KR-20260063352-A - Method for recommending regions of interest in biological tissue image and computing system performing the same

KR20260063352AKR 20260063352 AKR20260063352 AKR 20260063352AKR-20260063352-A

Abstract

A method for recommending regions of interest within a biological tissue image and a computing system for performing the same are disclosed. According to one aspect of the present invention, a computing system comprises the steps of: acquiring a biological tissue image; identifying cell instances within a search region that is at least part of the biological tissue image; calculating a predetermined characteristic value for each of the cell instances within the search region; selecting K partial regions from among the partial regions of the search region based on an assessment score to determine K Regions of Interest (ROI)—wherein K is a predetermined natural number, the partial region of the search region is a part of the biological tissue image having a predetermined shape and area, and the assessment score for each of the partial regions is a predetermined first numerical value calculated from the characteristic value of the cell instances within the respective partial region—and the computing system outputting the determined K regions of interest.

Inventors

양현석
황윤섭
장혜윤
곽태영
김선우

Assignees

주식회사 딥바이오

Dates

Publication Date: 20260507
Application Date: 20241030

Claims (20)

A computing system acquires a biological tissue image; The computing system identifies cell instances within a search area that is at least part of the biological tissue image; The above computing system calculates a predetermined characteristic value for each of the cell instances within the search area; The computing system determines K regions of interest (ROIs) by selecting K regions from among the regions of the search region based on an assessment score—wherein K is a predetermined natural number, the regions of the search region are parts of the biological tissue image having a predetermined shape and area, and the assessment score of each of the regions is a predetermined first numerical value calculated from the characteristic values of cell instances within the corresponding region; and A method comprising the step of the above computing system outputting the determined K regions of interest.
In paragraph 1, The step of identifying cell instances in the above-mentioned search area is, A step of performing segmentation on the above-mentioned search area to generate a cell component segmentation mask indicating whether each pixel of the above-mentioned search area corresponds to the cell nucleus, cytoplasm, or cell membrane; A step of generating a cell instance segmentation result mask for identifying each cell instance within the search area based on the cell component segmentation mask above—wherein, pixels constituting each cell instance of the cell instance segmentation result mask are assigned a label number of the cell instance to which the pixel belongs, pixels corresponding to the boundary of the cell instance are assigned a boundary unique number which is a unique number representing the boundary, and pixels corresponding to the background are assigned a background unique number which is a unique number representing the background; Based on the cell instance segmentation result mask, the step of generating a binary mask for each of the cell instances within the search area; and A method comprising the step of identifying each of the cell instances within the search area by applying a binary mask of each of the cell instances within the search area to the search area.
In paragraph 2, The step of generating a cell instance segmentation result mask to identify each cell instance within the above-mentioned search area is: A step of generating a cell background mask, which is a binary mask that distinguishes the internal part of the cell corresponding to the cell nucleus and cytoplasm from the other part, based on the cell component segmentation mask above; A step of generating a cell foreground mask, which is a binary mask distinguishing between a foreground portion and an extra portion—wherein the foreground portion of the cell foreground mask is a subset of the cell interior portion of the cell background mask; The computing system generates cell instance markers for a marker-controlled watershed based on the cell foreground mask and the cell background mask; and A method comprising the step of the computing system performing a marker-controlled watershed with the cell background mask and the cell instance marker to generate a cell instance segmentation result mask.
In paragraph 2, Based on the cell instance segmentation result mask, the step of generating a binary mask for each of the cell instances within the search area is: A method comprising the step of the computing system binarizing the pixels of the cell instance segmentation result mask into unique label numbers of each of the cell instances to generate a binary mask of each of the cell instances.
In paragraph 2, The above characteristic value is the staining intensity in a predetermined biomarker expression region, and The expression region of the above biomarker is at least one of the cell nucleus, cytoplasm, and cell membrane, and The step of calculating a predetermined characteristic value for each cell instance within the above-mentioned search area is: A method comprising the step of measuring the staining intensity in the biomarker expression region of each of the cell instances within the search region based on the cell component segmentation mask.
In paragraph 1, The step of selecting K sub-regions from the above-mentioned search area based on evaluation scores and determining them as K regions of interest is: A step of measuring the evaluation score of each of the first windows while searching the search area using a sliding window method having a first sliding window parameter—wherein the first sliding window parameter includes the shape and area of the window, and the shape and area of the first sliding window parameter are pre-set to be identical to a partial area of the search area; A step of determining at least some of the first windows as precision search areas by sorting them in order of evaluation score; and Until the above K regions of interest are determined, for each of the above precise search regions sequentially sorted according to the order of evaluation scores, the method includes the step of precisely searching the surrounding region of the said precise search region. The step of precisely searching the surrounding area of the above-mentioned precise search area is: A step of determining a surrounding area of a predetermined size having the same center point as the center point of the above-mentioned precision search area; A step of measuring evaluation scores of the second windows while searching the surrounding area using a sliding window method having second sliding window parameters, and determining a candidate region of interest based on the measured evaluation scores—wherein the second sliding window parameters include the shape and area of the window, and the shape and area of the second sliding window parameters are pre-set to be identical to a partial area of the search area; and A method comprising the step of determining the candidate region of interest as one of the K regions of interest when the candidate region of interest does not overlap at all with the previously determined regions of interest.
In paragraph 6, The first sliding window parameter and the second sliding window parameter further include a movement interval of the window, and A method characterized in that the movement interval of the second sliding window parameter is smaller than the movement interval of the first sliding window parameter.
In paragraph 6, The step of measuring the evaluation score of each of the first windows above is, The method further includes the step of measuring a threshold comparison value of each of the first windows—wherein the threshold comparison value of each of the first windows is a predetermined second numerical value calculated from the characteristic values of cell instances within the first window as a numerical value to be compared with a predetermined first threshold value that serves as a filtering criterion—and The step of determining at least some of the first windows as a precision search area by sorting them in order of evaluation score is: A method comprising the step of determining whether each of the first windows satisfies a predetermined first threshold condition expressed by the first threshold value, filtering at least some of the first windows, and sorting the filtered first windows based on an evaluation score to determine the precision search area.
In paragraph 8, The step of measuring the evaluation score of each of the above-mentioned second windows and determining candidate regions of interest based on the measured evaluation score is A step of measuring a threshold comparison value for each of the second windows; and A method comprising the step of determining the window with the highest evaluation score among the second windows satisfying the first threshold condition or a predetermined second threshold condition expressed by a predetermined second threshold value as the candidate region of interest.
In paragraph 6, the above method is, A step of generating a heatmap based on the location and evaluation score of each of the filtered first windows satisfying the first threshold condition among the first windows; and The method further includes the step of overlaying the heatmap on the biological tissue image and outputting it, The heatmap above is, It is a two-dimensional array having the same aspect ratio as the above biological tissue image, and A method in which each element of the above heatmap is assigned a value obtained by dividing the sum of the evaluation scores of filtered first windows belonging to the area on the biological tissue image covered by the element by the number of filtered first windows belonging to the area on the biological tissue image covered by the element.
In paragraph 1, The step of selecting K sub-regions from the above-mentioned search area based on evaluation scores and determining them as K regions of interest is: A step of measuring the evaluation score of each of the windows while searching the search area using a sliding window method in which the shape and area of the window are set to be identical to a partial area of the search area; and A method comprising the step of, for each of the K grades of evaluation scores, selecting a representative window from among the windows corresponding to the grade of evaluation score to determine the region of interest.
In Paragraph 11, The step of measuring the evaluation score of each of the above windows is, The method further includes the step of measuring a threshold comparison value for each of the above-mentioned windows—wherein the threshold comparison value of each of the above-mentioned windows is a predetermined second numerical value calculated from the characteristic values of cell instances within the window as a numerical value to be compared with a predetermined threshold value that serves as a filtering criterion—and The step of selecting a representative window among the windows corresponding to the grade based on the above evaluation score and determining it as a region of interest is: A step of sorting candidate windows whose evaluation scores correspond to the grade based on the threshold comparison value while the threshold comparison value satisfies a predetermined threshold condition expressed by the threshold value; and A method comprising the step of selecting the leading candidate window among the aligned candidate windows that has no overlap with the previously determined regions of interest as the representative window of the grade.
In Clause 12, the above method is, A step of generating a heatmap based on the location and evaluation score of each of the filtered windows satisfying the threshold condition among the windows; and The method further includes the step of overlaying the heatmap on the biological tissue image and outputting it, The heatmap above is, It is a two-dimensional array having the same aspect ratio as the above biological tissue image, and A method in which each element of the above heatmap is assigned a value obtained by dividing the sum of the evaluation scores of candidate windows belonging to the area on the biological tissue image covered by the element by the number of candidate windows belonging to the area on the biological tissue image covered by the element.
In paragraph 1, The above characteristic value is, A method comprising at least some of the expression intensity of a biomarker, the size of a cell or cell nucleus, the roundness of the shape of a cell or cell nucleus, the eccentricity when the cell or cell nucleus is approximated as an ellipse, and the solidity of the cell or cell nucleus.
A computer-readable recording medium having a computer program for performing a method described in any one of paragraphs 1 through 14.
As a computing system, A processor; and memory for storing computer programs, comprising, The above computer program is, A computing system that, when executed by the above processor, causes the computing system to perform the method described in any one of claims 1 to 14.
Acquisition module for acquiring biological tissue images; An identification module that identifies cell instances within a search area that is at least part of the above biological tissue image; A calculation module that calculates a predetermined characteristic value for each of the cell instances within the above-mentioned search area; A decision module for selecting K partial regions from among the partial regions of the above-mentioned search region based on evaluation scores and determining them as K regions of interest—wherein K is a predetermined natural number, the partial region of the above-mentioned search region is a part of the above-mentioned biological tissue image having a predetermined shape and area, and the evaluation score of each of the said partial regions is a predetermined numerical value calculated by the characteristic values of cell instances within the corresponding partial region; and A computing system comprising an output module that outputs the above-determined K regions of interest.
In Paragraph 17, The above identification module is, A cell component segmentation module that performs segmentation on the above-mentioned search area to generate a cell component segmentation mask indicating whether each pixel of the above-mentioned search area corresponds to the cell nucleus, cytoplasm, or cell membrane; A cell instance segmentation module for generating a cell instance segmentation result mask for identifying each cell instance within the search area based on the cell component segmentation mask above—wherein, pixels constituting each cell instance of the cell instance segmentation result mask are assigned a label number of the cell instance to which the pixel belongs, pixels corresponding to the boundary of the cell instance are assigned a boundary unique number which is a unique number representing the boundary, and pixels corresponding to the background are assigned a background unique number which is a unique number representing the background; A binary mask generation module that generates a binary mask for each of the cell instances within the search area based on the cell instance segmentation result mask; and A computing system comprising a cell instance identification module that identifies each of the cell instances within the search area by applying a binary mask of each of the cell instances within the search area to the search area.
In Paragraph 18, The above cell instance splitting module is, Based on the cell component segmentation mask above, a cell background mask is generated, which is a binary mask that distinguishes the internal part of the cell corresponding to the cell nucleus and cytoplasm from the other part, and Generate a cell foreground mask, which is a binary mask distinguishing between the foreground and the rest—wherein the foreground of the cell foreground mask is a subset of the cell interior of the cell background mask—, Based on the cell foreground mask and the cell background mask, cell instance markers for a marker-controlled watershed are generated, and A computing system that performs a marker-controlled watershed with the cell background mask and cell instance marker to generate a cell instance segmentation result mask.
In Paragraph 18, The above binary mask generation module is, A computing system that generates a binary mask for each of the cell instances by binarizing the pixels of the cell instance segmentation result mask into the unique label number of each of the cell instances.

Description

Method for recommending regions of interest in biological tissue image and computing system performing the same The present invention relates to a method for recommending a region of interest within a biological tissue image and a computing system for performing the same. In medicine, pathological diagnosis involving the interpretation of actual specimens is crucial for the accurate diagnosis of severe diseases and the prediction of prognosis. Traditional pathological diagnostic methods involve pathologists visually observing and interpreting diagnostic slides produced from specimens using an optical microscope. The method of converting pathology slides into digital images using a computer-connected microscope camera and then viewing and interpreting them on a monitor can be considered the origin of digital pathology. Recently, with the emergence of digital slide scanners, the method of converting an entire pathology slide into a single digital image and viewing and interpreting it on a computer monitor has become widely adopted. This digital pathology-based approach, which interprets scanned images via a monitor, is replacing the traditional method of visually examining specimen slides through an optical microscope. Meanwhile, identifying the tissue distribution of specific biomarkers is a critical task that accounts for a significant portion of pathological diagnosis. In particular, immunohistochemistry (IHC) staining-based methods for confirming biomarker distribution to identify the characteristics of tumor cells are currently widely utilized across many types of cancer. IHC staining is a technique that allows the distribution of target biomarkers to be visually observed by using dyes that express specific colors, such as brown or red, after binding to the target biomarker via an antigen-antibody binding method. Using this technology, proteins related to cancer diagnosis, prognosis, and treatment are stained, and the staining intensity and proportion of stained cells within the cancer cells are semi-quantitatively evaluated and subsequently utilized to determine treatment methods. Ki-67 is a protein that acts as a proliferation marker for breast cancer and is one of the major proteins stained with IHC along with HER2, ER, and PR during breast cancer diagnosis. The Ki-67 index is used to evaluate Ki-67 IHC results and represents the proportion of breast cancer cells identified as Ki-67 positive. Since it is impossible for pathologists to evaluate the Ki-67 index for every cell appearing on a pathology slide, they perform the evaluation by limiting the scope to representative areas of the slide; the following are some of the methods used for such evaluation. - Hotspot method: Evaluates the area with the highest staining rate - Global Method: Evaluates representative areas from the entire slide A brief description of each drawing is provided to help to better understand the drawings cited in the detailed description of the invention. FIG. 1 is a diagram illustrating a schematic system configuration for implementing a region of interest recommendation method according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating a method for recommending a region of interest according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating the process of determining K regions of interest to be recommended by a computing system according to one embodiment of the present invention (step S50 of FIG. 2). Figure 4 is a diagram illustrating the initial part of the process of searching using a sliding window method. Figure 5 is a flowchart illustrating an example of a specific process of step S220 of Figure 3. FIG. 6 is a flowchart illustrating another embodiment of step S50 of FIG. 3. Figure 7 is a flowchart illustrating the specific process of step S330 of Figure 6. FIG. 8a is a flowchart illustrating the process of a computing system according to an embodiment of the present invention generating and outputting a heatmap (step S70 of FIG. 3), and FIG. 8b and FIG. 8c are drawings illustrating examples of a heatmap overlaid on a biological tissue image, respectively. FIG. 9 is a flowchart illustrating in more detail the process (step S30) of a computing system according to one embodiment of the present invention identifying each cell instance from a biological tissue image. FIG. 10(a) illustrates an example of an IHC stained tissue image, FIG. 10(b) illustrates an example of visualizing a cell component segmentation mask corresponding to the IHC stained tissue image shown in FIG. 8(a), and FIG. 10(c) is an enlarged view of a portion of FIG. 10(b). Figure 11 is a diagram illustrating an example of the process of generating a cell component splitting mask. FIG. 12 is a flowchart illustrating a method for generating a cell instance splitting result mask according to one embodiment of the present invention. Figure 13 is a diagram illustrating an example of the process of generatin