Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, in an embodiment, a pathological section quality evaluation method based on machine learning is provided, and the pathological section quality evaluation method based on machine learning may be applied to both a terminal and a server, and this embodiment is exemplified by being applied to a server. The pathological section quality evaluation method based on machine learning specifically comprises the following steps:
and 102, acquiring a pathological section image, and digitally scanning the pathological section image to generate a digital pathological section image.
The pathological section image is an image obtained by non-invasive methods such as HE staining, Papanicolaou staining, special staining, immunohistochemistry, immunofluorescence or electron microscopy and the like on a human body or internal tissues of a part of the human body, and is used for carrying out pathological diagnosis and determining the benign and malignant degree, grouping the types, the malignant degree, judging prognosis, guiding clinical treatment and the like of diseases by combining clinical data. The digital pathological section image is a pathological section image in which a pathological section is displayed as a digital image and stored. Specifically, the digital pathology scanner may be used to digitally scan a pathology section image of a biopsy sample. For subsequent further processing based on the digital pathology slice image.
And 104, identifying the problem area of the digital pathological section image to obtain the pathological section image to be evaluated.
The problem area refers to a partial image of the digital pathological section, the quality of which does not meet a preset condition, for example, a cell nucleus and a serous stain of the pathological section corresponding to the digital pathological section image are not stained, or a problem area of the pathological section corresponding to the digital pathological section image, the thickness of which does not meet the condition, and the like. It can be understood that, when the digital pathological section image is used for disease diagnosis, if the digital pathological section image has a problem area, that is, the quality of the digital pathological section does not satisfy the preset condition, the pathological diagnosis is affected, which results in missed diagnosis and misdiagnosis. Specifically, the problem area may be determined according to the pixel size of the digital pathological section image and a preset pixel value, and the pathological section image to be evaluated may be obtained.
And 106, classifying the quality of the pathological section image to be evaluated through a preset quality classifier based on machine learning to obtain a quality category.
The quality category is used as index data for reflecting the quality problem type of the pathological section, and in one embodiment, the problem category includes but is not limited to cell nucleus, unclear serous staining, incomplete tissue block cutting, small cavity, uneven staining, thick section, knife shaking, folded section, collapsed section, too thick section, and the like. Machine learning is a method based on feature learning of data. The observations (e.g., images) can be represented using a variety of ways, such as a vector of intensity values for each pixel, or more abstractly as a series of edges, a region of a particular shape, and so forth. And tasks such as face recognition are easier to learn from instances using some specific representation methods. The machine learning is used for establishing and simulating a neural network for analyzing and learning the human brain, the neural network simulates the mechanism of the human brain to explain data such as images, sounds and texts, and unsupervised or semi-supervised feature learning and layered feature extraction efficient algorithms are used for replacing manually acquired features, so that the objectivity and the accuracy of a prediction result can be improved. The quality classifier of the present embodiment is used for classifying different pathological section images to be evaluated into one of a plurality of target quality categories (such as quality categories of uneven staining, thick and thick slices, jolting, creasing and collapsing slices, too thick slices, etc.). In particular, a classifier that can be classified using at least one machine learning model. The machine learning model may be one or more of the following: neural networks (e.g., convolutional neural networks, BP neural networks, etc.), logistic regression models, support vector machines, decision trees, random forests, perceptrons, and other machine learning models. As part of the training of such machine learning models, the training input is images corresponding to various effective pathological areas, and through training, a corresponding relationship classifier of pathological section images to be evaluated and various quality categories is established. The quality classifier is enabled to have the capability of judging which one of the quality classes corresponds to the input pathological section image to be evaluated. In this embodiment, the quality classifier is a multi-classifier, and a plurality of classification results are obtained, that is, the quality of the pathological section corresponding to the pathological section image is evaluated, and understandably, the quality of the pathological section corresponding to the pathological section image is evaluated in a machine learning manner, so that the automatic evaluation of the quality of the pathological section is realized, and the accuracy and objectivity of the quality evaluation of the pathological section are improved.
According to the pathological section quality evaluation method based on machine learning, the pathological section image is acquired and digitally scanned to generate a digital pathological section image; identifying the problem area of the digital pathological section image to obtain a pathological section image to be evaluated; the quality of the pathological section image to be evaluated is classified through a preset quality classifier based on machine learning to obtain a quality category, so that the automatic evaluation of the quality of the pathological section is realized, and the accuracy and the objectivity of the quality evaluation of the pathological section are improved.
As shown in fig. 2, in an embodiment, identifying a problem area of a digitized pathological section to obtain an image of the pathological section to be evaluated includes:
104A, performing binarization processing on the digital pathological section image to obtain a gray level image;
and step 104B, extracting a problem area with the gray value not meeting a preset gray threshold value from the gray image as a pathological section image to be evaluated.
In this embodiment, the problem area is determined according to the pixels of the digital pathological section image, and it can be understood that each preset condition corresponds to one preset pixel value, and the problem area can be identified by comparing the digital pathological section image with each preset pixel value. Specifically, the digital pathological section image is subjected to binarization processing, wherein the binarization processing includes, but is not limited to, global binarization, an optimal threshold method based on a histogram, or an OTSU extra large threshold method based on clustering, that is, color values of a color image are converted into a gray image, and an area where gray values do not meet a preset gray threshold is extracted from the gray image.
As shown in fig. 3, in an embodiment, before identifying the problem area of the digitized pathological section and obtaining the pathological section image to be evaluated, the method further includes:
step 108, sending the identified problem area image to an expert auditing end for auditing;
and step 110, determining the problem area image fed back after the examination as a pathological section image to be evaluated.
The expert auditing end refers to a user end for manual auditing by a professional doctor. Specifically, the server sends the problem area image identified in step 104 to a secondary manual review, then determines the problem area image fed back after the review as a pathological section image to be evaluated, and further improves the accuracy of identification of the pathological section image to be evaluated by adopting a manual review mode, thereby being beneficial to improving the accuracy of evaluation of the pathological section corresponding to the pathological section image to be evaluated.
As shown in fig. 4, in an embodiment, after obtaining the quality category, the method further includes:
step 112, acquiring a target quality category of the pathological section image to be evaluated, and determining a target improvement measure corresponding to the target quality category from a corresponding relation table of a preset quality category and the improvement measure;
and step 114, obtaining a target pathological section according to the pathological section corresponding to the pathological section image to be evaluated by the target improvement measure.
The preset quality category and improvement measure correspondence table refers to a mapping relationship table of each quality category and improvement measure stored in advance by the server, that is, each quality category corresponds to an improvement measure, for example, a cell nucleus and a quality category with unclear pulp staining, and the corresponding improvement measure is to replace dehydration liquid or prolong dehydration time; the tissue mass is not completely cut, the quality category of small cavities is correspondingly improved by improving a slicing method and finely trimming wax masses; the quality category of uneven dyeing is improved by replacing dehydration liquid or prolonging dehydration time and replacing dyeing liquid; the quality types of the thick slices and the vibration knife are improved correspondingly by improving the tissue dehydration and clamping wax blocks; the quality category of slice folding is improved correspondingly by fully unfolding the slice and raising the water temperature; quality categories of slice folding and collapse are improved correspondingly to avoid bubble generation; and the quality category of the slice which is too thick is correspondingly improved by adjusting the corresponding relation of the thickness of the slice and the like. Understandably, the target pathological section is obtained by the pathological section corresponding to the pathological section image to be evaluated according to the target improvement measure, so that the quality of the target pathological section is improved, and the pathological diagnosis effect is favorably improved.
As shown in fig. 5, in one embodiment, the method further comprises:
step 116, obtaining a training sample set, wherein the training sample set comprises a plurality of training problem area images and corresponding training quality categories;
and step 118, taking the image of the training problem area as the input of a preset classifier, taking the training quality category as the expected output, and training the preset classifier to obtain the trained quality classifier.
Specifically, a sample set of quality categories corresponding to problem areas marked by a doctor is acquired, training problem area images are used as input of a preset classifier, training quality categories are used as expected output, the preset classifier is trained, training quality categories corresponding to the training problem area images in the training sample set can be generated, and therefore the preset classifier is trained according to the expected output corresponding to the current training problem area images, and the trained quality classifier is obtained.
In the embodiment, the training sample set comprises a plurality of quality categories, the comprehensiveness of the training sample set is guaranteed, the quality categories trained by the training sample set can learn more comprehensive and accurate quality category classification rules, the efficiency of learning the preset classifier by the training machine is improved, and therefore the efficiency of evaluating the quality of the pathological section corresponding to the problem area image can be further improved.
As shown in fig. 6, in an embodiment, taking a training problem area image as an input of a preset classifier, taking a training quality category as an expected output, and training the preset classifier to obtain a trained quality classifier includes:
step 118A, obtaining a test sample set, wherein the test sample set comprises a test problem area image and a corresponding test quality category;
step 118B, inputting the image of the test problem area into a preset classifier;
step 118C, acquiring the output verification quality category, acquiring an error between the verification quality category and the quality pathology category, and determining that the training of a preset classifier is finished to obtain a quality classifier under the condition that the error is smaller than a preset error; or acquiring the training times corresponding to the preset classifier, and determining that the preset classifier is completely trained to obtain the quality classifier under the condition that the training times reach the maximum preset times.
Specifically, a test sample set is obtained, the test sample comprises a test problem area image and a corresponding test quality category, the test sample set is predicted by a trained machine learning classifier, a known classification result, namely an expected quality category, of the test sample set is obtained, the predicted result is compared with the known classification result to obtain the classification prediction accuracy of the corresponding machine learning classifier, the error between the verification quality category and the expected quality category is obtained, under the condition that the error is smaller than the preset error, the error can be used as a judgment index through the size of sensitivity and specificity, for example, when the sensitivity is greater than 99% or the specificity is greater than 95%, the preset classifier is determined to be completely trained, or the training frequency corresponding to the preset classifier is obtained, under the condition that the training frequency reaches the maximum preset frequency, and determining that the preset classifier is trained, acquiring a parameter value used by the machine learning classifier, and otherwise, continuously training the machine learning classifier by using the acquired parameter value and the test sample set.
In this embodiment, the parameter value is roughly positioned by using the test sample set, and the most appropriate parameter value can be found as far as possible by obtaining the error between the verified quality class and the expected quality class or the number of times of training, so that the training is performed by using the parameter value and the test sample set, and the trained machine learning classifier can distinguish the quality classes to achieve higher accuracy.
As shown in fig. 7, in an embodiment, after inputting the test question area image into a preset classifier and obtaining the output verification quality category, the method further includes:
step 120, sending the verification quality category to an expert auditing end for auditing;
and step 122, judging that the preset classifier is trained completely under the condition that the checking result is passed.
In the embodiment, the server sends the verification quality category to the expert auditing end for secondary manual auditing, and judges that the preset classifier is completely trained under the condition that the auditing result is passed, so that the accuracy of the quality classifier is further ensured in a manual rechecking mode.
As shown in fig. 8, in one embodiment, a system for machine learning-based pathological section quality evaluation is provided, the system comprising:
a scanning module 802, configured to acquire a pathological section image, perform digital scanning on the pathological section image, and generate a digital pathological section image;
the identification module 804 is used for identifying the problem area of the digital pathological section image to obtain a pathological section image to be evaluated;
and the evaluation module 806 is configured to classify the quality of the pathological section image to be evaluated by using a preset quality classifier based on machine learning, so as to obtain a quality category.
FIG. 9 is a diagram illustrating an internal structure of a computer device in one embodiment. The computer device may specifically be a server including, but not limited to, a high performance computer and a cluster of high performance computers. As shown in fig. 9, the computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program which, when executed by the processor, may cause the processor to implement a machine learning-based pathological section quality evaluation method. The internal memory may also store a computer program that, when executed by the processor, causes the processor to execute a method for machine learning-based pathological section quality assessment. Those skilled in the art will appreciate that the architecture shown in fig. 9 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, the method for evaluating the quality of a pathological section based on machine learning provided by the present application may be implemented in the form of a computer program, and the computer program may be executed on a computer device as shown in fig. 9. The memory of the computer device may store therein various program templates constituting the machine learning-based pathological section quality evaluation system. Such as a scanning module 802, an identification module 804, and an evaluation module 806.
A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the following steps when executing the computer program: acquiring a pathological section image, and carrying out digital scanning on the pathological section image to generate a digital pathological section image; identifying the problem area of the digital pathological section image to obtain a pathological section image to be evaluated; and classifying the quality of the pathological section image to be evaluated through a preset quality classifier based on machine learning to obtain a quality category.
In one embodiment, identifying the problem area of the digitized pathological section to obtain the pathological section image to be evaluated includes: carrying out binarization processing on the digital pathological section image to obtain a gray level image; and extracting a problem area with the gray value not meeting a preset gray threshold value from the gray image to be used as a pathological section image to be evaluated.
In one embodiment, before the identifying the problem area of the digitized pathological section to obtain the pathological section image to be evaluated, the method further includes: sending the identified problem area image to an expert auditing end for auditing; and determining the problem area image fed back after the examination as the pathological section image to be evaluated.
In one embodiment, after the obtaining the quality category, the method further includes: acquiring a target quality category of the pathological section image to be evaluated, and determining a target improvement measure corresponding to the target quality category from a corresponding relation table of a preset quality category and an improvement measure; and obtaining a target pathological section for the pathological section corresponding to the pathological section image to be evaluated according to the target improvement measure.
In one embodiment, the method further comprises: acquiring a training sample set, wherein the training sample set comprises a plurality of training problem area images and corresponding training quality categories; and taking the training problem area image as the input of a preset classifier, taking the training quality category as the expected output, and training the preset classifier to obtain the trained quality classifier.
In one embodiment, the training problem area image is used as an input of a preset classifier, the training quality category is used as an expected output, and the training of the preset classifier is performed to obtain the trained quality classifier, which includes: obtaining a test sample set, wherein the test sample set comprises a test problem area image and a corresponding test quality category; inputting the test problem area image into a preset classifier; acquiring an output verification quality category, acquiring an error between the verification quality category and a quality pathology category, and determining that the preset classifier is completely trained to obtain the quality classifier under the condition that the error is smaller than a preset error; or acquiring the training times corresponding to the preset classifier, and determining that the preset classifier is completely trained to obtain the quality classifier when the training times reach the maximum preset times.
In one embodiment, after the inputting the test problem area image into a preset classifier and acquiring the output verification quality category, the method further includes: sending the verification quality category to an expert auditing end for auditing; and judging that the preset classifier is trained completely under the condition that the auditing result is passed.
A computer-readable storage medium storing a computer program, the computer program when executed by a processor implementing the steps of: acquiring a pathological section image, and carrying out digital scanning on the pathological section image to generate a digital pathological section image; identifying the problem area of the digital pathological section image to obtain a pathological section image to be evaluated; and classifying the quality of the pathological section image to be evaluated through a preset quality classifier based on machine learning to obtain a quality category.
In one embodiment, identifying the problem area of the digitized pathological section to obtain the pathological section image to be evaluated includes: carrying out binarization processing on the digital pathological section image to obtain a gray level image; and extracting a problem area with the gray value not meeting a preset gray threshold value from the gray image to be used as a pathological section image to be evaluated.
In one embodiment, before the identifying the problem area of the digitized pathological section to obtain the pathological section image to be evaluated, the method further includes: sending the identified problem area image to an expert auditing end for auditing; and determining the problem area image fed back after the examination as the pathological section image to be evaluated.
In one embodiment, after the obtaining the quality category, the method further includes: acquiring a target quality category of the pathological section image to be evaluated, and determining a target improvement measure corresponding to the target quality category from a corresponding relation table of a preset quality category and an improvement measure; and obtaining a target pathological section for the pathological section corresponding to the pathological section image to be evaluated according to the target improvement measure.
In one embodiment, the method further comprises: acquiring a training sample set, wherein the training sample set comprises a plurality of training problem area images and corresponding training quality categories; and taking the training problem area image as the input of a preset classifier, taking the training quality category as the expected output, and training the preset classifier to obtain the trained quality classifier.
In one embodiment, the training problem area image is used as an input of a preset classifier, the training quality category is used as an expected output, and the training of the preset classifier is performed to obtain the trained quality classifier, which includes: obtaining a test sample set, wherein the test sample set comprises a test problem area image and a corresponding test quality category; inputting the test problem area image into a preset classifier; acquiring an output verification quality category, acquiring an error between the verification quality category and a quality pathology category, and determining that the preset classifier is completely trained to obtain the quality classifier under the condition that the error is smaller than a preset error; or acquiring the training times corresponding to the preset classifier, and determining that the preset classifier is completely trained to obtain the quality classifier when the training times reach the maximum preset times.
In one embodiment, after the inputting the test problem area image into a preset classifier and acquiring the output verification quality category, the method further includes: sending the verification quality category to an expert auditing end for auditing; and judging that the preset classifier is trained completely under the condition that the auditing result is passed.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.