Prognostic survival period prediction system based on lung biopsy histopathological features

Abstract

The present application relates to the technical field of medical image analysis, in particular to a prognosis survival period prediction system based on lung biopsy histopathological features. A data acquisition module acquires lung biopsy histopathological section gray scale images and clinical pathological data of a patient; a feature extraction module divides a tumor nucleus region according to pixel gray scale values, analyzes the distribution characteristics, divides a suspicious analysis area, and focuses on a special pathological significance area; then the local tumor nucleus region density, area and gray scale in the suspicious analysis area are analyzed to obtain pixel point pathological feature indexes, which reflect the tumor cell composition and growth state; a boundary analysis module divides a core region in the suspicious analysis area, extracts an initial tumor boundary line according to pathological feature changes, removes false noise points in combination with the position relationship and pathological features, and acquires a real tumor boundary line; and a prognosis prediction module effectively extracts tumor morphological features, combines clinical pathological data to perform survival period prediction, and comprehensively and accurately evaluates the prognosis of the patient.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis technology, specifically to a prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue. Background Technology

[0002] In the diagnosis and treatment of lung tumors, the pathological characteristics of lung biopsy tissue play a decisive role in assessing prognostic survival. The macroscopic and microscopic morphology of the tumor directly reflects its metastatic potential and biological invasiveness. For example, tumors with prominent spiculated or infiltrative margins have a significantly higher risk of recurrence than tumors with smooth borders. Therefore, inputting the morphological features of the tumor from grayscale images of pathological sections into a predictive model is an important approach to achieving accurate prognostic assessment.

[0003] Because lung tumor cells often infiltrate the surrounding normal lung interstitium in a linear or nest-like manner at the periphery, current technologies typically rely on manual annotation or conventional image segmentation algorithms based on grayscale thresholds to obtain the morphological features of tumors. However, during the preparation and staining of biopsy tissue sections, the tissue edges are prone to physical tearing, compression deformation, or staining artifacts, forming non-specific jagged edges or protruding noise. Therefore, manual annotation and conventional image segmentation algorithms based on grayscale thresholds often struggle to accurately identify true infiltration features and artifact noise points, resulting in reduced accuracy of subsequent prediction results. Summary of the Invention

[0004] To address the challenges of physical tearing, compression deformation, or staining artifacts at tissue edges during biopsy section preparation and staining, resulting in nonspecific jagged edges or raised noise, and the difficulty of accurately identifying true infiltration features and artifact noise points using manual annotation and conventional gray-scale threshold-based image segmentation algorithms, this invention aims to provide a prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue. The specific technical solution adopted is as follows: The data acquisition module is used to acquire grayscale images of pathological sections of lung biopsy tissue from patients, as well as clinical pathological data. The feature extraction module is used to divide the tumor nucleus region in the grayscale image of the pathological section according to the grayscale value of the pixel; to analyze the distribution characteristics of the pixel in the tumor nucleus region in the grayscale image of the pathological section and divide the suspicious analysis area; and to determine the pathological feature index of each pixel based on the density distribution, area size and grayscale performance of the tumor nucleus region within the local area of ​​each pixel in the suspicious analysis area. The boundary analysis module is used to divide the core region in each suspicious analysis area based on the pathological feature indicators between pixels; in each core region, it analyzes the changing characteristics of the pathological feature indicators of pixels to extract the initial tumor boundary line; based on the initial tumor boundary line, it analyzes the positional relationship between pixels and combines it with the pathological feature indicators of pixels to identify artifact noise points on the initial tumor boundary line; and removes artifact noise points on the initial tumor boundary line to obtain the true tumor boundary line. The prognostic prediction module is used to predict survival based on the morphological characteristics of the actual tumor boundary and in combination with clinicopathological data.

[0005] Furthermore, the method for obtaining the tumor nucleus region includes: The grayscale image of the pathological slide was segmented using the Otsu threshold segmentation method. Regions with grayscale values ​​less than the segmentation threshold were taken as target regions. Connectivity analysis was performed on the pixels in the target regions to obtain the cell nucleus region. In each cell nucleus region, the total number of pixels is used as the quantity factor, and the average gray value of all pixels is used as the gray factor. Based on the quantity factor of all nuclear regions, the Otsu threshold segmentation method is used to process the nuclear regions. The nuclear regions with the quantity factor greater than the quantity segmentation threshold are selected as the first candidate regions. Based on the gray factor of all nuclear regions, the Otsu threshold segmentation method is used to process the nuclear regions. The nuclear regions with the gray factor less than the gray segmentation threshold are selected as the second candidate regions. If a cell nucleus region is both a first candidate region and a second candidate region, then that cell nucleus region is considered a tumor nucleus region.

[0006] Furthermore, the method for obtaining the suspicious analysis area includes: In the grayscale image of the pathological slide, a neighborhood window of a preset size is constructed with each pixel as the center; Within the neighborhood window corresponding to each pixel, the total number of tumor nucleus regions that are completely located within the neighborhood window and the total number of tumor nucleus regions that are not completely located within the neighborhood window but have an intersection with the neighborhood window are counted and used as a quantity parameter; Among all the quantity parameters corresponding to all pixels, the largest quantity parameter is used as the denominator, and the quantity parameter corresponding to each pixel is used as the numerator. The resulting ratio is used as the tumor density of each pixel in the grayscale image of the pathological section. In the grayscale image of the pathological section, the Otsu threshold segmentation method is used to process the tumor density of all pixels, and the pixels with tumor density greater than the density segmentation threshold are taken as target pixels. Morphological dilation is performed on all target pixels to obtain all suspicious analysis regions.

[0007] Furthermore, the method for obtaining the pathological feature indicators includes: For each pixel within the suspected analysis area, multiple local neighborhoods of preset sizes are constructed; Within each preset size local neighborhood corresponding to each pixel, the total number of tumor nucleus regions completely located within the local neighborhood and tumor nucleus regions not completely located within the local neighborhood but intersecting with the local neighborhood is used as a quantity coefficient, and the tumor nucleus regions completely located within the local neighborhood are used as the region to be analyzed. Among all the quantity coefficients corresponding to all pixels, the largest quantity coefficient is used as the denominator, and the quantity coefficient corresponding to each pixel is used as the numerator. The resulting ratio is used as the density factor of each pixel in each suspicious analysis area. Within each suspicious analysis area, the density factor of each pixel is weighted and summed using the preset weights corresponding to the local neighborhoods under each preset size to obtain the weighted density of each pixel. In each suspected analysis area, select a pixel as the pixel to be tested. For the local neighborhood of the pixel to be tested at each preset size, calculate the coefficient of variation of the number of pixels in all the analysis areas within the local neighborhood as the number coefficient of variation; calculate the coefficient of variation of the gray value of the pixels in all the analysis areas within the local neighborhood as the gray value coefficient of variation. The value obtained by negatively correlated mapping of the sum of the quantity variation coefficient and the grayscale variation coefficient is used as the morphological consistency factor of the pixel under test at each preset size. The sum of the morphological consistency factors of the pixel under test at all preset sizes is normalized by multiplying the product of the weighted density corresponding to the pixel under test, and the pathological feature index of the pixel under test is obtained.

[0008] Furthermore, the method for obtaining the core region includes: In each suspicious analysis area, pixels with pathological feature indicators greater than the preset pathological threshold are used as feature points. Morphological closing operations and hole filling are performed on the feature points in each suspicious analysis area to obtain the core area.

[0009] Furthermore, the method for obtaining the initial tumor boundary line includes: In each core region, starting from the centroid and ending at each edge pixel of the core region, obtain the connecting line segment and the sequence of pixels that the connecting line segment passes through; In each pixel sequence, the difference between the pathological feature index of each pixel and the next adjacent pixel is calculated as the change value corresponding to each pixel. The mean of the change values ​​of all pixels is used as the mean feature value. The maximum value of the change value is searched in the pixel sequence and traversed from the starting point. The pixel corresponding to the first maximum value point that is greater than the mean feature value is used as the split point. The initial tumor boundary line is obtained by smoothly connecting the segmentation points in all pixel sequences.

[0010] Furthermore, the method for obtaining the artifact noise points includes: Uniform sampling was performed along each initial tumor boundary line to obtain multiple sampling points; Obtain the normal line corresponding to each sampling point on the initial tumor boundary line, and construct a preset window centered on each sampling point in the normal line direction. Within the preset window, calculate the difference between the pathological feature index of each pixel and the next adjacent pixel, and use it as the change parameter corresponding to the previous pixel in the two pixels. Within the preset window, take the pixel corresponding to the maximum change parameter as the new sampling point. Analyze the location distribution characteristics of new sampling points in the corresponding core region to screen out protrusion candidate points from all new sampling points; For each candidate protrusion, a sequence of pathological feature indicators of the pixels on the centroid line connecting it to the corresponding core region is obtained and fitted to a curve. The fitted curve is divided into several local windows. The mean of the fitted values ​​on the fitted curve within each local window is calculated as the fitted mean. The variance of the fitted mean of all local windows is used as the fluctuation factor of each candidate protrusion. The fluctuation factor of all candidate points is processed using the Otsu threshold segmentation method, and candidate points with fluctuation factors greater than the fluctuation segmentation threshold are identified as artifact noise points.

[0011] Furthermore, the method for obtaining the candidate protrusion points includes: Calculate the Euclidean distance between each new sampling point and the centroid of the corresponding core region, and obtain the mean of the Euclidean distances corresponding to all new sampling points as the mean distance. When the difference between the Euclidean distance corresponding to a new sampling point and the mean distance is greater than a preset distance change threshold, the new sampling point is selected as a candidate protrusion point.

[0012] Furthermore, the method for obtaining the actual tumor boundary line includes: On the initial tumor boundary line, a predetermined number of non-artifact noise points are selected on both sides of each artifact noise point as smoothing reference points; The smooth reference point is fitted with a polynomial curve using the least squares method to obtain a smooth curve segment. The artifact noise points are replaced with the corresponding projection points on the smooth curve segment. The projection points and non-artifact noise points are connected sequentially to obtain the real tumor boundary line.

[0013] Furthermore, the survival prediction based on the morphological characteristics of the actual tumor boundary line, combined with clinicopathological data, includes: Calculate the area of ​​the region enclosed by the actual tumor boundary line, and the perimeter of the actual tumor boundary line; The ratio of the square of the perimeter to the area is used as the irregularity of the actual tumor boundary line. The pixel point on the actual tumor boundary line that is farthest from the centroid of the core region is obtained, and the Euclidean distance from the pixel point to the centroid is calculated as the infiltration depth. The irregularity and the depth of invasion, combined with the patient's clinicopathological data, are input into a pre-trained Cox proportional hazards regression model to obtain a prognostic index. If the prognostic index is greater than or equal to the preset prognostic risk threshold, an early warning is required; otherwise, no early warning is required.

[0014] The present invention has the following beneficial effects: The data acquisition module first acquires grayscale images of pathological sections from the patient's lung biopsy tissue, along with clinicopathological data. Then, in the feature extraction module, because cell nuclei are stained darker, the tumor nucleus region is defined based on the pixel's grayscale value. Pixels belonging to lung tumor tissue in the image are surrounded by a large number of tumor cells. Therefore, in the global image, the distribution characteristics of pixels in the tumor nucleus region are analyzed, and a suspicious analysis area is defined, which helps to focus on areas that may have special pathological significance. Within the tumor core region, the density and grayscale of tumor cells usually exhibit more consistent characteristics. Therefore, in the suspicious analysis area, the density distribution, area size, and grayscale performance of the tumor nucleus region within the local area of ​​each pixel are analyzed to obtain pathological feature indicators for each pixel, reflecting information such as the tumor's cellular composition and growth status. Furthermore, in the boundary analysis module, a core region is defined within each suspicious analysis area based on pathological feature indicators between pixels. Since the pathological features significantly decrease as tumor tissue infiltrates outwards, the changes in pathological feature indicators of pixels in the core region are analyzed to extract the initial tumor boundary line. Then, based on the initial tumor boundary line, the positional relationships between pixels are analyzed and combined with pathological feature indicators to identify and remove artifact noise points, obtaining the true tumor boundary line. This fully considers the spatial and feature information of the image, accurately distinguishing between the true tumor boundary and artifact noise. Finally, in the prognostic prediction module, tumor morphological features are effectively extracted to reflect the tumor's growth pattern and invasiveness. Combined with clinicopathological data, survival prediction is performed, enabling a more comprehensive and accurate assessment of the patient's prognosis and improving the accuracy and reliability of survival prediction. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system block diagram of a prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue provided in one embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure of a prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue, provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a prognostic survival prediction system based on lung biopsy tissue pathological characteristics proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of a prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a system block diagram of a prognostic survival prediction system based on lung biopsy tissue pathological features according to an embodiment of the present invention. The system includes: a data acquisition module 101, a feature extraction module 102, a boundary analysis module 103, and a prognostic prediction module 104.

[0021] The data acquisition module 101 is used to acquire grayscale images of pathological sections of lung biopsy tissue from patients and clinical pathological data.

[0022] Experienced clinicians use specialized biopsy tools, such as puncture needles or bronchoscopes, to collect appropriate tissue samples from the patient's lung lesions. Strict aseptic techniques and operating procedures must be followed throughout the process to ensure the collected samples are representative and uncontaminated. The collected lung tissue samples are quickly placed in a container containing a fixative (such as 10% neutral formalin solution). The fixative rapidly penetrates the tissue, preventing autolysis and putrefaction, preserving cell morphology and structure, and providing favorable conditions for subsequent grayscale imaging of pathological sections. The fixed lung tissue samples are then sequentially immersed in ethanol solutions of different concentrations for dehydration, gradually replacing the water in the tissue with ethanol. Next, the samples are placed in a clearing agent such as xylene to make the tissue transparent, facilitating subsequent paraffin embedding and sectioning. The cleared samples are then immersed in molten paraffin for several hours to allow the paraffin to fully penetrate the tissue. Finally, the sample and paraffin are poured together into an embedding cassette. After the paraffin cools and solidifies, a paraffin block containing the lung tissue sample is formed. Use a microtome to cut the paraffin block into slices of uniform thickness (usually 3-5 micrometers). Gently place the slices into warm water to spread them out, removing any wrinkles and overlaps, so that they adhere smoothly to the glass slide.

[0023] To clearly display cellular structures and tissue components in grayscale images of pathological sections, staining is necessary. A commonly used staining method is hematoxylin-eosin (HE) staining. Hematoxylin stains the cell nucleus blue, while eosin stains the cytoplasm and extracellular matrix pink. After staining, the morphology and structure of cells and tissues are clearer, facilitating observation and diagnosis by pathologists.

[0024] The stained pathological sections are observed under a microscope. The microscope's focal length and magnification are adjusted to ensure that the cells and tissue structures in the sections are clearly displayed in the field of view. The microscope is connected to a professional image acquisition device, such as a digital camera or a high-resolution microscope camera. Parameters such as image resolution, color mode, and exposure time are set according to the acquisition requirements. To obtain a grayscale image, the color mode is set to grayscale mode to ensure that the acquired image accurately reflects the grayscale information of the tissue. This yields a grayscale image of the pathological section from the patient's lung biopsy.

[0025] At the same time, it is also necessary to obtain the patient's clinical pathology data and query the patient's basic information through the hospital's electronic medical record system, including name, gender, age, smoking history, tumor TNM stage (according to the 8th edition of the AJCC staging system), tumor histological subtype, degree of differentiation, and whether the patient has received adjuvant therapy. This information is an important basis for identifying the patient and conducting subsequent follow-up.

[0026] In this embodiment of the invention, the collection and acquisition of patients' personal information data are all authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals.

[0027] The feature extraction module 102 is used to divide the tumor nucleus region in the grayscale image of the pathological slide according to the grayscale value of the pixel; to analyze the distribution characteristics of the pixel in the tumor nucleus region in the grayscale image of the pathological slide and divide the suspicious analysis area; and to determine the pathological feature index of each pixel based on the density distribution, area size and grayscale performance of the tumor nucleus region within the local area of ​​each pixel in the suspicious analysis area.

[0028] In grayscale images of pathological sections, different tissue components exhibit varying grayscale values ​​due to differences in their cellular structure, density, and staining characteristics. The tumor nucleus, the core of tumor growth and development, contains the main cell population and crucial biological information. It is typically composed of densely packed tumor cells, and its grayscale value differs significantly from surrounding normal tissue or other lung tumor tissues. In hematoxylin-eosin (HE) staining, hematoxylin primarily binds to nucleic acids within the cell nucleus, staining them deep blue or bluish-purple. Therefore, in grayscale images, the grayscale value of the nucleus region is significantly lower than that of the cytoplasm and extracellular matrix. Thus, the tumor nucleus region can be first delineated in the grayscale image of the pathological section based on the grayscale values ​​of individual pixels.

[0029] Preferably, in one embodiment of the present invention, the method for obtaining the tumor nucleus region includes: Otsu's thresholding method is an adaptive image segmentation algorithm that can automatically find the optimal grayscale segmentation threshold between the foreground and background, avoiding the instability caused by manually setting a fixed threshold. Furthermore, based on the aforementioned analysis, the pixel grayscale values ​​in the cell nucleus region are relatively low. Therefore, Otsu's thresholding method is used to segment the grayscale image of the pathological slide, and regions with grayscale values ​​less than the adaptive segmentation threshold (which is the threshold automatically found by Otsu's thresholding method) are designated as target regions. Connectivity analysis is then performed on the pixels in the target regions to obtain the cell nucleus regions. At this point, each cell nucleus region represents a single cell nucleus object.

[0030] The aforementioned cell nuclei may contain not only tumor cell nuclei but also a large number of normal lung interstitial cell nuclei, lymphocyte nuclei, etc. Since normal cell nuclei are typically smaller and lighter-stained, while tumor cell nuclei are typically larger and darker-stained, the total number of pixels in each nucleus region is used as the quantity factor, and the average grayscale value of all pixels is used as the grayscale factor. The quantity factor reflects the size of the nucleus region; a larger value indicates a larger area. The grayscale factor characterizes the average staining depth of the nucleus region.

[0031] Then, based on the quantity factor of all nuclear regions, the Otsu threshold segmentation method is used to process them. Nuclear regions with a quantity factor greater than the quantity segmentation threshold (which is the threshold automatically found by the Otsu threshold segmentation method) are selected as the first candidate regions. Based on the gray factor of all nuclear regions, the Otsu threshold segmentation method is used to process them. Nuclear regions with a gray factor less than the gray factor segmentation threshold (which is the threshold automatically found by the Otsu threshold segmentation method) are selected as the second candidate regions.

[0032] Based on the foregoing analysis, tumor cell nuclei are characterized by their large area and deep staining. Therefore, if a certain nucleus region is both the first and second candidate region, it means that the nucleus region meets the aforementioned characteristics and is thus considered a tumor nucleus region.

[0033] It should be noted that Otsu's threshold segmentation method and connected component analysis are both well-known techniques, and the specific process will not be elaborated here.

[0034] Clinical studies have shown that certain special areas within a tumor, such as the infiltrative area at the tumor margin and the area surrounding the necrotic region within the tumor, are often closely related to the tumor's prognosis. Therefore, further analysis of the pixel distribution characteristics in the tumor core region within the grayscale image of pathological sections can be conducted to delineate suspicious analysis areas and achieve more precise localization.

[0035] Preferably, in one embodiment of the present invention, the method for obtaining the suspicious analysis area includes: Tumor cells do not exist in isolation in tissues, but rather aggregate and interact with each other. Therefore, in the grayscale image of a pathological section, a neighborhood window of a preset size is constructed with each pixel as the center. This allows for consideration of local information within a certain range around the pixel, facilitating a more comprehensive understanding of the distribution characteristics of the tumor nucleus within that local range.

[0036] Within the neighborhood window corresponding to each pixel, the total number of tumor nuclei regions completely located within the neighborhood window and those not completely located within the neighborhood window but intersecting with it are counted as a quantity parameter. This quantity parameter reflects the degree of aggregation of tumor cell nuclei regions within the neighborhood window of that pixel. Among the quantity parameters corresponding to all pixels, the largest quantity parameter is used as the denominator, and the quantity parameter corresponding to each pixel is used as the numerator. The resulting ratio is used as the tumor density of each pixel in the grayscale image of the pathological section. The higher the tumor density, the more significant the aggregation of tumor nuclei regions around that pixel.

[0037] Then, in the grayscale image of the pathological section, the tumor density of all pixels is processed using the Otsu threshold segmentation method. Pixels with a tumor density greater than the density segmentation threshold (which is the threshold automatically found by the Otsu threshold segmentation method) are taken as target pixels. Morphological dilation is then performed on all target pixels to obtain all suspicious analysis areas.

[0038] It should be noted that the size of the neighborhood window is set to a circular window with a radius of 50 pixels centered on each pixel. The specific size can be adjusted according to the implementation scenario and is not limited here. The Otsu threshold segmentation method and morphological dilation operation are well-known techniques and will not be described in detail here.

[0039] At this point, the analyzable area in the grayscale image of the suspected pathological section was obtained, which included the core region of the tumor cells and the invasion front.

[0040] In the core region of a tumor, the density of tumor cells is usually high and they exhibit a specific arrangement pattern. However, the density of tumor cells decreases in the infiltrated part. Furthermore, tumor cells often grow clonally. Therefore, the density distribution, area size, and grayscale appearance of the tumor nucleus region tend to have high consistency. Thus, in the suspicious analysis area, we can analyze the consistency of the density distribution, area size, and grayscale appearance of the tumor nucleus region within the local area of ​​each pixel to determine the pathological feature indicators of each pixel, which can be used to reflect whether the pixel is located at the tumor infiltration boundary.

[0041] Preferably, in one embodiment of the present invention, the method for obtaining pathological feature indicators includes: While tumor cells exhibit a consistent distribution, differences in cell density and nuclear arrangement may exist in the tumor core and invasion front. Therefore, for each pixel within the suspected analysis area, multiple local neighborhoods of preset sizes are constructed to comprehensively analyze the distribution of tumor cells from multiple perspectives. It should be noted that in this embodiment of the invention, the local neighborhood is a circle centered on each pixel with a preset radius. The preset size can be set to 10, 20, or 30 pixels. In this embodiment, three preset sizes are used; the specific number and value can be adjusted according to the implementation scenario and are not limited here.

[0042] Then, within each preset size local neighborhood corresponding to each pixel, the total number of tumor nucleus regions that are completely located within the local neighborhood and the total number of tumor nucleus regions that are not completely located within the local neighborhood but have an intersection with the local neighborhood are used as a quantity coefficient. The quantity coefficient reflects the quantity characteristics of tumor nucleus regions within the local neighborhood corresponding to each pixel. The larger the value, the denser the distribution of tumor nucleus regions within the local neighborhood at that preset size.

[0043] Among all the quantity coefficients corresponding to all pixels, the largest quantity coefficient is used as the denominator, and the quantity coefficient corresponding to each pixel is used as the numerator. The resulting ratio is used as the density factor of each pixel in each suspicious analysis area. The larger the density factor, the higher the distribution density of the tumor core region.

[0044] Then, within each suspicious analysis region, the density factor of each pixel is weighted and summed using the preset weights corresponding to the local neighborhoods under each preset size (multiplying the preset weights corresponding to the local neighborhoods under the same preset size with the density factors and then summing them up). This yields the weighted density of each pixel. The weighted density integrates the reference value and importance of the local neighborhoods under each preset size, thus the weighted density can more accurately reflect the density characteristics of tumor cells, and the larger the value, the higher the density. In this embodiment of the invention, since the distribution of tumor cells is heterogeneous, a small window can capture more detailed local information, while a large window can capture overall information. Therefore, a small window should have a higher weight. Thus, the preset weights of the local neighborhoods corresponding to 10 pixels, 20 pixels, and 30 pixels are 0.5, 0.3, and 0.2, respectively.

[0045] Given that the dense clusters of lymphocytes and proliferating fibroblasts in lung tissue can also form high-density areas, relying solely on the density of the tumor nucleus region may lead to misjudgment. Furthermore, during the clonal growth of tumor cells, they typically exhibit a certain degree of morphological and size uniformity. Therefore, within the local neighborhood corresponding to each pixel at each preset size, the tumor nucleus region completely located within the local neighborhood is taken as the region to be analyzed.

[0046] Within each suspected analysis region, a pixel is randomly selected as the test pixel. For the local neighborhood of the test pixel at each preset size, the coefficient of variation of the number of pixels in all analysis regions within that local neighborhood is calculated as the quantity coefficient of variation, and the coefficient of variation of the grayscale value of the pixels in all analysis regions within that local neighborhood is calculated as the grayscale coefficient of variation. The coefficient of variation measures the dispersion of the data; therefore, the larger the quantity coefficient of variation, the more consistent the number of pixels in the analysis region, and the more uniform their size and distribution. Similarly, the larger the grayscale coefficient of variation, the more consistent the grayscale performance of the pixels in the analysis region. The sum of the quantity coefficient of variation and the grayscale coefficient of variation is negatively correlated and mapped to the morphological consistency factor of the test pixel at each preset size. Based on the previous analysis, a larger morphological consistency factor is considered to be more consistent with the clonal growth characteristics of tumor cells. The negative correlation mapping here can be performed using the formula... ,in, Let x represent an exponential function with the natural constant e as the base, and let x represent the independent variable.

[0047] Finally, the sum of the morphological consistency factors of the pixels under test at all preset sizes is normalized by multiplying the product of the weighted density of the pixels under test, resulting in the pathological feature index of the pixels under test. This index comprehensively considers the density and morphological consistency of tumor cells, and can more accurately reflect the biological characteristics of tumor cells. The larger the pathological feature index, the more obvious the tumor pathological features of the pixels under test. Normalization is a technique well known to those skilled in the art. The normalization function can be linear normalization or standard normalization, such as maximum and minimum value normalization, where the maximum and minimum values ​​can be extracted from a database of all pixels.

[0048] The boundary analysis module 103 is used to divide the core region in each suspicious analysis area according to the pathological feature indicators between pixels; in each core region, it analyzes the changing characteristics of the pathological feature indicators of pixels and extracts the initial tumor boundary line; based on the initial tumor boundary line, it analyzes the positional relationship between pixels and combines it with the pathological feature indicators of pixels to identify artifact noise points on the initial tumor boundary line; and removes artifact noise points on the initial tumor boundary line to obtain the true tumor boundary line.

[0049] As tumor tissue transitions towards its boundary, the density of the tumor nucleus decreases, and the clonal growth characteristics of the cells also weaken. Therefore, in each suspicious analysis area, the core region can be delineated based on the pathological feature indicators between pixels.

[0050] Preferably, in one embodiment of the present invention, the method for obtaining the core region includes: In each suspicious analysis area, pixels with pathological feature indicators greater than the preset pathological threshold are used as feature points. Morphological closing operations and hole filling are performed on the feature points in each suspicious analysis area to obtain the core area.

[0051] It should be noted that the preset pathological threshold is set to 0.65, and the specific value can be adjusted according to the implementation scenario, which is not limited here; morphological closure operation and cavity filling are well known techniques, and will not be elaborated here.

[0052] Tumors do not grow uniformly; rather, they gradually thin out from the core area outwards, meaning the boundary between the infiltrated area and the surrounding normal tissue gradually becomes blurred. In this step, the pathological features of each pixel can be analyzed in each core area to extract the initial tumor boundary line, which is used to initially define the extent of the tumor.

[0053] Preferably, in one embodiment of the present invention, the method for obtaining the initial tumor boundary line includes: The centroid is the geometric center and can represent the overall position of a region. Therefore, the centroid of the core region is obtained in each core region. Starting from the centroid and ending at each edge pixel of the core region, the line segments and the sequence of pixels that the line segments pass through are obtained. By analyzing the pixel sequence, the pathological features of the tumor tissue from the inside to the outside are analyzed.

[0054] As the tumor approaches its boundary from the centroid, tumor features show a significant decrease. Therefore, in each pixel sequence, the difference between the pathological feature indicators of each pixel and its next adjacent pixel is calculated as the change value for each pixel. A positive change value, and the larger the value, the greater the decrease in the pathological feature indicators between the two adjacent pixels, indicating that the tumor boundary is very likely to be reached. Thus, the mean of the change values ​​of all pixels is used as the mean feature value. The maximum value of the change value is searched in the pixel sequence, and the process is traversed starting from the starting point (centroid). The pixel corresponding to the first maximum value greater than the mean feature value is used as the segmentation point, which is very likely to be a pixel on the tumor boundary.

[0055] Finally, the segmentation points in all pixel sequences are smoothly connected to obtain the initial tumor boundary line.

[0056] The initial tumor boundary line obtained by smooth connection may classify some pixels located inside the tumor as outside the edge. In this process, it may be affected by various factors, such as noise during image acquisition and the complexity of tissue structure, resulting in some pixels on the initial boundary that do not belong to the real tumor boundary, i.e., artifact noise points. These artifact noise points will blur the real boundary of the tumor and affect the accuracy of subsequent processes.

[0057] On the real tumor boundary, the positional relationship of pixels usually follows a certain pattern. However, noise spikes are more prominent in shape than the real tumor boundary. Furthermore, when the boundary point is a noise spike, the pathological feature index will also decrease. Therefore, based on the initial tumor boundary line, we analyze the positional relationship between pixels and combine it with the pathological feature index of the pixels to identify artifact noise points on the initial tumor boundary line.

[0058] Preferably, in one embodiment of the present invention, the method for obtaining artifact noise points includes: The initial tumor boundary line is usually composed of a large number of continuous pixels. Directly analyzing all pixels would be computationally intensive and unnecessary. Therefore, we first perform uniform sampling along each initial tumor boundary line (e.g., sampling once every other pixel) to obtain multiple sampling points.

[0059] Obtain the normal (a straight line perpendicular to the tangent) corresponding to each sampling point on the initial tumor boundary line. Construct a pre-defined window (size 11, meaning the window includes each sampling point and 5 pixels on each side of the sampling point along the normal) centered on each sampling point along the normal direction. Within the pre-defined window, calculate the difference in pathological feature indicators between each pixel and its next adjacent pixel (the difference in pathological feature indicators between pixels closer to the centroid and pixels farther from the centroid). This difference is used as the change parameter for the preceding pixel. The change parameter is positive, and the larger the value, the more likely the preceding pixel is a boundary point. Therefore, within the pre-defined window, the pixel corresponding to the largest change parameter is selected as the new sampling point. After this update, all new sampling points will more accurately fall at locations with more significant changes.

[0060] Then, the positional distribution characteristics of the new sampling points in the corresponding core region are analyzed to screen out protrusion candidate points from all the new sampling points: the Euclidean distance between each new sampling point and the centroid of the corresponding core region is calculated, and the mean of the Euclidean distances of all new sampling points is obtained as the distance mean. When the difference between the Euclidean distance of a new sampling point and the distance mean is greater than the preset distance change threshold, it is considered that the positional distribution of the new sampling point has deviated significantly, that is, it may be an abnormal protrusion on the boundary caused by noise. Therefore, the new sampling point is regarded as a protrusion candidate point.

[0061] Furthermore, the obtained protrusion candidate points may also be real tumor boundary spurs. However, for real tumor boundary spurs, under the influence of infiltration features, the changes in pathological feature indicators between them and surrounding pixels should be relatively consistent. Conversely, for real noise spurs, they are often isolated. Therefore, for each protrusion candidate point, the sequence of pathological feature indicators of the pixels on the centroid line connecting it to the corresponding core region is obtained and a curve is fitted. The fitted curve is divided into several local windows (for example, each local window consists of 3 pixels). The mean of the fitted values ​​on the fitted curve within each local window is calculated as the fitted mean. The variance of the fitted mean of all local windows is used as the fluctuation factor of each protrusion candidate point. The larger the fluctuation factor, the more likely the protrusion candidate point is to be a noise point.

[0062] Therefore, the Otsu threshold segmentation method is used to process the fluctuation factor of all protrusion candidate points, and protrusion candidate points with fluctuation factors greater than the fluctuation segmentation threshold are identified as artifact noise points.

[0063] It should be noted that the preset distance change threshold can be set to 5 pixels. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0064] The initial tumor boundary line may contain artifact noise points. Therefore, artifact noise points can be removed from the initial tumor boundary line to obtain the true tumor boundary line.

[0065] Preferably, in one embodiment of the present invention, the method for obtaining the actual tumor boundary line includes: Since artifact noise points can interfere with the true shape of the tumor boundary, directly processing them may yield inaccurate results. Therefore, on the initial tumor boundary line, a preset number of non-artifact noise points are selected on both sides of each artifact noise point as smoothing reference points.

[0066] Then, the least squares method is used to perform polynomial curve fitting on the smooth reference point to obtain the fitted smooth curve segment. The artifact noise points are replaced with the corresponding projection points on the smooth curve segment. Finally, the projection points and non-artifact noise points are connected in sequence to obtain the real tumor boundary line.

[0067] It should be noted that the preset quantity is set to 8, and the specific value can be adjusted according to the implementation scenario, which is not limited here; the least squares method is a well-known technique, and the specific process will not be described in detail here.

[0068] The prognostic prediction module 104 is used to predict survival based on the morphological characteristics of the actual tumor boundary line and in combination with clinicopathological data.

[0069] After obtaining the true tumor boundary, morphological indicators that reflect the invasiveness of the tumor can be extracted based on its morphological characteristics, and then combined with the patient's clinicopathological data to predict survival.

[0070] Preferably, in one embodiment of the present invention, survival prediction based on the morphological characteristics of the actual tumor boundary line and in combination with clinicopathological data includes: Calculate the area enclosed by the true tumor boundary line and the perimeter of the true tumor boundary line. The area directly reflects the size of the space occupied by the tumor in a two-dimensional plane and is an important indicator for measuring the size of the tumor. The perimeter describes the length of the tumor boundary line and is related to the outline of the tumor.

[0071] Then, the ratio of the square of the perimeter to the area is used as the irregularity of the actual tumor boundary. Irregularity is a comprehensive index that can characterize the degree of irregularity of the tumor boundary. The larger the value, the more complex and irregular the tumor boundary.

[0072] Obtain the pixel furthest from the centroid of the core region on the actual tumor boundary line, and calculate the Euclidean distance from this pixel to the centroid as the invasion depth. The invasion depth describes the extent to which the tumor infiltrates from the core region into the surrounding tissues. The larger the value, the wider the range of tumor infiltration, and the more severe the invasion of the surrounding normal tissues.

[0073] Finally, the irregularity and depth of invasion, combined with the patient's clinicopathological data, are input into a pre-trained Cox proportional hazards regression model to obtain a prognostic index. The prognostic index should be between 0 and 1, with a higher value indicating a higher risk of recurrence. Therefore, if the prognostic index is greater than or equal to the preset prognostic risk threshold, an alert is required; otherwise, no alert is needed.

[0074] It should be noted that the preset prognostic risk threshold is 0.75, and the specific value can be adjusted according to the implementation scenario, which is not limited here; the training process of the Cox proportional hazards regression model is a well-known technique and will not be described in detail here.

[0075] Unless otherwise specified, the normalized values ​​in the embodiments of this invention are generally in the range of 0 to 1.

[0076] In summary, the data acquisition module first acquires grayscale images of pathological sections from the patient's lung biopsy tissue, along with clinicopathological data. Then, in the feature extraction module, since cell nuclei are stained darker, the tumor nucleus region is defined based on the pixel's grayscale value. Pixels belonging to lung tumor tissue in the image are surrounded by a large number of tumor cells. Therefore, in the global image, the distribution characteristics of pixels in the tumor nucleus region are analyzed, and a suspicious analysis area is defined, which helps to focus on areas that may have special pathological significance. Within the tumor core region, the density and grayscale of tumor cells usually exhibit more consistent characteristics. Therefore, in the suspicious analysis area, the density distribution, area size, and grayscale performance of the tumor nucleus region within the local area of ​​each pixel are analyzed to obtain pathological feature indicators for each pixel, reflecting information such as the tumor's cellular composition and growth status. Furthermore, in the boundary analysis module, a core region is defined within each suspicious analysis area based on pathological feature indicators between pixels. Since the pathological features significantly decrease as tumor tissue infiltrates outwards, the changes in pathological feature indicators of pixels in the core region are analyzed to extract the initial tumor boundary line. Then, based on the initial tumor boundary line, the positional relationships between pixels are analyzed and combined with pathological feature indicators to identify and remove artifact noise points, obtaining the true tumor boundary line. This fully considers the spatial and feature information of the image, accurately distinguishing between the true tumor boundary and artifact noise. Finally, in the prognostic prediction module, tumor morphological features are effectively extracted to reflect the tumor's growth pattern and invasiveness. Combined with clinicopathological data, survival prediction is performed, enabling a more comprehensive and accurate assessment of the patient's prognosis and improving the accuracy and reliability of survival prediction.

[0077] Please see Figure 2This illustration shows a schematic diagram of the system structure of a prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue provided by an embodiment of the present invention. It includes a processor 200, a memory 201, a bus 202, and a communication interface 203. The processor 200, communication interface 203, and memory 201 are connected via the bus 202. The memory 201 may include a high-speed random access memory, and the bus 202 may be an ISA bus, PCI bus, or EISA bus, etc. The processor 200 may be an integrated circuit chip with signal processing capabilities. The memory 201 stores at least one instruction, at least one program, code set, or instruction set. When the processor loads and executes the at least one instruction, at least one program, code set, or instruction set, it implements the steps in the aforementioned modules.

[0078] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0079] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A prognostic survival prediction system based on the pathological characteristics of lung biopsy tissue, characterized in that, The system includes: The data acquisition module is used to acquire grayscale images of pathological sections of lung biopsy tissue from patients, as well as clinical pathological data. The feature extraction module is used to divide the tumor nucleus region in the grayscale image of the pathological section according to the grayscale value of the pixel; to analyze the distribution characteristics of the pixel in the tumor nucleus region in the grayscale image of the pathological section and divide the suspicious analysis area; and to determine the pathological feature index of each pixel based on the density distribution, area size and grayscale performance of the tumor nucleus region within the local area of ​​each pixel in the suspicious analysis area. The boundary analysis module is used to divide the core region in each suspicious analysis area based on the pathological feature indicators between pixels; in each core region, it analyzes the changing characteristics of the pathological feature indicators of pixels to extract the initial tumor boundary line; based on the initial tumor boundary line, it analyzes the positional relationship between pixels and combines it with the pathological feature indicators of pixels to identify artifact noise points on the initial tumor boundary line; and removes artifact noise points on the initial tumor boundary line to obtain the true tumor boundary line. The prognostic prediction module is used to predict survival based on the morphological characteristics of the actual tumor boundary and in combination with clinicopathological data.

2. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The method for obtaining the tumor nuclear region includes: The grayscale image of the pathological slide was segmented using the Otsu threshold segmentation method. Regions with grayscale values ​​less than the segmentation threshold were taken as target regions. Connectivity analysis was performed on the pixels in the target regions to obtain the cell nucleus region. In each cell nucleus region, the total number of pixels is used as the quantity factor, and the average gray value of all pixels is used as the gray factor. Based on the quantity factor of all nuclear regions, the Otsu threshold segmentation method is used to process the nuclear regions. The nuclear regions with the quantity factor greater than the quantity segmentation threshold are selected as the first candidate regions. Based on the gray factor of all nuclear regions, the Otsu threshold segmentation method is used to process the nuclear regions. The nuclear regions with the gray factor less than the gray segmentation threshold are selected as the second candidate regions. If a cell nucleus region is both a first candidate region and a second candidate region, then that cell nucleus region is considered a tumor nucleus region.

3. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The method for obtaining the suspicious analysis area includes: In the grayscale image of the pathological slide, a neighborhood window of a preset size is constructed with each pixel as the center; Within the neighborhood window corresponding to each pixel, the total number of tumor nucleus regions that are completely located within the neighborhood window and the total number of tumor nucleus regions that are not completely located within the neighborhood window but have an intersection with the neighborhood window are counted and used as a quantity parameter; Among all the quantity parameters corresponding to all pixels, the largest quantity parameter is used as the denominator, and the quantity parameter corresponding to each pixel is used as the numerator. The resulting ratio is used as the tumor density of each pixel in the grayscale image of the pathological section. In the grayscale image of the pathological section, the Otsu threshold segmentation method is used to process the tumor density of all pixels, and the pixels with tumor density greater than the density segmentation threshold are taken as target pixels. Morphological dilation is performed on all target pixels to obtain all suspicious analysis regions.

4. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The methods for obtaining the pathological feature indicators include: For each pixel within the suspected analysis area, multiple local neighborhoods of preset sizes are constructed; Within each preset size local neighborhood corresponding to each pixel, the total number of tumor nucleus regions completely located within the local neighborhood and tumor nucleus regions not completely located within the local neighborhood but intersecting with the local neighborhood is used as a quantity coefficient, and the tumor nucleus regions completely located within the local neighborhood are used as the region to be analyzed. Among all the quantity coefficients corresponding to all pixels, the largest quantity coefficient is used as the denominator, and the quantity coefficient corresponding to each pixel is used as the numerator. The resulting ratio is used as the density factor of each pixel in each suspicious analysis area. Within each suspicious analysis area, the density factor of each pixel is weighted and summed using the preset weights corresponding to the local neighborhoods under each preset size to obtain the weighted density of each pixel. In each suspected analysis area, select a pixel as the pixel to be tested. For the local neighborhood of the pixel to be tested at each preset size, calculate the coefficient of variation of the number of pixels in all the analysis areas within the local neighborhood as the number coefficient of variation; calculate the coefficient of variation of the gray value of the pixels in all the analysis areas within the local neighborhood as the gray value coefficient of variation. The value obtained by negatively correlated mapping of the sum of the quantity variation coefficient and the grayscale variation coefficient is used as the morphological consistency factor of the pixel under test at each preset size. The sum of the morphological consistency factors of the pixel under test at all preset sizes is normalized by multiplying the product of the weighted density corresponding to the pixel under test, and the pathological feature index of the pixel under test is obtained.

5. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The method for obtaining the core region includes: In each suspicious analysis area, pixels with pathological feature indicators greater than the preset pathological threshold are used as feature points. Morphological closing operations and hole filling are performed on the feature points in each suspicious analysis area to obtain the core area.

6. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The method for obtaining the initial tumor boundary line includes: In each core region, starting from the centroid and ending at each edge pixel of the core region, obtain the connecting line segment and the sequence of pixels that the connecting line segment passes through; In each pixel sequence, the difference between the pathological feature index of each pixel and the next adjacent pixel is calculated as the change value corresponding to each pixel. The mean of the change values ​​of all pixels is used as the mean feature value. The maximum value of the change value is searched in the pixel sequence and traversed from the starting point. The pixel corresponding to the first maximum value point that is greater than the mean feature value is used as the split point. The initial tumor boundary line is obtained by smoothly connecting the segmentation points in all pixel sequences.

7. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The method for obtaining artifact noise points includes: Uniform sampling was performed along each initial tumor boundary line to obtain multiple sampling points; Obtain the normal line corresponding to each sampling point on the initial tumor boundary line, and construct a preset window centered on each sampling point in the normal line direction. Within the preset window, calculate the difference between the pathological feature index of each pixel and the next adjacent pixel, and use it as the change parameter corresponding to the previous pixel in the two pixels. Within the preset window, take the pixel corresponding to the maximum change parameter as the new sampling point. Analyze the location distribution characteristics of new sampling points in the corresponding core region to screen out protrusion candidate points from all new sampling points; For each candidate protrusion, a sequence of pathological feature indicators of the pixels on the centroid line connecting it to the corresponding core region is obtained and fitted to a curve. The fitted curve is divided into several local windows. The mean of the fitted values ​​on the fitted curve within each local window is calculated as the fitted mean. The variance of the fitted mean of all local windows is used as the fluctuation factor of each candidate protrusion. The fluctuation factor of all candidate points is processed using the Otsu threshold segmentation method, and candidate points with fluctuation factors greater than the fluctuation segmentation threshold are identified as artifact noise points.

8. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 7, characterized in that, The method for obtaining the candidate protrusion points includes: Calculate the Euclidean distance between each new sampling point and the centroid of the corresponding core region, and obtain the mean of the Euclidean distances corresponding to all new sampling points as the mean distance. When the difference between the Euclidean distance corresponding to a new sampling point and the mean distance is greater than a preset distance change threshold, the new sampling point is selected as a candidate protrusion point.

9. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The method for obtaining the actual tumor boundary line includes: On the initial tumor boundary line, a predetermined number of non-artifact noise points are selected on both sides of each artifact noise point as smoothing reference points; The smooth reference point is fitted with a polynomial curve using the least squares method to obtain a smooth curve segment. The artifact noise points are replaced with the corresponding projection points on the smooth curve segment. The projection points and non-artifact noise points are connected sequentially to obtain the real tumor boundary line.

10. The prognostic survival prediction system based on lung biopsy tissue pathological characteristics according to claim 1, characterized in that, The survival prediction based on the morphological characteristics of the actual tumor boundary line, combined with clinicopathological data, includes: Calculate the area of ​​the region enclosed by the actual tumor boundary line, and the perimeter of the actual tumor boundary line; The ratio of the square of the perimeter to the area is used as the irregularity of the actual tumor boundary line. The pixel point on the actual tumor boundary line that is farthest from the centroid of the core region is obtained, and the Euclidean distance from the pixel point to the centroid is calculated as the infiltration depth. The irregularity and the depth of invasion, combined with the patient's clinicopathological data, are input into a pre-trained Cox proportional hazards regression model to obtain a prognostic index. If the prognostic index is greater than or equal to the preset prognostic risk threshold, an early warning is required; otherwise, no early warning is required.

Images