A pathological image analysis system and method based on image processing technology

Abstract

The present invention belongs to the technical field of pathological image analysis, and specifically relates to a pathological image analysis system and method based on image processing technology, which selects samples of batch pathological images, determines an adaptive image segmentation method based on the visual features of the pathological image samples, pays attention to the recognition effect when performing pathological recognition after the sample image segmentation using the adaptive image segmentation method, and then adjusts the segmentation method of the remaining pathological images based on the recognition effect, thereby realizing closed-loop feedback adjustment of the batch pathological image segmentation method. This method can not only timely discover poor recognition performance and avoid continuous occurrence of erroneous segmentation, but also the adjustment of the segmentation method is based on feedback data, has reliable data support, does not require a large amount of manual review and correction work, and does not rely on the experience of pathologists for review, thereby maximizing the pertinence and effectiveness of the segmentation method adjustment.

Description

Technical Field

[0001] The present invention belongs to the technical field of pathological image analysis, and in particular relates to a pathological image analysis system and method based on image processing technology. Background Art

[0002] Pathological analysis is an indispensable part of modern medicine and plays a key role in the diagnosis, treatment and research of diseases. Pathological analysis can significantly improve the accuracy of diagnosis and treatment. With the development of digital pathology and computer vision technology, digital pathological images have been widely used in pathological analysis. This transformation not only realizes the automated detection and quantification of pathological features, reducing the workload of pathologists, but also ensures the accuracy of measurements such as cell size and morphological features. In addition, digital pathological images are easy to store and transmit, which promotes remote consultation and cooperation among multidisciplinary teams.

[0003] Pathological analysis based on pathological images includes image preprocessing, image segmentation, pathological recognition and other processes. Image segmentation is a key technical link in digital pathological image analysis. Its purpose is to separate different objects or areas in the image for subsequent pathological recognition. When multiple patients have similar symptoms or signs, hospitals need to perform pathological analysis on the tissue samples of these patients to confirm whether a certain disease exists. In this case, in order to ensure the consistency of the image and the reliability of the analysis, the hospital may choose to use the same microscope scanner and scan the batch of tissues under the same imaging conditions to form a batch of pathological images. When performing image segmentation on batch pathological images, in order to improve efficiency, the same segmentation method is usually used to segment all pathological images. This method lacks real-time self-feedback adjustment of the pathological recognition effect after segmentation. Using a single segmentation method may lead to insensitivity to certain features of the pathological image, so that when the recognition algorithm performs poorly, the error will not be discovered and corrected in time, resulting in an increase in mis-segmentation. In addition, when the recognition effect is not ideal after all the batch pathological images are segmented, pathologists need to perform a lot of manual review and correction work. On the one hand, the workload is increased invisibly. On the other hand, manual review may rely on the experience of pathologists, making it difficult to ensure the consistency of the results. Summary of the invention

[0004] The purpose of the present invention is to improve the deficiencies in the prior art and to provide a pathological image analysis system and method based on image processing technology. By establishing a real-time feedback mechanism for segmentation results, the segmentation method of batch pathological images is dynamically optimized and adjusted, thereby effectively making up for the problems existing in the prior art.

[0005] The objectives of the present invention can be achieved through the following technical solutions: According to one aspect of an embodiment of the present application, a pathological image analysis system based on image processing technology is provided, comprising the following modules: a pathological image sample selection module, for selecting samples of batch pathological images of the same tissue scanned by the same microscope scanner under the same imaging conditions, and performing image preprocessing.

[0006] The image segmentation method determination module is used to extract visual features from the pathological image samples after image preprocessing, thereby determining the adaptive image segmentation method of the pathological image samples.

[0007] The image segmentation operation module is used to perform image segmentation on the pathological image sample according to the adapted image segmentation method to obtain a main body image retained after segmentation.

[0008] The pathology recognition module is used to perform pathology recognition on the main image retained by the pathology image sample and obtain indicators of the pathology recognition effect, specifically the recognition time and the degree of consistency between the recognition result and the expected biomarker, and evaluate the pathology recognition effect based on this.

[0009] The pathology recognition log analysis module is used to retrieve the log files generated by pathology recognition, extract the operation steps and timestamps corresponding to each log record, and perform pathology recognition image processing time ratio analysis.

[0010] Pairing library for image segmentation comparisons corresponding to various visual features.

[0011] The preprocessing optimization demand judgment module is used to judge whether there is an image preprocessing optimization demand from the analysis of the proportion of pathology recognition image processing time of pathology image samples, and output the optimization direction if so.

[0012] The image segmentation replacement demand judgment module is used to judge whether there is a need to replace the image segmentation method based on the pathology recognition effect of the pathology image sample, and if there is a need for replacement, determine the image segmentation method to be replaced.

[0013] The preprocessing optimization module is used to perform image preprocessing on the remaining pathological images according to the image preprocessing optimization orientation.

[0014] The image segmentation and recognition optimization module is used to perform image segmentation and pathology recognition on the remaining pathological images after image preprocessing by using a changed image segmentation method.

[0015] According to another aspect of an embodiment of the present application, a pathological image analysis method based on image processing technology is provided, comprising the following steps: S1, selecting samples of batch pathological images of the same tissue scanned by the same microscope scanner under the same imaging conditions, and performing image preprocessing.

[0016] S2. Extracting visual features from the pathological image samples after image preprocessing, thereby determining an adapted image segmentation method for the pathological image samples.

[0017] S3. Perform image segmentation on the pathological image sample according to the adapted image segmentation method to obtain a main body image retained after segmentation.

[0018] S4. Perform pathological recognition on the main image retained by the pathological image sample, and obtain indicators of the pathological recognition effect, specifically the recognition time and the degree of consistency between the recognition result and the expected biomarker, and evaluate the pathological recognition effect based on this.

[0019] S5. Retrieve the log file generated by pathology recognition, extract the operation steps and timestamps corresponding to each log record, and perform pathology recognition image processing time ratio analysis.

[0020] S6. From the analysis of the proportion of pathology recognition image processing time of pathology image samples, determine whether there is a need for image preprocessing optimization, and if so, output the optimization direction.

[0021] S7. Based on the pathology recognition effect of the pathology image sample, it is judged whether there is a need to change the image segmentation method. If there is a need to change, the image segmentation method to be replaced is determined.

[0022] S8. Perform image preprocessing on the remaining pathological images according to the image preprocessing optimization orientation.

[0023] S9. Use the changed image segmentation method to perform image segmentation and pathology recognition on the remaining pathological images after image preprocessing.

[0024] In combination with all the above-mentioned technical solutions, the positive effects of the present invention are as follows: 1. The present invention determines an adaptive image segmentation method by performing sample selection on batch pathology images, and then pays attention to the recognition effect when performing pathology recognition after sample image segmentation using the adaptive image segmentation method. The segmentation method of the remaining pathology images is adjusted based on the recognition effect, thereby realizing a closed-loop feedback adjustment of the batch pathology image segmentation method. Not only can poor performance of pathology recognition be discovered in a timely manner to avoid continuous occurrence of erroneous segmentation, but on the other hand, the adjustment of the segmentation method is based on real-time feedback data, which not only improves the accuracy and reliability of segmentation, but also reduces the need for manual intervention and improves overall work efficiency.

[0025] 2. When the present invention uses an adaptive image segmentation method to perform pathology recognition on the selected pathology image samples after segmentation, it not only pays attention to the recognition effect but also adds an analysis of the proportion of image processing time in the recognition process. This can provide feedback reference for the optimization of image preprocessing before image segmentation, and can timely discover the deficiencies in the preprocessing and optimize it, thereby improving the recognition efficiency for subsequent pathology recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0027] Figure 1 It is a schematic diagram of system module connection of the present invention.

[0028] Figure 2 It is a diagram of the steps for implementing the method of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] Example 1

[0031] The present invention provides a pathological image analysis system based on image processing technology, comprising a pathological image sample selection module, an image segmentation mode determination module, an image segmentation operation module, a pathological recognition module, a pathological recognition log analysis module, a preprocessing optimization demand judgment module, an image segmentation replacement demand judgment module, a preprocessing optimization module, an image segmentation recognition optimization module and a pairing library.

[0032] In the above-mentioned pathological image sample selection module is connected to the image segmentation method determination module, the image segmentation method determination module is connected to the image segmentation operation module, the image segmentation operation module is connected to the pathology recognition module, the pathology recognition module is connected to the pathology recognition log analysis module, the pathology recognition log analysis module is connected to the preprocessing optimization demand judgment module, the pathology recognition module is connected to the image segmentation replacement demand judgment module, the preprocessing optimization demand judgment module is connected to the preprocessing optimization module, the image segmentation replacement demand judgment module is connected to the image segmentation recognition optimization module, and the pairing library is connected to the image segmentation method determination module.

[0033] The pathological image sample selection module is used to select samples of batch pathological images of the same tissue scanned by the same microscope scanner under the same imaging conditions, and perform image preprocessing.

[0034] It should be added that the same imaging can be performed using the same staining method and the same magnification.

[0035] In the specific implementation of the above scheme, the sample selection process is as follows: the number of batch pathology images is counted, and each pathology image is numbered. The numbering can be numbered according to the order of image scanning, for example, starting from 1 and numbering continuously to the total number. In one example, if there are 1000 pathology images, the numbers are 1 to 1000.

[0036] The number of samples is determined from the number of batch pathology images according to the sample selection ratio.

[0037] Applied to the above scheme, the sample selection ratio can be determined according to the needs of the research or task. For example, 10% of the samples are selected, and the number of samples is 100 in the example of 1000 pathological images.

[0038] The interval numbers in the batch pathology images are determined according to the number of samples, and samples are selected based on this.

[0039] Further applied to the above scheme, if the number of samples is 100, the interval is 1000 divided by 100, which is 10. That is, starting from image numbered 1, one image is selected as a sample every 10 numbers. The final selected sample numbers are 1, 11, 21, 31...991, totaling 100 images.

[0040] It should be understood that when selecting samples for batch pathological images of the same tissue scanned by the same microscope scanner under the same imaging conditions, it is considered that the cell distribution, staining status and tissue structure of the pathological images obtained in this case have a certain degree of similarity, which means that the variation between images is relatively small, and the main purpose of sample selection is to ensure that the selected samples can represent the characteristics of the entire batch of images and be used for subsequent analysis. While ensuring the representativeness of the samples, the simplicity and efficiency of the selection process must also be considered. Therefore, the batch pathological images are not stratified according to cell distribution, staining status and tissue structure for sample selection. Instead, a systematic sampling method is adopted. On the one hand, it is simple and easy and does not require complex stratification or clustering algorithms. On the other hand, since there is no obvious periodic arrangement between the images, the representativeness of the samples can be well guaranteed under certain conditions.

[0041] In the optimization of the above scheme, the purpose of image preprocessing is to improve image quality, enhance useful features, and reduce interference factors, so as to improve the accuracy and efficiency of subsequent image segmentation and pathology recognition. The image preprocessing of the selected pathology image samples specifically includes image enhancement, image resizing, image denoising, etc. In order to improve the efficiency of image preprocessing and because the subsequent pathology recognition process is in an unknown state, the image preprocessing of the pathology image samples is generally processed according to the default processing method, and there may be a situation where the processing effect is not good and secondary processing is performed in the subsequent pathology recognition.

[0042] An image segmentation method determination module is used to extract visual features from the pathological image samples after image preprocessing, thereby determining an adapted image segmentation method for the pathological image samples;

[0043] As an achievable manner of the above solution, visual features are extracted from the pathological image samples after image preprocessing, specifically as follows: a background separation operation is performed on each pathological image sample to separate cells in the pathological image from the background.

[0044] The cell detection algorithm is used to identify cells, count the number of cells identified in the image, and calculate the number of cells per unit area, i.e., the cell density.

[0045] Each pathological image sample is converted into a color space to construct a color histogram of the pathological image, thereby calculating the staining uniformity.

[0046] In the possible implementation of the above scheme, color conversion is to convert the image from RGB color space to HSV or Lab color space in order to better separate color information. RGB color space is the most commonly used way to represent color, which is based on different combinations of the three primary colors of red, green and blue. RGB color space is very suitable for display devices (such as monitors, cameras, etc.), but in some image processing tasks, RGB color space cannot separate color information well.

[0047] The HSV color space divides colors into three separate components: Hue: Indicates the type of color, such as red, green, blue, etc. Saturation: Indicates the purity or vividness of the color. The higher the saturation, the purer the color. Lightness / Brightness: Indicates the brightness or lightness of the color. One advantage of the HSV color space is that it can separate the type, purity, and brightness of the color, which allows for more flexibility in adjusting and separating these attributes when processing images.

[0048] The Lab color space is a perception-based color space designed to mimic the way the human visual system perceives color. The Lab color space divides color into three separate components: Brightness: Indicates how light or dark the color is. A channel: Represents the color range from green to red. B channel: Represents the color range from blue to yellow. An important feature of the Lab color space is that its color components are more consistent with human visual perception, so it is very useful in tasks such as color matching and color conversion.

[0049] Converting an image from RGB color space to HSV or Lab color space can help us better separate color information.

[0050] In a further possible implementation of the above scheme, the specific construction process of constructing a color histogram after the color space conversion is completed is as follows: first, the dimension of the histogram is determined according to the color space used. For example, the HSV color space contains three components (H, S, V), and the Lab color space also contains three components (L, a, b). Then, the number of intervals for each component is determined. Generally, the number of intervals for each component can be selected according to actual conditions, such as 8, 16 or 256.

[0051] Then calculate the frequency of each color component by traversing each pixel in the image and counting the number of pixels that fall into each interval. Finally, use the statistical results to draw a histogram to show the color distribution.

[0052] In a further possible implementation of the above scheme, the coloring uniformity is calculated by calculating the variance of the color histogram, wherein the variance of the color histogram is a statistic used to measure the degree of discreteness of the color distribution in the image, and the variance reflects the degree of deviation between the color value and its average value. The larger the variance, the more dispersed the color distribution in the image; the smaller the variance, the more concentrated and uniform the color distribution is. The variance is a dimensional parameter, and when the variance is used to calculate the coloring uniformity, the variance needs to be dimensionless. Specifically, a variance threshold can be set, and the calculated variance is compared with the variance threshold. When the variance is less than the variance threshold, the coloring uniformity is 1. When the variance reaches the variance threshold, the exponential function is constructed with three quarters as the base, and the result of subtracting the variance from the variance threshold and dividing it by the variance threshold is used as the exponent to obtain the calculated value as the coloring uniformity.

[0053] A gray-level co-occurrence matrix is ​​constructed for each pathological image sample, and texture features are extracted from it to calculate the complexity of the tissue structure.

[0054] In the optimization implementation of the above scheme, the texture features are extracted from each pathological image sample to construct a grayscale co-occurrence matrix. The specific operations are as follows: each pathological image sample is grayed and normalized, and a grayscale co-occurrence matrix is ​​constructed according to the set distance and angle, from which texture features such as contrast, energy, and homogeneity are extracted. The contrast reflects the degree of difference in grayscale values ​​between adjacent pixels in the image. If the contrast is high, it means that the grayscale difference between adjacent pixels in the image is large, which usually indicates that the tissue structure is more complex and there are more edges and details; the energy reflects the uniformity of the grayscale value distribution in the image. If the energy is low, it means that the grayscale value distribution in the image is uneven, which usually indicates that the tissue structure is more complex and there are more changes; the homogeneity reflects the similarity between adjacent pixels in the image. If the homogeneity is low, it means that the similarity between adjacent pixels in the image is low, which usually indicates that the tissue structure is more complex and there are more changes.

[0055] Since the extracted contrast, energy, and homogeneity are dimensionless parameters, the contrast, energy, and homogeneity can be weighted averaged to obtain the complexity of the tissue structure. , where the values ​​of energy and homogeneity are usually greater than 0 in practice, so the above calculation formula is valid.

[0056] Cell density, staining uniformity and tissue structure complexity are all used as visual features of pathological image samples.

[0057] It should be understood that cell density, staining uniformity and tissue structure complexity are used as visual features of pathological image samples because these features can provide key information about pathological images and determine the difficulty of image segmentation and the applicable segmentation method.

[0058] As a further improvement of the above technical solution, the adaptive image segmentation method of the pathological image samples is determined as follows: the visual features of each pathological image sample are matched with the image segmentation control relationships corresponding to various visual features in a pre-constructed pairing library, and the adaptive image segmentation method corresponding to the cell density, staining uniformity and tissue structure complexity as the segmentation basis is obtained in each pathological image sample.

[0059] In the above improved example, the image segmentation comparison relationship corresponding to various visual features is as follows: for the visual feature of cell density, the distance between high-density cells is small, and cell overlap or aggregation is prone to occur. High-density images can use methods based on connected component analysis. The distance between low-density cells is large, and the cells are relatively isolated. Low-density images can use methods based on edge detection; for the visual feature of staining uniformity, when the staining in the image is relatively uniform and the color change is not large, a color threshold-based segmentation method can be used. When there are obvious color changes and uneven staining in the image, a more complex segmentation method needs to be used, such as a segmentation model based on deep learning; for the visual feature of tissue structure complexity, when the image contains multiple tissue types and has many structural levels, advanced segmentation methods, such as a segmentation model based on deep learning, are needed. When there are fewer tissue types in the image and the structural level is simple, a simpler segmentation method, such as a segmentation method based on edge detection, can be used.

[0060] The adaptive image segmentation methods corresponding to the segmentation basis based on cell density, staining uniformity and tissue structure complexity in each pathological image sample are classified according to the same segmentation basis to form a set of adaptive image segmentation methods based on cell density, staining uniformity and tissue structure complexity. Then, each segmentation method in the set of adaptive image segmentation methods is counted and its occurrence frequency is recorded to obtain the occurrence frequency of each adaptive image segmentation method in the set.

[0061] The adapted image segmentation methods in different sets (i.e., sets corresponding to different visual feature combinations) are compared, and the occurrence frequency of each adapted image segmentation method between sets is recorded.

[0062] The frequency of occurrence of each adaptive image segmentation method within the set and between sets is combined to calculate the frequency index of each adaptive image segmentation method. The specific calculation formula is: , there are only 3 sets formed, so the frequency of occurrence between sets is less than or equal to 3.

[0063] According to the frequency indexes of the adaptive image segmentation methods, the adaptive image segmentation method corresponding to the maximum frequency index is selected as the adaptive image segmentation method of the pathological image sample.

[0064] It should be understood that according to the frequency of occurrence of each adaptive image segmentation method within and between sets, the segmentation method corresponding to the maximum frequency index is selected as the adaptive image segmentation method for the pathological image sample. This selection method can find the most frequently occurring segmentation method, increase the stability of the selection results, and may have better robustness.

[0065] The image segmentation operation module is used to perform image segmentation on the pathological image sample according to the adaptive image segmentation method to obtain a main body image retained after segmentation.

[0066] The pathology recognition module is used to perform pathology recognition on the main image retained by the pathology image sample, and obtain pathology recognition effect indicators, specifically the recognition time and the degree of consistency between the recognition result and the expected biomarker, and evaluate the pathology recognition effect based on this.

[0067] In an optional implementation of the above scheme, the process of obtaining the pathology recognition effect indication is as follows: capture the timestamp of the first log record and the timestamp of the last log record from the pathology recognition log file of each pathology image sample, thereby obtaining the pathology recognition time of each pathology image sample, and take the average pathology recognition time as the pathology recognition time of the pathology image sample.

[0068] It is important to know that various errors or abnormalities may be encountered during pathology identification. Therefore, in order to conduct audit tracking, log files are generally generated. The log files record every step in the processing process, including the status of input data, the results of intermediate processing steps, the final output results, etc. This is very useful for auditing and reproducing the processing process.

[0069] Each pathological image sample is similarly predicted using the biomarkers corresponding to the historical pathological images to form an expected biomarker set for each pathological image sample.

[0070] It is important to know that biomarkers refer to indicators of molecular, cellular or functional changes that can reflect specific physiological or pathological states in an organism. Biomarkers can be used as the result of pathological identification, and by detecting the presence or level of biomarkers, doctors can be assisted in making a diagnosis. Exemplarily, biomarkers can be cancer markers, such as carcinoembryonic antigen, alpha-fetoprotein, prostate-specific antigen, etc.

[0071] In the specific implementation of the above scheme, each pathological image sample is predicted similarly using the biomarkers corresponding to the historical pathological images as follows: first, the historical pathological images and their corresponding biomarker data are collected; then, a similarity model is constructed between each pathological image sample and the historical pathological images to perform a similarity analysis, from which similar historical pathological images corresponding to each pathological image sample are screened out; then, the biomarkers corresponding to the similar historical pathological images of each pathological image sample are deduplicated and used as the biomarkers predicted for each pathological image sample.

[0072] It should be noted that in order to enhance the efficiency and reliability of biomarker prediction, it is necessary to ensure that the collected historical pathology images and the current pathology image samples are from the same tissue type, and the number of historical pathology images must be large enough.

[0073] The pathological recognition results of each pathological image sample are matched with the corresponding expected biomarker set, and the proportion of successfully matched pathological image samples is counted as the degree of consistency between the recognition results of the pathological image samples and the expected biomarkers.

[0074] The present invention uses the proportion of successfully matched pathological image samples as a quantitative indicator, which can intuitively reflect the degree of consistency between the recognition results and the expected biomarkers, and provides an intuitive evaluation standard that is easy to understand and explain.

[0075] In another specific implementation of the above scheme, the pathology recognition effect evaluation process is as follows: the pathology recognition effect indicator of the pathology image sample is substituted into the evaluation formula Obtain the pathology recognition effect coefficient , where , They represent the recognition time of pathological image samples and the consistency between the recognition results and the expected biomarkers, respectively. It represents the reference recognition time, which can be the average pathology recognition time of similar historical pathology images. Represents a natural constant. From this formula, it can be seen that the shorter the identification time, the greater the consistency between the identification result and the expected biomarker, and the greater the pathology identification effect coefficient.

[0076] The pathology recognition log analysis module is used to retrieve the log files generated by pathology recognition, extract the operation steps and timestamps corresponding to each log record, and perform pathology recognition image processing time ratio analysis.

[0077] The above-mentioned analysis of the proportion of pathology recognition image processing time is as follows: the timestamp of each log record in the pathology recognition log file of each pathology image sample is compared with the timestamp of the next log record to obtain the operation time of the corresponding operation step of each log record.

[0078] Identify whether the operation step corresponding to each log record is an image processing step, and then calculate the ratio of the operation time of the log record corresponding to the image processing step to the pathology recognition time of the corresponding pathology image sample to obtain the pathology recognition image processing time ratio of each pathology image sample.

[0079] The pairing library is used to store image segmentation comparison relationships corresponding to various visual features.

[0080] The preprocessing optimization demand judgment module is used to judge whether there is an image preprocessing optimization demand from the analysis of the pathology recognition image processing time ratio of the pathology image samples. If so, the optimization direction is output. The specific implementation refers to the following process: the pathology recognition image processing time ratio of each pathology image sample is compared with the set critical ratio. When the pathology recognition image processing time ratio of a pathology image sample reaches the critical ratio, it means that the pathology image sample spends a lot of time on image processing during the pathology recognition process. The number of pathology image samples that reach the critical ratio is screened out, and compared with the selected number of pathology image samples to obtain the ratio of samples with long processing time.

[0081] It should be explained that the above scheme sets the critical ratio as a quantitative indicator to make the evaluation process more objective and measurable, and the critical ratio should not be set too small or too large. When the setting is too small, even normal image processing steps may be marked as time-consuming, resulting in an increase in false alarms. When the setting is too large, even pathological image samples that take a long time may not be marked, which will cause some actual performance bottlenecks to be ignored and cannot be discovered in time. Therefore, in order to ensure that the critical ratio is set reasonably, we can analyze historical data to find the average time ratio of image processing steps in the pathology recognition process, and then use the average time ratio as the basis for setting. For example, assuming that we analyze historical data and find that the average time ratio of image processing steps in the pathology recognition process is approximately 0.5, then we can set the critical ratio to 0.6.

[0082] The ratio of samples with long processing time is compared with a preset ratio threshold. For example, the ratio threshold is 0.8. If the ratio threshold is reached, it means that most of the selected pathological image samples spend a lot of time on image processing during the pathology identification process. In this case, it is judged that there is a need for image preprocessing optimization. Otherwise, it is judged that there is no need for image preprocessing optimization.

[0083] The pathological image samples that reach the critical proportion are recorded as reference pathological image samples, and then the operation time corresponding to the log record of the image processing step in the reference pathological image sample and the pathology recognition time of the corresponding pathological image sample are proportionally calculated to obtain the time proportion corresponding to each image processing step in pathology recognition of the reference pathological image sample, and the image processing step corresponding to the longest time proportion is extracted as the optimized image processing step of the reference pathological image.

[0084] In the example of the above scheme, it is assumed that the image processing steps of the reference pathology image sample in pathology identification include image enhancement and image denoising, where the duration corresponding to image enhancement accounts for 0.3, and the duration corresponding to image denoising accounts for 0.38. At this time, image denoising is used as the optimized image processing step of the reference pathology image.

[0085] The optimized image processing steps of each reference pathological image are compared, and the occurrence ratio values ​​of the optimized image processing steps are summarized and then compared with the effective ratio value. Exemplarily, the effective ratio value is 0.8, and the optimized image processing steps that reach the effective ratio value are extracted as the optimization direction in image preprocessing.

[0086] The image segmentation replacement demand judgment module is used to judge whether there is a need to change the image segmentation method based on the pathology recognition effect of the pathology image sample. If there is a replacement demand, the replacement image segmentation method is determined. The specific implementation is as follows: the pathology recognition effect coefficient of the pathology image sample is compared with the pre-configured standard effect coefficient. Exemplarily, the standard effect coefficient is 80%. If the pathology recognition effect coefficient does not reach the standard effect coefficient, it is judged that there is a need to change the image segmentation method. Otherwise, it is judged that there is no need to change the image segmentation method.

[0087] When it is judged that there is a need to replace the image segmentation method, the adaptive image segmentation method with the second largest frequency index is extracted from the frequency indexes of the adaptive image segmentation methods as the replacement image segmentation method.

[0088] The preprocessing optimization module is used to perform image preprocessing on the remaining pathological images according to the image preprocessing optimization orientation, and is specifically implemented to strengthen the processing of the image preprocessing optimization orientation in the image preprocessing of the remaining pathological images.

[0089] The image segmentation and recognition optimization module is used to perform image segmentation and pathology recognition on the remaining pathological images after image preprocessing using a changed image segmentation method.

[0090] In the innovative implementation of the above scheme, when it is judged that there is a need for image preprocessing optimization and a need to change the image segmentation method, the remaining pathology images can be subjected to secondary sample selection, and then image preprocessing optimization and image segmentation method change can be performed in the secondary pathology image samples. After the segmentation is completed, pathology recognition can be performed and the pathology recognition effect can be evaluated. Finally, based on the pathology recognition effect, the need to change the image segmentation method can be re-evaluated, thereby achieving dynamic optimization of pathology image segmentation and further improving the accuracy and efficiency of pathology recognition.

[0091] Example 2

[0092] Reference Figure 2 As shown, the present invention proposes a pathological image analysis method based on image processing technology, comprising the following steps: S1, selecting samples of batch pathological images of the same tissue scanned by the same microscope scanner under the same imaging conditions, and performing image preprocessing.

[0093] S2. Extracting visual features from the pathological image samples after image preprocessing, thereby determining an adapted image segmentation method for the pathological image samples.

[0094] S3. Perform image segmentation on the pathological image sample according to the adapted image segmentation method to obtain a main body image retained after segmentation.

[0095] S4. Perform pathological recognition on the main image retained by the pathological image sample, and obtain indicators of the pathological recognition effect, specifically the recognition time and the degree of consistency between the recognition result and the expected biomarker, and evaluate the pathological recognition effect based on this.

[0096] S5. Retrieve the log file generated by pathology recognition, extract the operation steps and timestamps corresponding to each log record, and perform pathology recognition image processing time ratio analysis.

[0097] S6. From the analysis of the proportion of pathology recognition image processing time of pathology image samples, determine whether there is a need for image preprocessing optimization, and if so, output the optimization direction.

[0098] S7. Based on the pathology recognition effect of the pathology image sample, it is judged whether there is a need to change the image segmentation method. If there is a need to change, the image segmentation method to be replaced is determined.

[0099] S8. Perform image preprocessing on the remaining pathological images according to the image preprocessing optimization orientation.

[0100] S9. Use the changed image segmentation method to perform image segmentation and pathology recognition on the remaining pathological images after image preprocessing.

[0101] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A pathological image analysis system based on image processing technology, characterized in that: Includes the following modules: A pathology image sample selection module is used to select samples of batch pathology images of the same tissue scanned by the same microscope scanner under the same imaging conditions, and perform image preprocessing; An image segmentation method determination module is used to extract visual features from the pathological image samples after image preprocessing, thereby determining an adapted image segmentation method for the pathological image samples; Pairing library, used to store image segmentation comparison relationships corresponding to various visual features; Match the visual features of each pathological image sample with the image segmentation control relationships corresponding to various visual features in the pre-built pairing library, and obtain the adaptive image segmentation method corresponding to the cell density, staining uniformity and tissue structure complexity as the segmentation basis in each pathological image sample; An image segmentation operation module is used to perform image segmentation on the pathological image sample according to the adapted image segmentation method to obtain a main body image retained after segmentation; The pathology recognition module is used to perform pathology recognition on the main images retained by the pathology image samples and obtain the indicators of the pathology recognition effect, specifically the recognition time, recall rate, and the consistency between the recognition results and the expected biomarkers, and evaluate the pathology recognition effect based on this; The pathology recognition log analysis module is used to retrieve the log files generated by pathology recognition, extract the operation steps and timestamps corresponding to each log record, and perform pathology recognition image processing time ratio analysis; A preprocessing optimization requirement evaluation module is used to determine whether there is an image preprocessing optimization requirement from the analysis of the pathology recognition image processing time ratio of the pathology image samples, and output the optimization direction if so; An image segmentation replacement demand judging module is used to judge whether there is a need to replace the image segmentation method based on the pathology recognition effect of the pathology image sample, and if there is a need to replace, determine the image segmentation method to be replaced; A preprocessing optimization module, used for performing image preprocessing on the remaining pathological images according to the image preprocessing optimization orientation; The image segmentation and recognition optimization module is used to perform image segmentation and pathology recognition on the remaining pathological images after image preprocessing by using a changed image segmentation method.

2. A pathological image analysis system based on image processing technology as claimed in claim 1, characterized in that: The sample selection process is as follows: Count the number of batch pathological images and number each pathological image; Determine the number of samples from the number of batch pathology images according to the sample selection ratio; The interval numbers in the batch pathology images are determined according to the number of samples, and samples are selected based on this.

3. A pathological image analysis system based on image processing technology as claimed in claim 1, characterized in that: The process of extracting visual features is as follows: Perform background separation on each pathological image sample, use cell detection algorithm to identify cells, and count the number of cells identified in the image to calculate cell density; Performing color space conversion on each pathological image sample to construct a color histogram of the pathological image, thereby calculating the staining uniformity; A gray-level co-occurrence matrix is ​​constructed for each pathological image sample, and texture features are extracted from it to calculate the complexity of tissue structure. Cell density, staining uniformity and tissue structure complexity are all used as visual features of pathological image samples.

4. A pathological image analysis system based on image processing technology as claimed in claim 3, characterized in that: The process of determining the appropriate image segmentation method for pathological image samples is as follows: The adaptive image segmentation methods corresponding to the segmentation basis of cell density, staining uniformity and tissue structure complexity in each pathological image sample are classified according to the same segmentation basis to form an adaptive image segmentation method set based on cell density, staining uniformity and tissue structure complexity as the segmentation basis, and then each segmentation method in the adaptive image segmentation method set is counted and its occurrence frequency is recorded to obtain the occurrence frequency of each adaptive image segmentation method in the set; Compare the adapted image segmentation methods in different sets, and record the occurrence frequency of each adapted image segmentation method between sets; Combining the occurrence frequencies of each adaptive image segmentation method within a set and between sets, and counting the frequency index of each adaptive image segmentation method; According to the frequency indexes of the various adaptive image segmentation methods, the adaptive image segmentation method corresponding to the maximum frequency index is selected as the adaptive image segmentation method of the pathological image sample.

5. A pathological image analysis system based on image processing technology as claimed in claim 1, characterized in that: The process of obtaining the pathology recognition effect indication is as follows: Capturing the timestamp of the first log record and the timestamp of the last log record from the pathology recognition log file of each pathology image sample, thereby obtaining the pathology recognition duration of each pathology image sample, and taking the average pathology recognition duration as the pathology recognition duration of the pathology image sample; Performing similarity prediction on each pathological image sample using the biomarkers corresponding to the historical pathological images to form an expected biomarker set for each pathological image sample; The pathological recognition results of each pathological image sample are matched with the corresponding expected biomarker set, and the proportion of successfully matched pathological image samples is counted as the degree of consistency between the recognition results of the pathological image sample and the expected biomarkers.

6. A pathological image analysis system based on image processing technology as claimed in claim 1, characterized in that: The pathology recognition effect evaluation process is as follows: Substitute the pathology recognition effect indicators of pathology image samples into the evaluation formula Obtain the pathology recognition effect coefficient , where , They represent the recognition time of pathological image samples and the consistency between the recognition results and the expected biomarkers, respectively. Indicates the reference recognition time. Represents a natural constant.

7. A pathological image analysis system based on image processing technology as claimed in claim 1, characterized in that: The analysis process of the pathology recognition image processing time ratio is as follows: Compare the timestamp of each log record in the pathology recognition log file of each pathology image sample with the timestamp of the next log record to obtain the operation duration of the operation step corresponding to each log record; Identify whether the operation step corresponding to each log record is an image processing step, and then calculate the proportion of the pathology recognition image processing time of each pathology image sample by adding up the operation time of the log records corresponding to the image processing step and calculating the proportion of the pathology recognition image processing time of each pathology image sample.

8. A pathological image analysis system based on image processing technology as claimed in claim 7, characterized in that: The process of judging whether there is a need for image preprocessing optimization and outputting the optimization direction if there is a need is as follows: The proportion of the pathology recognition image processing time of each pathology image sample is compared with the set critical proportion, and the number of pathology image samples reaching the critical proportion is screened out, and compared with the number of selected pathology image samples to obtain the ratio of samples with long processing time; The ratio of samples with long processing time is compared with a preset ratio threshold. If the ratio threshold is reached, it is judged that there is a need for image preprocessing optimization; otherwise, it is judged that there is no need for image preprocessing optimization; The pathological image samples reaching the critical proportion are recorded as reference pathological image samples, and then the operation time of the log record corresponding to the image processing step in the reference pathological image samples is proportionally calculated to the pathological recognition time of the corresponding pathological image samples, so as to obtain the time proportion corresponding to each image processing step in pathological recognition of the reference pathological image samples, and the image processing step corresponding to the longest time proportion is extracted as the optimized image processing step of the reference pathological image; The optimized image processing steps of each reference pathological image are compared, and the occurrence ratio values ​​of the optimized image processing steps are summarized and then compared with the effective ratio values, so as to extract the optimized image processing steps that reach the effective ratio value as the optimization direction in image preprocessing.

9. A pathological image analysis system based on image processing technology as claimed in claim 4, characterized in that: The image segmentation method for determining the replacement is implemented as follows: The adaptive image segmentation method with the second largest frequency index is extracted from the frequency indexes of the adaptive image segmentation methods as the replacement image segmentation method.

10. A pathological image analysis method based on image processing technology, characterized in that: The following steps are involved: S1, selecting samples of batch pathological images of the same tissue scanned by the same microscope scanner under the same imaging conditions, and performing image preprocessing; S2, extracting visual features from the pathological image samples after image preprocessing, thereby determining an adapted image segmentation method for the pathological image samples; Match the visual features of each pathological image sample with the image segmentation control relationships corresponding to various visual features in the pre-built pairing library, and obtain the adaptive image segmentation method corresponding to the cell density, staining uniformity and tissue structure complexity as the segmentation basis in each pathological image sample; S3, performing image segmentation on the pathological image sample according to the adapted image segmentation method to obtain a main image retained after segmentation; S4. Perform pathology recognition on the main images retained by the pathology image samples, and obtain indicators of pathology recognition effect, specifically, recognition time, recall rate, and consistency between the recognition results and the expected biomarkers, and evaluate the pathology recognition effect based on this; S5, retrieve the log file generated by pathology recognition, extract the operation steps and timestamps corresponding to each log record, and perform pathology recognition image processing time ratio analysis; S6. Judging whether there is a need for image preprocessing optimization from the analysis of the proportion of pathology recognition image processing time of the pathology image samples, and outputting optimization directions if so; S7. judging whether there is a need to change the image segmentation method based on the pathology recognition effect of the pathology image sample, and determining a replacement image segmentation method if there is a need to change; S8, performing image preprocessing on the remaining pathological images according to the image preprocessing optimization orientation; S9. Use the changed image segmentation method to perform image segmentation and pathology recognition on the remaining pathological images after image preprocessing.

Images