters exceeds the threshold (NO in step S53), it stores in the temporary storage area that the confidence level of the gate being processed is low (step S56). The control unit 11 determines whether or not all gates have been processed (step S57). When the control unit 11 determines that all gates have not been processed (NO in step S57), it returns the process to step S51 and processes the unprocessed gates. When the control unit 11 determines that all gates have been processed (YES in step S57), it refers to the temporary storage area and determines whether or not the confidence levels of all gates are high (step S58). When the control unit 11 determines that the confidence levels of all gates are high (YES in step S58), it stores in the temporary storage area that the confidence level (overall confidence level) for the estimation result of the gate area in the scatter diagram being processed is high (step S59). When the control unit 11 determines that the confidence level of all gates is not high, but that the confidence level of some gates is low (NO in step S58), it stores in a temporary storage area that the confidence level of the entire estimation result for the scatter diagram being processed is low (step S60).The control unit 11 returns the process to the caller.

A specific example of the gate region estimation process will be described. Figures 16A and 16B are explanatory diagrams showing an example of the estimation result of the gate region. Like Figure 11, Figure 16 is an example of a scatter plot in CD45 gating. Figure 16A is an example of an input scatter plot, and Figure 16B shows the results of the gate region in the input scatter plot estimated by five regression models. In Figures 16A and 16B, the horizontal axis is SSC and the vertical axis is FL3. Here, one of the gate regions obtained as an estimation result is displayed. In Figure 16B, the gate region is overwritten on the input scatter plot. AI-1 shows the estimation result of the first regression model 1341, AI-2 shows the estimation result of the second regression model 1342, and so on, and AI-5 shows the estimation result of the fifth regression model 1345. The parameters of the ellipse that is the gate region are written below the scatter plot. From top to bottom, these are the central X coordinate (Cx), central Y coordinate (Cy), major axis length (Dx), minor axis length (Dy), and inclination angle (θ).

Figures 17A and 17B are explanatory diagrams showing examples of the degree of scatter. Figures 17A and 17B show the estimation results when different scatter plots are input. In both Figures 17A and 17B, the horizontal axis is SSC and the vertical axis is FL3. Figure 17A shows the five plots shown in Figure 16B drawn as one figure. That is, Figure 17A shows the gate regions output by the five regression models superimposed on the input scatter plot. Figure 17B is the same as Figure 17A. The numbers on the right side of the scatter plot indicate the center coordinates that specify the ellipse and the SD of the major and minor axes. Figure 17A is an example where the estimation results have a small variation, and Figure 17B is an example where the estimation results have a large variation. Here, if the SD of each of Cx, Cy, Dx, and Dy is 20 or less, the confidence level is high, and if there is even one item that exceeds 20, the confidence level is low. In this case, the results shown in Figure 17A have all SDs of 20 or less, so the confidence level is determined to be high. On the other hand, in the result shown in FIG. 17B, since the SD of Cx and Dx exceeds 20, the confidence level is determined to be low.

Next, an example of the confidence level determination when multiple gate regions are estimated is shown. Figure 18 is an explanatory diagram showing an example of the gate region estimation result. Figure 18 shows the estimation results of three gate regions, G1 to G3. Each gate region is drawn on a scatter plot with the gate regions output by five regression models overlapping. The table below the scatter plot shows the SD of each parameter. Gates G1 and G2 are determined to have high confidence levels because all SDs are 20 or less. Gate G3 is determined to have low confidence levels because the SD of Cx exceeds 20. When multiple gate regions are estimated, if the confidence levels of the estimation results of all gate regions are high, the overall confidence level is high, and if there is even one low confidence level in the estimation results of the gate regions, the overall confidence level is low. If judged by this definition, the estimation results shown in Figure 18 are determined to have low confidence levels overall.

Next, the screen display of the estimation result of the gate area will be described. Figures 19A and 19B are explanatory diagrams showing examples of the estimation result display screen. Figure 19A is an example of a screen when the confidence level is high. Figure 19B is an example of a screen when the confidence level is low. The estimation result display screen includes a scatter diagram 191, confidence level 192, and confidence level icon 193. The scatter diagram 191 displays the estimation result of the gate area in a scatter diagram. Here, the gate area to be displayed is one area selected by a predetermined algorithm from the five estimated areas output by the five regression models. The confidence level 192 displays the confidence level for the entire judgment result. In Figure 19, high confidence level is displayed as High, and low confidence level is displayed as Low. The confidence level icon 193 expresses the confidence level with a face icon. If the confidence level is high, a smiling icon is displayed, and if the confidence level is low, a troubled face icon is displayed. In addition, in the scatter diagram 191, the gate regions to be displayed may be all five estimated regions output by the five regression models, similar to FIG.

Figure 20 is an explanatory diagram showing an example of an ID list screen. The ID list screen displays a list of IDs assigned to each test in association with the confidence level of the estimation results for the gate area. The ID list screen includes an ID display 201 and a confidence level display 202. The ID display 201 displays, for example, a reception number. The confidence level display 202 displays, for example, A when the confidence level is high and a when the confidence level is low. When an ID is selected from the ID display 201, the estimation result display screen shown in Figure 19 is displayed.

In this embodiment, the gate region estimation result is output with a confidence level attached. This has the effect of enabling operation suited to the on-site operating environment, such as by referring to the confidence level display 202 and having a highly skilled inspector give priority to carefully checking "A" and taking time to analyze "a" which is likely to be incorrect.

In this embodiment, one scatter plot is input to each regression model, but this is not limited to this and two or more scatter plots may be input. Also, the scatter plot is not limited to two dimensions and may be three or more dimensions.

The degree of dispersion is the standard deviation of the parameters of the figure representing the gate region, and in the case of an ellipse, the standard deviation of the central coordinates and the length of the semimajor axis, but is not limited to this. The area of the gate region estimated by five regression models may also be taken as the degree of dispersion. For example, when five estimated gate regions are superimposed on a scatter plot, the area of the region containing the five regions and the area of the region where the five overlap are calculated, and the ratio of the latter area to the former area is taken as the degree of dispersion. In this case, the smaller the value, the greater the variation. 1 is the maximum value, when all five regions match.

(Embodiment 2)
This embodiment relates to a method for selecting one gate region to be presented to a user from five gate regions estimated by five regression models. In ensemble learning, the output results of multiple learning models are combined to obtain a final result. Since multiple learning models are used in ensemble learning, it has the effect of reducing output fluctuations. It is known to be particularly effective in neural networks, where accuracy fluctuations are likely to occur with each learning.

Ensemble learning is particularly effective in neural networks, which are prone to fluctuations in accuracy with each learning, and is a technique used in various competitions. However, the output of the regression model that estimates the gate area is multi-output and needs to be evaluated in a composite manner. Simply combining the outputs of multiple models by averaging them together does not improve accuracy. When the gate area is estimated by multiple regression models trained under different conditions, the estimation results differ due to differences in learning. Therefore, when the average value of each parameter is calculated from the estimation results output by each regression model and the final gate area to be displayed is determined, it is highly likely that it will be unclear to the inspector which part of the scatter diagram is the gate to be surrounded. Therefore, in this embodiment, one of the five gate areas estimated by the five regression models is selected as the optimal gate area to be presented to the user. In this embodiment, the hardware configuration and the generation process of the first regression model 1341 to the fifth regression model 1345 are the same as those in the first embodiment. In the following explanation, the differences from the first embodiment will be mainly explained.

The preparation process is the same as in the first embodiment, so the description will be omitted. The operation process will be described below. FIG. 21 is a flowchart showing another example of the procedure of the gate region estimation process. A process for selecting a gate region is added to the process shown in FIG. 14. The control unit 11 acquires a scatter diagram (step S71). The control unit 11 inputs the acquired scatter diagram to each regression model (step S72). The control unit 11 acquires the estimated output of the gate region of each regression model (step S73). The control unit 11 removes the gate region including an outlier from the selection target (step S74). The median is calculated for each parameter of the gate region output by the five regression models. The gate region that deviates from the median in even one item is removed from the selection target. Note that step S74 is not a required process and may be omitted. The control unit 11 calculates the feature amount for each gate region (step S75). The feature amount is the number of cells in the gate, the area of the gate region, the cell density in the gate, the cell purity in the gate, etc. The feature amount will be supplemented later. The control unit 11 selects the optimal gate based on the feature amount (step S76). The control unit 11 calculates the degree of dispersion (step S77). The control unit 11 determines the confidence level (step S78). The contents of steps S77 and S78 are the same as those in the first embodiment, so the description will be omitted. The control unit 11 stores the selected gate region and the confidence level (step S79), and ends the gate region estimation process.

Next, an example of excluding gate regions containing outliers from selection is shown. Figures 22A and 22B are explanatory diagrams showing an example of excluding outlier gate regions. Figure 22A shows the gate regions output by five regression models superimposed on a scatter plot. Of the gate regions, gate region Gj is a different size from the other gate regions, and is therefore excluded from selection as a gate region containing an outlier. Figure 22B is a scatter plot that displays only gate region Gj that was excluded from selection.

Next, cell purity, which is one of the feature quantities, will be explained. In the test, it is desirable that each gate contains only one type of cell. The cell type can be roughly estimated from the information on FSC, SSC, and CD45. Therefore, the cell population is roughly classified into classes based on the information on FSC, SSC, and FL3, and the cell purity is defined as which class is most abundant in the target gate and what percentage of cells of that class are contained in the gate. As a specific example, in the distribution of FSC, SSC, and FL3, a three-dimensional automatic clustering method, k-means, is applied to create n small groups. n is a natural number. Here, n=10. Figure 23 is an explanatory diagram showing an example of 10 small groups. The pentagonal marks indicate the centers of each small group used in k-means. In Figure 23, the horizontal axis is a two-dimensional display with SSC and the vertical axis is FL3, but in reality it is a three-dimensional clustering with FSC as the axis normal to the paper. In Figure 23, there are many cells of class Cb in the gate area G. Therefore, the cell purity of the gate region G is the proportion of class Cb contained in the gate region G. That is, the cell purity is calculated by finding the class of cells contained most frequently in the target gate region, and dividing the number of cells of that class contained in the target gate region by the total number of cells of that class.

Figure 24 is a flowchart showing an example of the procedure of the gate selection process. The gate selection process corresponds to step S76 in Figure 21. The control unit 11 performs cell clustering (step S91). For example, as described above, in the distribution of FSC, SSC, and FL3, three-dimensional automatic clustering is performed by k-means to divide the cells into 10 classes. The control unit 11 selects a gate region to be processed from the five gate regions output by each of the five regression models (step S92). The control unit 11 obtains the number of cells for each class contained in the selected gate region, and identifies the class with the largest number of cells (step S93). The control unit 11 calculates the cell purity (step S94). The control unit 11 determines whether or not there is an unprocessed gate region (step S95). If the control unit 11 determines that there is an unprocessed gate region (YES in step S95), the process returns to step S92 and processes the unprocessed gate region. When the control unit 11 determines that there is no unprocessed gate region (NO in step S95), it selects a gate region to be output (step S96). The control unit 11 selects the gate region with the highest cell purity from among the five gate regions. The control unit 11 ends the gate selection process.

Figures 25A and 25B are explanatory diagrams showing an example of gate region selection. Figure 25A shows gate regions output by five regression models superimposed on a scatter plot. The numbers on the right side of Figure 25A indicate the cell purity of each gate region. In this case, the cell purity of the gate region output by AI-3, i.e., the third regression model 1343, is the highest at 0.66, so that gate region is selected. Figure 25B is an example of a scatter plot in which only the selected gate region is superimposed. Similar processing is also possible when outputting multiple gate regions. Details will be described later.

Next, we will discuss features other than cell purity. Cell count is the number of cells contained in the gate region. Area is the area of the shape that shows the gate region in a two-dimensional scatter plot. Cell density is the number of cells divided by the area.

Measurement values other than the analysis axis may be used as features. The analysis axis is the horizontal and vertical axis of a two-dimensional scatter plot. In flow cytometry testing, cell tumors are identified from measurement values in all dimensions. Therefore, even within a gate, measurement values in other dimensions (other than the analysis axis) can serve as indicators for determining the optimal gate to narrow down cell tumors.

In the above example, the analysis axes are SSC and FL3 (CD45). In this case, examples of measurements other than the analysis axes are FSC and FL1 (CD34). At this time, CD34 is measured at FL1, and the optimal gate is selected based on the average value. For example, the average value of AI-1 is 0.21, the average value of AI-2 is 0.16, the average value of AI-3 is 0.18, and the average value of AI-4 is 0.20. The gate area output by AI-5 is excluded as an outlier gate area. At this time, the gate area output by AI-1 is selected.

Image information may be used as a feature. When selecting an optimal gate based on the cell density within the gate, bias in cell distribution may affect the selection of an appropriate gate. To avoid this, the distribution situation is converted into an image to obtain the feature. An example of processing is explained below. To treat the contents of the scatter plot as image information, here the luminance L, the areas where cells exist are represented by black pixels (L<255) and the rest by white pixels (L=255).

Figure 26 is an explanatory diagram showing an example of gate area selection based on luminance information. The top left of Figure 26 is a scatter plot with a superimposed display of the gate area output by AI-1. The bottom left of Figure 26 shows the features of the gate area output by AI-1. The top right of Figure 26 is a scatter plot with a superimposed display of the gate area output by AI-2. The bottom right of Figure 26 shows the features of the gate area output by AI-2. Note that the features of the gate areas output by AI-3 to AI-5 are smaller than those of the gate areas output by AI-1 and AI-2, so there is no possibility of any of them being selected. In the example of Figure 26, the gate area to be selected is the area output by AI-1.

In the example of Figure 26, if cell density is used as the feature for selecting the gate region, AI-1 is 1.0 and AI-2 is 1.1, so the gate region output by AI-2 will be selected. However, if the black-white ratio (= number of white pixels/number of black pixels) is used as the feature for selecting the gate region, AI-1 will be 0.7 and AI-2 will be 0.5, so the gate region output by AI-1 will be selected. In the example of Figure 26, there is a bias in the cell distribution, so using cell density as a feature would result in an inappropriate selection. The black-white ratio reflects cell density to some extent while reducing the influence of cell number, resulting in an appropriate selection result.

In the above description of this embodiment, the case where one gate area is obtained has been described. To obtain multiple gate areas, the processing needs to be expanded, and there are two possible methods for this expansion. Method 1 is a method in which an optimal output is selected from multiple AIs that output all gates. For example, this method selects the estimation results output by AI-1 for all gate areas.

Figure 27 is a flowchart showing another example of the procedure of the gate region selection process. Figure 27 shows the selection process when multiple gate regions are set in one scatter plot, and is the process by the above-mentioned method 1. The control unit 11 selects a regression model to be processed (step S111). The control unit 11 selects a gate region to be processed from the multiple gate regions output by the selected regression model (step S112). The control unit 11 judges the confidence level of the selected gate region (step S113). The confidence level is judged as described above. The control unit 11 stores the confidence level in a temporary storage area (step S114). The control unit 11 judges whether or not there is an unprocessed gate region (step S115). If the control unit 11 judges that there is an unprocessed gate region (YES in step S115), the process returns to step S112 and processes the unprocessed gate region. If the control unit 11 judges that there is no unprocessed gate region (NO in step S115), the control unit 11 judges the confidence level of the entire gate region output by the selected regression model (step S116). For example, when there are three gate regions to be set and the confidence levels of each gate region are high, high, and low, the number of highs, 2, is set as the overall confidence level. The control unit 11 stores the confidence levels in a temporary storage area (step S117). The control unit 11 judges whether there is an unprocessed regression model (step S118). When the control unit 11 judges that there is an unprocessed regression model (YES in step S118), the process returns to step S111 and processes the unprocessed regression model. When the control unit 11 judges that there is no unprocessed regression model (NO in step S118), the control unit 11 selects a regression model based on the confidence level for each regression model (step S119). The control unit 11 outputs the gate region output by the selected regression model (step S120) and ends the process.

Method 2 is a method of selecting one optimal output for each gate from multiple AIs that output all gates. For example, the first gate selects the estimation result output by AI-1, the second gate selects the estimation result output by AI-4, and the third gate selects the estimation result output by AI-5. Method 1 has the advantage that there is little overlap between gates because one AI outputs each gate, but it has the disadvantage that if one gate is misaligned, the other gates are likely to be misaligned as well. Method 2 has the advantage that there is little impact from misalignment of other gates because selection is made gate by gate, but it has the disadvantage that gates are likely to overlap if selected under conditions where there is no information on other gates. However, the impact of this disadvantage can be mitigated by ingeniously adding information on other gates.

A process to reduce the impact of the shortcomings of method 2 is described below. The following conditions are used in gate selection to compare selected gates with and without adding other gate information. Condition 1: The feature "cell purity" is used as the criterion for determining the optimal gate. Condition 2: A gate is selected according to its usefulness (as analysis gates generally have a usefulness). At that time, cells contained in other gates are excluded from the feature calculation. Gate selection is performed when only condition 1 is applied, and when conditions 1 and 2 are applied. If there is no gate overlap in either case, that selection result is adopted. Note that the feature in condition 1 can be a feature other than cell purity as long as it is related to the gate region.

Figure 28 is a flowchart showing another example of the procedure of the gate region selection process. Figure 28 shows the selection process when multiple gate regions are set in one scatter plot, and is the process by the above-mentioned method 2. The control unit 11 selects a gate region to be processed from multiple gate regions (step S131). For example, when three gate regions are set, they are designated as the first gate, the second gate, and the third gate, respectively. The order numbers are determined by the usefulness, such as the order of importance in reporting the test results. At this time, the control unit 11 processes the first gate, the second gate, and the third gate in this order. The control unit 11 selects a regression model to be processed (step S132). The control unit 11 calculates the feature amount, for example, the cell purity, for the gate region output by the selected regression model (step S133). The control unit 11 determines whether or not there is an unprocessed regression model (step S134). When the control unit 11 determines that there is an unprocessed regression model (YES in step S134), the process returns to step S132 and processes the unprocessed regression model. When the control unit 11 determines that there is no unprocessed regression model (NO in step S134), it selects a gate region to be finally output from among the gate regions output by any of the regression models based on the feature amount (step S135). The control unit 11 stores information on the selected gate region in a temporary storage area (step S136). The control unit 11 determines whether or not there is an unprocessed gate region (step S137). When the control unit 11 determines that there is an unprocessed gate region (YES in step S137), it returns the process to step S131 and processes the unprocessed gate region. When the control unit 11 determines that there is no unprocessed gate region (NO in step S137), it outputs all gate regions based on the selection information stored in the temporary storage area (step S138) and ends the process.

In this embodiment, it is possible to select the optimal gate region from the estimated gate region results output by multiple regression models. Note that, in the above embodiment, CD45 gating in LLA is used as an example, but CD45 gating in malignant lymphoma analysis (MLA) tests can also be performed in a similar procedure.

The technical features (constituent elements) described in each embodiment can be combined with each other, and by combining them, new technical features can be formed.
The embodiments disclosed herein are illustrative in all respects and should not be considered as limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the scope and meaning equivalent to the claims.

REFERENCE SIGNS LIST 10 Flow cytometer 1 Processing section 11 Control section 12 Main memory section 13 Auxiliary memory section 131 Measurement value DB
132 Feature Information DB
133 Gate DB
1341 First regression model 1342 Second regression model 1343 Third regression model 1344 Fourth regression model 1345 Fifth regression model 135 Threshold DB
136 Confidence DB
14 Input unit 15 Display unit 16 Communication unit 17 Reading unit 1P Control program 1a Portable storage medium 1b Semiconductor memory 2 Measurement unit 3 Learning server

Claims (13)

A group of scatter plots including multiple scatter plots obtained by flow cytometry measurements with different measurement items is obtained,
The obtained scatter plot group is input to each of a plurality of learning models that have been trained based on training data including the scatter plot group and the gate region;
A gate region estimation program for causing a computer to perform a process of outputting estimated gate regions obtained from each of the plurality of learning models. The gate region estimation program according to claim 1 , further comprising: determining a degree of confidence based on a plurality of the estimated gate regions. 3. The gate region estimation program according to claim 2, further comprising: determining a degree of confidence based on a degree of dispersion of each of a plurality of variables indicating each of the plurality of estimated gate regions. The gate region estimation program according to claim 3, characterized in that the degree of dispersion of each variable is compared with a plurality of predetermined thresholds, a degree of confidence for each variable is determined in a plurality of stages, and the degree of confidence of the estimated gate region is determined based on the degree of confidence for each variable. 5. The gate region estimation program according to claim 2, further comprising: determining a degree of confidence of an estimated gate region for each measurement of a plurality of samples; and outputting identification information for identifying the samples in association with the determined degree of confidence. selecting one learning model based on the estimated gate regions obtained from each of the plurality of learning models;
The gate region estimation program according to claim 1 , further comprising: outputting an estimated gate region output by the selected learning model. The gate region estimation program according to claim 6 , further comprising: selecting a learning model based on the number of cells included in each of the estimated gate regions output by a plurality of the learning models. The gate region estimation program according to claim 6 , further comprising: selecting a learning model based on the areas of the estimated gate regions output from a plurality of said learning models. Clustering the measured cells based on a scatter diagram obtained from the plurality of measurement items;
The gate region estimation program according to claim 6, further comprising: selecting a learning model for each of the estimated gate regions output by a plurality of the learning models based on a cell purity determined using a result of the clustering. obtaining a plurality of estimated gate regions from each of the plurality of learning models;
Selecting one learning model for each group including a plurality of the estimated gate regions related to each other;
10. The gate region estimation program according to claim 1, further comprising:outputting the estimated gate region output by each of the selected learning models. determining a confidence level based on the plurality of estimated gate regions;
selecting one learning model based on the estimated gate regions obtained from each of the plurality of learning models;
2. The gate region estimation program according to claim 1, further comprising: outputting an estimated gate region output by the selected learning model and the degree of confidence. The computer
A group of scatter plots including multiple scatter plots obtained by flow cytometry measurements with different measurement items is obtained,
The obtained scatter plot group is input to each of a plurality of learning models that have been trained based on training data including the scatter plot group and the gate region;
and outputting the estimated gate regions obtained from each of the plurality of learning models. An acquisition unit that acquires a scatter plot group including a plurality of scatter plots obtained by flow cytometry measurements having different measurement items;
and an output unit that inputs the acquired scatter plot group to each of a plurality of learning models that have been trained based on training data including a scatter plot group and a gate region, and outputs the estimated gate region obtained from each of the plurality of learning models.

Images