JP7684143B2 - IMAGE PROCESSING METHOD, PROGRAM, AND RECORDING MEDIUM - Google Patents

Description

This invention relates to an image processing method for analyzing images of cell clusters, and in particular to a technique for separating cell cluster regions and protrusion regions from images of cancer cell spheroids.

In the fields of cancer research and drug discovery screening, cancer cells are artificially cultured to produce cancer cell spheroids, which are then observed and evaluated. It is known that highly malignant cancer cells form protrusions called invadopodia, and even in artificial culture environments that mimic the in vivo environment, the formation of invadopodia extending radially from the spheroids has been observed. Changes in the number and length of invadopodia can be used as indicators for evaluating the efficacy of chemicals against cancer cells.

Until now, invadopodia have been measured using images of cancer cell spheroids obtained using an optical microscope, but because spheroids with a three-dimensional structure are observed as two-dimensional images, the measurement accuracy is insufficient. To address this issue, for example, Patent Document 1 proposes a method for evaluating the protrusion formation ability of cells by acquiring cross-sectional images by imaging a sample with a confocal microscope to which fluorescent labels corresponding to the protrusions have been added, and then constructing a three-dimensional image from the cross-sectional images of multiple cross sections to measure the protrusions.

JP 2019-088200 A

The above conventional techniques require the step of fluorescently staining the sample. In addition, because cancer cell spheroids are usually cultured embedded in a gel-like medium, adding a fluorescent reagent causes the gel to self-luminesce, which can be a factor that hinders observation. Furthermore, there is a risk that variations in the degree of staining may affect the quantitative evaluation results.

Here, for example, by using Optical Coherence Tomography (OCT) technology, it is possible to obtain a three-dimensional image without staining the sample. This eliminates the need for staining and also reduces the variation in evaluation due to staining variations. On the other hand, OCT images cannot distinguish between the cell masses and protrusions that make up the cancer cell spheroids. Therefore, in order to perform quantitative measurements, it is necessary to separate these structures from each other in the image.

For example, some general-purpose image processing application software has the function of separating such tree-like structures from the imaged object and quantitatively measuring them. However, according to the knowledge of the present inventors, this type of application is not sufficient in terms of accuracy when it comes to the purpose of extracting the protrusions of cancer cell spheroids. In other words, morphological processing involving contraction and expansion of image objects is generally applied to separate tree-like structures. Here, the protrusions in cancer cell spheroids are particularly thick at the base, and the boundary with the cell mass is unclear. For this reason, in general morphological processing, this part is recognized as part of the cell mass, and the number and length of the protrusions may be estimated lower than they actually are.

Thus, with regard to the goal of isolating protrusions equivalent to invadopodia from three-dimensional images of cancer cell spheroids and quantitatively measuring them, the image processing technology that makes this possible has not yet been put to practical use.

This invention has been made in consideration of the above problems, and aims to provide an image processing technique that can accurately separate areas corresponding to cell masses and areas corresponding to protrusions from a three-dimensional image of a cancer cell spheroid.

One aspect of the present invention is an image processing method for identifying a region corresponding to a cell mass and a region corresponding to the protrusion from an original image including a three-dimensional image of a cancer cell spheroid on which a protrusion is formed, and in order to achieve the above object, the method includes the steps of: extracting an object region occupied by an object corresponding to the cancer cell spheroid from the original image, and creating a binary image in which the object region is binarized so that it has a higher brightness than other regions; distance-transforming the binary image to create a distance image in which each pixel in the object region is assigned a brightness value such that the pixel that is the greater the distance to the closest edge of the object region has a higher brightness; performing maximum value filtering on the distance image to create a filtered image; and segmenting the object region into regions in the filtered image that have a brightness higher than a predetermined threshold value, which corresponds to the cell mass.

The invention configured in this way makes it possible to separate cell clusters and protrusions with high accuracy, and in particular, prevents the thick base of a protrusion from being mistakenly recognized as part of a cell cluster, a problem that occurs in conventional image processing applications. The reasons for this will be described in detail later, but are outlined below.

The erosion (shrinkage) process used in general morphology processing shrinks the outer edges of objects in an image, and has little effect on the interior of thick objects such as invadopodia in cancer cell spheroids. Therefore, unless the outer edges are significantly shrunk, the base of the protrusion is treated as one part of the cell mass area. As an alternative to this, the present invention performs distance conversion processing. Therefore, even if it is the internal area of the thick base of the protrusion, the brightness value assigned is relatively small compared to the area near the center of the cell mass, which is a structure much larger than the protrusion.

Furthermore, in the present invention, maximum value filtering is performed as a process corresponding to the dilation (expansion) process in conventional morphology processing. As a result, while the brightness values of the pixels on the outer edge of the cell cluster increase, the brightness values of the pixels inside the protrusions that are far from the center of the cell cluster do not increase as much. As a result, the brightness values of each pixel are polarized between the inside of the cell cluster and the inside of the protrusions.

Therefore, by introducing an appropriate threshold value for the brightness value of each pixel, the object region can be divided into a region corresponding to a cell cluster and a region corresponding to a protrusion. Even inside a thick protrusion, if it is far from the center of the cell cluster, the brightness value is low and it is recognized as part of the protrusion, so misrecognition as occurs with conventional technology can be avoided.

In this way, the image processing according to the present invention is configured to fully utilize the morphological characteristics unique to cancer cell spheroids that have protrusions, and this configuration makes it possible to accurately separate the cell mass region from the protrusion region.

Another aspect of the present invention is a program for causing a computer to execute each step of the image processing method described above. Another aspect of the present invention is a computer-readable recording medium that non-temporarily stores the program. In an invention configured in this way, it is possible to use, for example, an existing computer device as the executing entity of the present invention.

As described above, according to the present invention, it is possible to accurately separate the cell mass region and the protrusion region from a three-dimensional image of a cancer cell spheroid, for example, an OCT image.

FIG. 1 is a diagram showing an example of the configuration of a computer device capable of executing an image processing method according to the present invention. FIG. 2 is a diagram illustrating a cancer cell spheroid that is the subject of analysis in this embodiment. 1 is a flowchart showing an embodiment of an image processing method according to the present invention. FIG. 11 is a diagram showing changes in an image resulting from the processing up to this point. FIG. 2 is a diagram for explaining the principle of area division in the present embodiment. FIG. 13 is a diagram showing an example of a result of region division. FIG. 13 illustrates an example of a GUI screen suitable for adjusting a threshold value. FIG. 11 is a diagram showing the effect of region division according to the present embodiment.

Figure 1 shows an example of the configuration of a computer device capable of executing the image processing method according to the present invention. The image processing according to the present invention can be realized by, for example, implementing a control program for executing each processing step of the image processing described below in a computer device 1 having a hardware configuration such as that shown in Figure 1, and having the computer device 1 execute the control program.

The computer device 1 has a typical hardware configuration, for example, as a personal computer, and includes a CPU (Central Processing Unit) 10, memory 14, storage 15, input device 16, display unit 17, interface 18, and disk drive 19.

The CPU 10 executes a control program prepared in advance to realize, in software, an image processing unit 11 as a functional block for executing image processing described below. Dedicated hardware for realizing the image processing unit 11 may also be provided. The memory 14 temporarily stores various data generated during the calculation process of the CPU 10. The storage 15 stores, on a long-term basis, the control program to be executed by the CPU 10, as well as image data of the original image to be processed and image data after processing.

The input device 16 is for receiving instructions input from an operator and includes, for example, a mouse and a keyboard. The display unit 17 is, for example, a liquid crystal display that has a function for displaying images, and displays various information such as the original image, the processed image, and messages to the operator. A touch panel in which the input device and the display unit are integrated may also be provided.

The interface 18 exchanges various types of data with external devices via electrical communication lines. The disk drive 19 accepts an external recording disk 2 on which various types of data such as image data and control programs are recorded. The image data, control programs, etc. stored on the recording disk 2 are read by the disk drive 19 and stored in the storage 16. The disk drive 19 may also have a function of writing data generated within the computer device 1 to the recording disk 2.

The control program for causing the computer device 1 to execute the image processing of this embodiment may be read by the disk drive 19 accessing the recording disk 2 on which it is recorded, or may be provided from an external device via the interface 18. The same applies to the original image data.

The specific contents of the image processing method executed by the computer device 1 will be described below. Note that this computer device 1 does not have an imaging function, and the images to be processed are captured by another device and acquired from the outside, for example, via a telecommunications line. However, for example, a control program may be implemented in the control unit of a device that has an imaging function, and both imaging and image processing may be performed by this imaging device.

The image processing in this embodiment is a process for analyzing the three-dimensional image obtained by imaging the cancer cell spheroid, and separating the area of the cell mass from the area of the protrusions called invadopodia that extend around it. The ultimate goal is to obtain quantitative information such as the number, length, thickness, and degree of branching of the invadopodia.

Figure 2 is a diagram illustrating a cancer cell spheroid that is the subject of analysis in this embodiment. Figure 2(a) is an example image of an artificially cultured cancer cell spheroid. More specifically, Figure 2(a) is an example image of a cancer cell spheroid cultured until invadopodia are formed in a culture environment simulating an in vivo environment, which is imaged using an optical coherence tomography (OCT) imaging device, and a three-dimensional image of the cancer cell spheroid is created from multiple tomographic images. Figure 2(b) is a schematic diagram showing the structure of a cancer cell spheroid in a simplified manner.

As shown in these figures, in a cancer cell spheroid S, a cell mass C is formed in the center, where many cells are gathered in an approximately spherical shape, and protrusions P extend radially from the surface. Furthermore, the protrusions may branch out in many different ways. Cancer cell spheroids artificially created in this way may be used, for example, as model cells in drug discovery screening for the development of cancer treatment drugs. In other words, by administering a chemical substance that is a drug candidate to the cancer cell spheroid and examining the changes that occur, the efficacy of the chemical substance can be evaluated.

Changes in cancer cell spheroids can be evaluated, for example, by measuring changes in the size and shape of cell clusters and invadopodia. To enable such evaluation, there is a need to separately identify the area occupied by the cell cluster C and the area occupied by the protrusion P from a three-dimensional image of the cancer cell spheroid. The image processing of this embodiment is aimed at this.

Figure 3 is a flow chart showing one embodiment of an image processing method according to the present invention. This process is realized by the CPU 10 of the computer device 1 executing a control program that has been read out in advance from the recording disk 2 and recorded in the storage 16. First, an image including a three-dimensional image of a cancer cell spheroid with invadopodia formed is acquired as the original image to be processed (step S101). The image may be newly captured, or image data that has been captured in advance and stored may be recalled.

The original image is, for example, a three-dimensional image created from a tomographic image obtained by OCT imaging, as shown in Figure 2(a). However, there are no particular limitations on the imaging method, as long as it can similarly obtain a three-dimensional image of the cancer cell spheroid.

From the original image thus obtained, image objects corresponding to the cancer cell spheroids are extracted (step S102), and based on the result, the original image is binarized by assigning different brightness values to the areas occupied by the image objects and other areas (step S103). Simply put, for example, the area of the image object can be assigned maximum brightness, and the other areas can be assigned minimum brightness (for example, zero).

For example, a semantic segmentation method can be suitably applied to extract image objects. However, this is not limiting, and various methods for dividing an image into regions based on the brightness value of each pixel can be applied. In addition, when a predetermined threshold value is set for the brightness value of each pixel and a process is performed to extract regions with higher or lower brightness than the threshold, the result of the extraction itself is a binarized image.

In the binarized image, the cell cluster regions and protrusion regions are not yet separated. In addition, the sample may contain objects other than cancer cell spheroids. Therefore, further image processing is performed to separate these.

Specifically, a distance transform process is first performed on the binarized image (step S104). Distance transform process is a common image process that can be performed by general-purpose image processing software, and represents each pixel in an image in an image object by the distance from that pixel to the nearest black pixel (i.e., a pixel with a brightness value of zero). For example, the distance transform of an image can be performed using the Distance Transform function provided in the OpenCV (Open Source Computer Vision) library.

Due to the nature of the distance transform process, in the transformed image, pixels closer to the center of large image objects are given higher values, while pixels within small image objects and pixels near the edges of image objects are given lower values. Outside the image objects, the value is zero.

A maximum value filter process is performed on the luminance values of the image after the distance transformation (step S105). As described later, it is preferable that the window size in the filter process is determined according to the size of the protrusion in the original image, but the window size at this point can be, for example, a preset initial value. By performing the maximum value filter process, the luminance value of each pixel is replaced with the luminance value of the pixel with the highest luminance among the pixels surrounding it.

The image after maximum value filtering is then divided into regions based on a predetermined threshold (step S106), dividing the image objects into regions corresponding to cell clusters and regions corresponding to protrusions. The processing up to this point makes it possible to divide the objects in the image into regions corresponding to cell clusters and regions corresponding to protrusions. The processing from step S107 onwards is an adjustment process that optimizes the processing parameters to obtain better segmentation results. This processing will be explained later.

Figure 4 shows the changes in the image that accompany the processing up to this point. Note that while actual image processing is carried out in three-dimensional image space based on three-dimensional image data, or so-called voxel data, for the sake of clarity in the explanation on paper, for the sake of convenience, we will use one of the tomographic images that is the source of the three-dimensional image data for the explanation. The same applies to Figures 5 and 6 below. Note that below, image objects will simply be referred to as "objects".

Image Ia shown in FIG. 4(a) is an example of an original image, specifically a tomographic image obtained by OCT imaging of a cancer cell spheroid. In FIG. 4(a), there is a large object in the center of the image that corresponds to the cancer cell spheroid. It is estimated that the large central area corresponds to the cell mass, and the elongated peripheral area corresponds to the protrusion.

There are a number of small isolated objects around a larger object. These isolated objects may be either actually isolated within the specimen or may be a continuous structure with the central object, but appear to be isolated in a cross-sectional image. In other words, it is important to note that there are cases where a protrusion that is a continuous structure with the cell mass in the three-dimensional image appears to be isolated in the cross-sectional image. This is also true for the following images.

Image Ib shown in FIG. 4(b) is an example of an image after image Ia has been binarized (step S103). In binarized image Ib, areas extracted as image objects are shown in white (i.e., high brightness), and other areas are shown in black (i.e., low brightness).

Image Ic shown in FIG. 4(c) is an example of an image (i.e., a distance image) after image Ib has been subjected to distance transformation (step S104), and the larger the value given by the process, the higher the brightness. The center of a large object is particularly bright, and the closer to the periphery, the darker it is. For small objects, the distance to the periphery is not large, so even the center does not become very bright.

Compared to the binarized image Ib, the object appears to have shrunk, with parts with particularly fine structures appearing thinner, and in that sense it is similar to erosion processing in conventional morphological processing. However, unlike erosion processing, which shrinks the object itself, the size of the object does not change; the brightness of its peripheral areas has simply decreased.

Image Id shown in FIG. 4(d) is an example of an image after maximum value filtering (step S105) is performed on image Ic after distance transformation. By maximum value filtering, high-luminance areas are enlarged compared to image Ic. Low-luminance areas on the periphery are also enlarged by an amount equivalent to the window size by maximum value filtering. In this sense, this process can be said to be an alternative to dilation processing in conventional morphology processing.

Figure 5 is a diagram for explaining the principle of region division in this embodiment. As a simplified model of a cancer cell spheroid, as shown in Figure 5(a), a structure in which a relatively thin protrusion P1 and a thicker protrusion P2 extend in opposite directions from a roughly circular cell mass C is used to explain the principle of region division in this embodiment.

Figure 5(b) shows the luminance distribution in the image after distance transformation processing, specifically, the luminance distribution on the line drawn through the center of the cell cluster C along the extension direction of the protrusions P1 and P2, shown by the dotted line in Figure 5(a). As can be seen, the luminance is highest at the center of the cell cluster C, and decreases closer to the periphery. Without the protrusions, the luminance would decrease to zero at the periphery of the cell cluster C, as shown by the dotted line in Figure 5(b), but the presence of the protrusions P1 and P2 prevents the decrease in luminance at the periphery of the cell cluster C. Even in the protrusions P1 and P2, the luminance value is high in the wider parts, and decreases as the luminance value approaches the narrower tip.

Figure 5(c) shows the luminance distribution on the same line as above in the image after maximum value filtering. By performing maximum value filtering, the profile after distance conversion shown in Figure 5(b) (shown by a dotted line in Figure 5(c)) expands by a width corresponding to the window size W, as shown by a solid line. As can be seen from this figure, in the region corresponding to the cell cluster C, the luminance is high inside, so there is a relatively large increase in luminance after filtering. The increase is particularly noticeable in the periphery. In comparison, there is only a slight increase in luminance in the regions corresponding to the protrusions P1 and P2, which have a thinner structure than the cell cluster C and therefore a lower luminance value. This difference is an important factor for accurately separating the cell cluster and the protrusions.

As shown in FIG. 5(d), an appropriate threshold value Th is introduced, and regions with brightness values higher than the threshold value Th are regarded as cell clusters C, and regions with brightness values lower than the threshold value Th are regarded as regions corresponding to protrusions P1 and P2, and region division is performed. As is clear from a comparison of FIG. 5(a) and FIG. 5(d), these regions are separated from each other at a position close to the boundary between the actual cell cluster C and protrusions P1 and P2. Naturally, the boundary between the divided regions varies greatly depending on the threshold value, so it is a prerequisite that the threshold value is set appropriately.

As can be seen from Figure 5(c), when threshold-based segmentation is performed on a profile (dotted line) that does not undergo maximum value filtering, the area corresponding to cell cluster C is estimated to be smaller than it actually is. As shown by the solid line in Figure 5(c), this problem can be avoided by performing maximum value filtering to selectively increase the brightness value inside cell cluster C, especially in its periphery.

In the image processing of this embodiment, a distance transformation process is performed to separate the fine structures corresponding to the protrusions from the areas corresponding to the cell masses. With this method, the weight of the periphery of the object is relatively small compared to the center, but unlike the erosion process in the conventional method, no part of the object is erased. In other words, the information for identifying the contour of the object is maintained at this point.

Then, maximum value filtering is performed as a process equivalent to the dilation process in conventional methods. The effect of this on increasing brightness values is large in cell clusters, which are relatively large structures and have high brightness values due to the distance conversion process, but is small in protrusions, which are smaller structures. Therefore, while there is a significant increase in brightness values in the periphery of the cell cluster, the increase in brightness values is limited in the periphery of the protrusions.

Therefore, in the subsequent region segmentation process, regions with a higher brightness than an appropriately set threshold value can be regarded as regions corresponding to cell clusters and extracted. In other words, since cell clusters are given relatively high brightness values up to the pixels on their periphery, it is possible to distinguish them from protrusions, for which the effect of increasing brightness due to maximum value filter processing is small, by simply comparing them with the threshold value. Note that maximum value filter processing causes the periphery of the protrusion to expand somewhat. For this reason, if regions with a lower brightness than the threshold value are directly regarded as protrusion regions, a larger region than the actual size will be extracted. To prevent this, it is desirable to treat the region obtained by deleting the regions regarded as corresponding to cell clusters from the image before maximum value filter processing, for example the original image or the binarized image, as the protrusion region.

In this way, the combination of distance transformation processing and maximum value filter processing in this embodiment clarifies the brightness contrast between the cell clusters and the protrusions, and enables the subsequent threshold-based region segmentation to properly separate the cell clusters and the protrusions.

As a comparative example, FIG. 5(e) shows a schematic diagram of the results of erosion processing in region segmentation using a combination of conventional erosion processing and dilation processing. By uniformly shrinking the periphery of the object by an appropriate amount, it is relatively easy to erase the portion corresponding to the relatively thin protrusion P1 from the object. On the other hand, the wide protrusion P2 remains as part of the object, which causes the problem with the prior art mentioned above, namely, the problem that the base of a thick protrusion is regarded as part of the cell mass.

In principle, it is possible to eliminate such protrusions by increasing the amount of shrinkage. However, if the amount of shrinkage is too large, a large amount of information about the original shape is lost, and the reproducibility of the object shape after dilation processing is significantly reduced.

In contrast, the image processing of this embodiment sequentially performs binarization, distance conversion, and maximum value filtering on the original image, and finally performs region segmentation based on a threshold value, thereby making it possible to mitigate the above problems.

Figure 6 shows an example of the results of region segmentation. Figure 6(a) is an example of an image in which regions of the binarized image Ib that have been determined to correspond to cell clusters C are shown in white. Figure 6(b) is an example of an image in which regions of the binarized image Ib that have been determined to correspond to protrusions P are shown in white. For example, by applying these as mask images to the original image Ia, it is possible to extract the regions corresponding to cell clusters C and the regions corresponding to protrusions P from the original image Ia.

Returning to FIG. 3, the image processing of this embodiment will be described further. Through the processing up to step S106, the object has been divided into an area corresponding to the cell mass C and an area corresponding to the protrusion P. However, unless the processing parameters, specifically the window size W in the maximum value filter processing and the threshold value Th in the area division processing, are set appropriately, the division results will not be good. Therefore, the image processing of this embodiment is configured to allow these processing parameters to be adjusted as necessary.

Specifically, the result image obtained by the processing up to step S106 is displayed on the display unit 17 and presented to the user (step S107), and operation input from the user regarding parameter changes is accepted. If the user wishes to change the threshold setting for region division (YES in step S108), a new threshold setting input is accepted (step S211), region division processing is executed based on the new threshold (step S106), and the result is displayed (step S107). Also, if the user wishes to change the window size setting in the maximum value filter processing (YES in step S109), the new setting input is accepted (step S212), and maximum value filter processing is re-executed based on the new window size (step S105). Furthermore, region division processing is re-executed (step S106), and the result is displayed (step S107).

If the user does not wish to change the parameters (NO in both steps S108 and S109), the result image at that point is stored in storage 15 (step S110), and the process ends. In this way, the parameters are optimized, making it possible for the user to obtain the segmentation results desired.

Of these parameters, the window size W can be estimated to some extent from the size of the protrusions in order to separate the protrusions from the cell mass. Specifically, it is known that a value of about 1/2 to 1 time the thickness of the protrusions is suitable. Note that the actual maximum value filter process is a spherical filter process in three-dimensional image space. Therefore, the window size W referred to here is a concept equivalent to the diameter of a spherical window in spherical filter process. The radius of the spherical window may be taken as the window size.

In cancer cell spheroids, invadopodia can have various thicknesses. In this case, the "thickness of the protrusion" related to the determination of the window size can be, for example, the thickness (width) of the thickest part of the various protrusions. In this way, the window size in maximum value filtering can be narrowed down to a certain extent at the point when the external shape of the cancer cell spheroid can be grasped. In other words, it is thought that there will be no need to frequently change the window size when optimizing parameters.

On the other hand, the threshold value Th is an important parameter for determining the boundary between the cell cluster and the protrusion, and requires fine adjustment. For this reason, in the process shown in FIG. 3, the processing loop for adjusting the threshold value is placed inside the processing loop for adjusting the window size. In addition, as shown below, a user interface screen (Graphical User Interface; GUI) is provided to assist the user in adjusting the threshold value.

Figure 7 shows an example of a GUI screen suitable for adjusting the threshold. All of Figures 7(a) to 7(d) are examples of images resulting from region segmentation in step S106, in which areas of the object in the image that are deemed to correspond to cell clusters have been colored to be brighter (whiter on paper) than areas that are deemed to be protrusions.

In addition, in each of these images, a histogram H of the luminance values in the image and a gauge G showing the setting value of the threshold value Th for the luminance values are added to the image Ip that has been colored as described above. A comparison of these images shows that the boundary between the cell mass and the protrusion changes significantly depending on the setting of the threshold value Th.

The user can change the setting of the threshold Th by operating the gauge G via the input device 16. When the threshold is changed, the region division is executed again, and the result image is displayed each time. The user can adjust the threshold Th while checking the result image, making it possible to find the optimal threshold Th in a short time.

Figure 8 shows the effect of region division according to this embodiment. Figure 8(a) is an image in which the image resulting from the image processing of this embodiment is applied to the original three-dimensional image to distinguish between the regions occupied by cell clusters and the regions occupied by protrusions in the three-dimensional image. Here, an image is shown in which the three-dimensional image is projected into two dimensions. Specifically, the darker colored area in the center is the region considered to be the cell region, and the lighter area is the region considered to be the protrusion. It can be said that the region has been accurately divided into the central cell cluster region and the protrusion region extending radially outward from the surface.

Figure 8 (b) is a comparative example, showing the results of dividing the cell cluster region and the protrusion region by combining erosion processing and dilation processing, which are conventional methods. Even if the amount of contraction in erosion and the amount of expansion in dilation are sufficiently adjusted, there are still areas that are incorrectly recognized as cell clusters, especially in some of the thick protrusions, as shown by the arrows in the figure.

As described above, in the image processing of this embodiment, the following processes are sequentially performed on the three-dimensional image of the cancer cell spheroid: binarization based on the object extraction results, distance conversion, maximum value filtering, and region segmentation based on a threshold value. This makes it possible to separate the cell mass region and the protrusion (invadopodia) region with greater accuracy than conventional methods.

If cell clusters and invadopodia can be separated with such precision, the results can be used for various quantitative evaluations, such as the number, thickness, and length of invadopodia, and the degree of branching. For example, if the image of the cell cluster is erased from the three-dimensional image, and an image is created that leaves only the image of the protrusions, measurements for such evaluations can be easily performed based on that image. In particular, this embodiment reduces the type of misrecognition in which thick invadopodia are deemed to be cell clusters, making it possible to avoid the problem of the length of the invadopodia being estimated to be shorter than it actually is.

The present invention is not limited to the above-described embodiment, and various modifications other than those described above can be made without departing from the spirit of the present invention. For example, the computer device 1 of the above-described embodiment is pre-implemented with a control program for executing the image processing according to the present invention. In other words, the present invention may be implemented as a software program for causing a computer device to execute each processing step of the above-described processing.

Such a program can be distributed, for example, by downloading it via a telecommunications line such as the Internet, or by distributing a computer-readable recording medium on which the program is recorded. In addition, for example, by loading the program into an existing microscope imaging device via an interface, the present invention can be implemented by the device.

For example, in the above embodiment, the window size in the maximum value filter process and the threshold value in the region segmentation process can be changed as processing parameters. As for the window size, as described above, it is possible to estimate an appropriate value to some extent from the size (thickness) of the invadopodia in the cancer cell spheroid to be analyzed. Therefore, it is also possible to use a fixed value that is predetermined according to the size of the target object, or to allow the user to set only the initial value.

In addition to the image processing of the above embodiment, other image processing, such as noise removal processing and smoothing processing, may be performed in appropriate combination to eliminate images of minute objects other than the analysis target and image noise.

In addition, in the image processing of the above embodiment, a binarized image is used in which the inside of the image object in the three-dimensional image has high brightness and the background has low brightness. In the distance transformation process, distance transformation is performed so that the center of the object has high brightness, and maximum value filtering is performed. In the region segmentation process, regions with brightness higher than a threshold value are regarded as cell mass regions. However, even if the brightness relationship is inverted in these processes, it is technically equivalent.

The image processing of the above embodiment is directed to analyzing a three-dimensional image of a cancer cell spheroid imaged by OCT. However, the imaging method is not limited to this, and similar processing is possible for images captured by various imaging methods that can three-dimensionally represent the three-dimensional shape of the surface of the object to be analyzed. Furthermore, the image processing of this embodiment can be applied not only to cancer cell spheroids, but also to the analysis of various cell clusters that have protrusions extending outward from the surface of a similar cluster of cells.

As described above by way of example of specific embodiments, in the image processing method according to the present invention, for example, in distance transformation, each pixel in an object region may be assigned a luminance value higher than the luminance value assigned to pixels outside the object region. Such processing can be realized using standard functions of general-purpose image processing software, and is fully practical as the distance transformation processing of the present invention.

As a method for extracting object regions from an original image, for example, a semantic segmentation method can be applied. With this configuration, it is possible to extract object regions with high accuracy even when the object to be extracted is an indefinite shape such as a cancer cell spheroid.

For example, a step of displaying the image after region division on a display device may be further provided. In this way, the results of the processing are presented to the user as an image, allowing the user to confirm the results of the processing. For example, this can be used to adjust parameters such as:

For example, the threshold value can be made changeable by the user. Setting the threshold value changes the position of the boundary defined between the cell clumps and the protrusions, which greatly affects the success or failure of separation of the two. By making the threshold a changeable parameter and adjusting it according to the state of the cancer cell spheroids being analyzed, it becomes possible to separate the cell clumps and the protrusions more accurately.

Similarly, the window size in maximum value filtering may be changeable by user operation. The window size in maximum value filtering is a parameter related to the contrast in brightness values used to separate cell clusters and protrusions, and particularly affects the reproducibility of their shapes after separation. By making this adjustable as necessary, it becomes possible to separate cell clusters and protrusions more accurately.

For example, the maximum value filter process and region division may be configured to be repeatedly executed in this order. As described above, the window size, which is a processing parameter related to the maximum value filter process, and the threshold value, which is a processing parameter related to the region division, are both related to the accuracy of separation between cell clusters and protrusions. When these are changed, it is preferable to re-execute the maximum value filter process or region division, and by repeatedly executing these two processes, it becomes possible to successively bring the two parameters closer to the optimal values.

The image processing method according to the present invention may also be configured to, for example, erase areas corresponding to cell clusters from the original image or the binarized image to create an image showing areas corresponding to protrusions. With such a configuration, a three-dimensional image is created in which only the areas corresponding to protrusions are shown. Based on this image, it becomes possible to easily perform quantitative evaluation of the protrusions.

For example, the original image may be an image acquired by imaging using an optical coherence tomography (OCT) imaging method. Optical coherence tomography (OCT) imaging allows for non-destructive and non-invasive imaging of cells and spheroids. For this reason, for example, by periodically imaging a cultured cancer cell spheroid, it is possible to observe changes in the cancer cell spheroid over time. Furthermore, images captured using OCT are suitable for use in the image processing according to the present invention.

This invention can be applied to the quantitative evaluation of artificially cultured cancer cell spheroids, and is suitable for use in the fields of medicine and drug discovery, for example.

1 Computer device 2 Recording disk (recording medium)
10 CPU
15 Storage 16 Input device 17 Display unit C Cell mass P, P1, P2 Invadopodia (protrusions)

Claims (11)

1. An image processing method for identifying an area corresponding to a cell mass and an area corresponding to a protrusion from an original image including a three-dimensional image of a cancer cell spheroid having a protrusion, the method comprising:
A step of extracting an object region occupied by an object corresponding to the cancer cell spheroid from the original image, and creating a binarized image by binarizing the object region so that it has a higher brightness than other regions;
a step of performing distance transformation on the binarized image to generate a distance image in which a luminance value is assigned to each pixel in the object region such that the luminance of a pixel increases as the pixel is closer to the nearest edge of the object region;
performing a maximum filtering process on the range image to generate a filtered image;
and segmenting an area of the object region in the filtered image that has a brightness higher than a predetermined threshold, the area corresponding to the cell mass, from the object region. The image processing method of claim 1, wherein the distance transformation assigns a luminance value to each pixel in the object region that is higher than the luminance value assigned to pixels outside the object region. The image processing method according to claim 1 or 2, in which the object region is extracted from the original image by semantic segmentation. The image processing method according to any one of claims 1 to 3, further comprising a step of displaying the image after the region division on a display device. The image processing method according to any one of claims 1 to 4, wherein the threshold value can be changed and set by a user operation. The image processing method according to any one of claims 1 to 5, wherein the window size in the maximum value filter process can be changed and set by a user operation. The image processing method according to claim 5 or 6, wherein the maximum value filtering process and the region division can be repeatedly performed in this order. The image processing method according to any one of claims 1 to 7, further comprising the step of erasing the area corresponding to the cell mass from the original image or the binarized image to create an image showing the area corresponding to the protrusion. The image processing method according to any one of claims 1 to 8, wherein the original image is an image acquired by imaging using an optical coherence tomography method. A program for causing a computer to execute each step of the image processing method according to any one of claims 1 to 9. A computer-readable recording medium on which the program according to claim 10 is non-temporarily recorded.

Images