TWI836280B - Medical image analysis method and device - Google Patents

Abstract

A medical image analysis method includes: reading an original medical image; performing image classification and objection detection on the original medical image to generate a first classification result and a plurality of detection results by a plurality of complementary AI models; performing object feature integration and quantification transformation on first and second detection results among the plurality of detection results to generate a quantification result by a feature integration and transformation module; and performing machine learning on the first classification result and the quantification result to generate and display an image determination result by a machine learning module.

Description

Medical image analysis methods and devices

The present invention relates to a medical image analysis method and device.

In order to improve the efficiency of doctors' image interpretation and reduce human errors, using artificial intelligence (AI) to assist doctors in image interpretation has become one of the key projects in the development of smart medicine in the medical field.

AI image-assisted diagnosis mainly uses three technologies: object detection (Object Detection), segmentation (Segmentation) and classification (Classification). In terms of classification technology, it can distinguish benign and malignant tumors and classify the severity of diseases in a relatively simple manner, allowing for effective allocation of medical resources. Under normal circumstances, classifiers (classifiers) such as: Deep Neural Networks (DNN), support vector machines (SVM, a type of machine learning), random forests, etc. are usually used. Perform lesion severity classification on medical images.

However, the classification, symptoms, and physiological tissues of diseases are usually correlated with each other, and the correlation method, degree of correlation, etc. are difficult for even experienced doctors to give specific clear-cut judgment criteria, and are not easy to obtain by analytical methods. This type of problem is very suitable for applying AI (artificial intelligence) technology to solve it through learning labeled data, but past experience has found that it is difficult to achieve the best accuracy if only a single AI classification model or AI detection model of symptom objects is used. If the classification model can be used to assist the shortcomings of the detection model (and vice versa, that is, the functions of these models can complement each other, it will definitely improve the shortcomings of using a single model and thus improve the misjudgment rate.

In addition, if the amount of training data is sparse, or because in the early stages of the disease, the symptoms are usually very small, if you use a classifier solely to classify medical images, it may not be possible because the variation area between normal images and abnormal images is extremely small. Effective identification of these symptoms leads to incorrect disease classification results.

According to an embodiment of this case, a medical image analysis method is proposed, including: reading an original medical image; using a plurality of complementary artificial intelligence models to perform image classification and object detection on the original medical image to obtain a first classification result and a plurality of object detection results; a feature integration conversion module performs integration and quantitative conversion of object features on a first detection result and a second detection result among the object detection results to obtain a quantification result; and a machine learning module performs machine learning on the quantification result and the first classification result to obtain and display an image interpretation result.

According to another embodiment of the present invention, a medical image analysis device is provided, comprising a processor; and a display unit coupled to the processor. The processor is configured to: read an original medical image; use a plurality of complementary artificial intelligence models to perform image classification and object detection on the original medical image to obtain a first classification result and a plurality of object detection results; use a feature integration conversion module to integrate and quantify object features of a first detection result and a second detection result among the object detection results to obtain a quantitative result; and use a machine learning module to perform machine learning on the quantitative result and the first classification result to obtain an image interpretation result, which is displayed on the display unit.

In order to better understand the above and other aspects of the present invention, the following embodiments are specifically described in detail with reference to the accompanying drawings:

The technical terms in this specification refer to the idioms in the technical field. If there are explanations or definitions for some terms in this specification, the explanation or definition of this part of the terms shall prevail. Each embodiment of the present disclosure has one or more technical features. Under the premise that implementation is possible, a person with ordinary skill in the art can selectively implement some or all of the technical features in any embodiment, or selectively combine some or all of the technical features in these embodiments.

Currently, diabetic macular edema (DME) is defined as the presence of any hard exudates (HE) within one disc diameter from the center of the macula. Conversely, the absence of any hard exudates within one disc diameter from the center of the macula is considered non-referable diabetic macular edema.

In addition, computed tomography (CT) is used before surgery to assess the resectability of pancreatic cancer tumors, which can be divided into five grades, from Grade 0 to Grade 4. The grade is related to the degree to which the tumor contacts the blood vessels. The greater the degree to which the tumor contacts the blood vessels, the greater the grade.

As shown in Figure 1, in the original medical image 100, a circle is virtualized with the macular center 110 as the center and the diameter R of the optic disc 111 as the radius. Hard leakage HE appears in this virtual circle. The condition is considered to have diabetic macular edema.

FIG. 2 shows a functional schematic diagram of a medical image analysis device 200 according to an embodiment of the present invention. The medical image analysis device 200 includes a processor 210, a database 220, and a display unit 230. The processor 210 is coupled to the database 220 and the display unit 230. The processor 210 reads the original medical image from the database 220, performs analysis and image interpretation, and then displays the image interpretation result on the display unit 230. The processor 210 is, for example, a central processing unit (CPU), or other programmable general-purpose or special-purpose micro control unit (MCU), microprocessor, digital signal processor (DSP), programmable controller, application specific integrated circuit (ASIC), graphics processing unit (GPU), arithmetic logic unit (ALU), complex programmable logic device (CPLD), field programmable gate array (FPGA) or other similar components or a combination of the above components. The display unit 230 is, for example but not limited to, a device with display function such as a liquid crystal display (LCD).

Figure 3 shows a schematic diagram of medical image analysis according to an embodiment of this case. As shown in Figure 3, the original medical image RMI is input to artificial intelligence (AI) models for a variety of different tasks, such as, but not limited to, classification model 310, disease symptom detection model 320, and physiological tissue detection model 330.

The classification model 310 is a disease severity classification (class) model used to analyze the entire original medical image RMI to obtain the disease classification result 340.

The symptom detection model 320 is a disease-related symptom detection model. The original medical image RMI is analyzed to obtain a symptom detection result 350. The symptom detection result 350 includes the location, area, confidence value and number of each symptom type of each symptom. Optionally, the area of each symptom can be obtained by calculating the vertical length and horizontal length of the bounding box of the symptom selected by the symptom detection model 320. Here, the symptom is, for example, but not limited to, hard leakage.

The physiological tissue detection model 330 is a physiological tissue (including organs and tissues) detection model related to the disease, and analyzes the original medical image RMI to obtain the physiological tissue detection result 360. The physiological tissue detection result 360 includes the position, vertical length (first length), horizontal length (second length), and confidence value of each physiological tissue. Optionally, the area of each physiological tissue can be calculated by the vertical length and horizontal length of the bounding box selected by the physiological tissue detection model 330. Here, the physiological tissue is, for example, but not limited to, the optic disc, the center of the macula, etc.

The symptom detection result 350 and the physiological tissue detection result 360 are input to the Features Integration and Transformation module 370 to integrate and quantitatively transform the object features (such as but not limited to symptoms and physiological tissues). In detail, the Features Integration and Transformation module 370 uses the symptom detection result and the object detection result of the physiological tissue to obtain a scalar or vector through transformation. The quantitative transformation result (scalar or vector) of the Features Integration and Transformation module 370 is input to the machine learning module 380 together with the disease classification result 340. The machine learning module 380 can find a plurality of judgment rules through a machine learning (ML) algorithm to obtain an image interpretation result 390. The image interpretation result 390 may be input to the display unit 230 for display. The image interpretation result may indicate the probability of DME, for example.

The object detection result (scalar or vector) and the disease classification result 340 can be used as a plurality of training data (each training data includes an object detection result (scalar or vector) and a disease classification result 340) to train the machine learning module 380 to obtain image interpretation results.

In one embodiment of the present case, the feature integration conversion module 370 and the machine learning module 380 can be implemented by using, for example, a chip, a circuit block in a chip, a firmware circuit, a circuit board containing a plurality of electronic components and wires, or a storage medium storing a plurality of sets of program codes, or can be implemented by executing corresponding software or programs in electronic devices such as computer systems and servers. The feature integration conversion module 370 and the machine learning module 380 are executed by the processor 210 of FIG. 2 .

Figure 4 shows a schematic diagram of medical image analysis according to another embodiment of the present case. As shown in Figure 4, the original medical image RMI is input to artificial intelligence (AI) models for a variety of different tasks, such as, but not limited to, classification model 310, disease symptom detection model 320, physiological tissue detection model 330 and classification. Model 410.

As shown in FIG. 4 , the physiological tissue detection results 360 (ie, the target area) generated by the physiological tissue detection model 330 are input to the classification model 410 . For example, the classification model 410 takes the macula region of the original medical image RMI as input. Alternatively, the macular center area may be selected by the bounding box of the physiological tissue detection result 360, or may be selected by using the macular center and the optic disc diameter, where the macular center and the optic disc diameter are It is calculated using the physiological tissue detection results 360 and analyzed by the classification model 410 to generate the disease classification result 420. One of the important points of this embodiment is that the central area of the macula is defined by the physiological tissue detection model 330 and is used as the input of the classification model 410 . In particular, during the training or/and prediction process of the classification model 410, the defined central area of the macula will be regarded as the input image, rather than the entire original medical image RMI. Compared with using the entire original medical image RMI as the input of the classification model 410, even after undergoing an image reduction operation (image reduction operation) that is often used to save computing time in the deep learning and/or prediction process, using the above The way to define the central area of the macula is more likely to be detected by the classification model 410. This effect is particularly useful when the lesion or anatomic landmark is very small relative to the RMI of the entire original medical image. This ensures that the classification model 410 has the ability to learn to recognize small objects in the central area of the macula.

As shown in Figure 4, the disease classification results 420 of the classification model 410 are also input to the machine learning module 380.

The operating principle of the feature integration conversion module 370 will now be described.

FIG. 5A shows a schematic diagram of dividing the fundus image into four quadrants with the center of the optic disc as the center point in an embodiment of the present invention. In FIG. 5A, _S1 to _S4 represent the first to fourth quadrants, respectively. In addition, _Sn represents the nth physiological tissue, such as the center of the macula, quadrant, etc. (n is a positive integer). In an embodiment of the present invention, n=1~6, but the range of n can be adjusted according to the actual application. That is, in the embodiment of the present invention, the quadrant also represents a physiological tissue. represents the i-th symptom among the m-th symptom, such as Represents the first hard leak, Represents the third hard leak. Representative symptoms Relative distance from physiological tissue S _n , 0 ≦ ≦1. For example, Representative symptoms Relative distance from physiological tissue S _n .

In an embodiment of the present invention, taking the relationship between quadrants (n=1-4) and symptoms as an example, the quantization method of the feature integration conversion module 370 takes into account any combination of the correlation between the symptom and the quadrant, the symptom area, the confidence value, etc. In an embodiment of the present invention, the quantization degree is positively correlated with the degree of angular membership of the quadrant, the area, the confidence value, or any combination thereof. In an embodiment of the present invention, for example but not limited to, the first positive relationship between multiple quadrants and symptoms can be expressed by formula (1): Formula (I).

In formula (1), the parameters p, q, and r can be obtained through data learning. The degree of quadrant belonging of the angle can be represented by a fuzzy function (for example, but not limited to, as shown in Figure 6). Symptoms Angle with the horizontal axis (radian), Represents the degree to which the angle belongs to the quadrant. Representative symptoms The relative area of Represents confidence value. ( ) represents the angle belonging to the quadrant function. Among them, 0≦ , , ( )≦1. In one embodiment of the present invention, the plane coordinates used are the center of the optic disc as the origin, the horizontal line as the horizontal axis, and the vertical line as the vertical axis. That is, the first positive relationship between the quadrants and the symptom is related to a first operation result ( ), the area and a second parameter (parameter q) a second operation result ( ), and a third operation result ( ).

In one embodiment of the present case, taking the relationship between a general physiological tissue and a symptom as an example, the quantification method of the feature integration conversion module 370 takes into account any combination of the correlation between the symptom and the physiological tissue (for example, the inverse of the distance between the symptom and the physiological tissue), the symptom area, the confidence value, etc. In one embodiment of the present case, the quantification degree value is positively correlated with (1) the inverse of the distance between the symptom and the physiological tissue, (2) the area, (3) the confidence value, or any combination thereof. In one embodiment of the present case, for example but not limited to, the second positive relationship between the physiological tissue and the symptom can be expressed by formula (II): Formula (II).

In formula (2), the parameters p', q', and r' can be obtained through data learning. That is, the second positive relationship between the physiological tissue and the symptom is related to the fourth operation result (the inverse of the distance between the symptom and the physiological tissue) and the fourth parameter (parameter p'). ), the area and the fifth parameter (parameter q') a fifth operation result ( ), and a sixth operation result ( ).

FIG. 5B shows a non-diabetic fundus image in another embodiment of the present invention. In FIG. 5B , S ₁ and S ₂ represent two physiological tissues, such as blood vessels and pancreas _{, respectively} . Represents the i-th symptom among the m-th symptom (e.g. pancreatic tumor). Representative symptoms Relative distance from physiological tissue S _n , 0 ≦ ≦1. For example, Representative symptoms Relative distance from physiological tissue S ₁ .

Similarly, with reference to FIG. 5B, in an embodiment of the present invention, taking the relationship between a general physiological tissue (such as a blood vessel or pancreas) and a symptom (such as a pancreatic tumor) as an example, the quantification method of the feature integration conversion module 370 takes into account any combination of the correlation between the symptom and the physiological tissue (for example, the inverse of the distance between the symptom and the physiological tissue), the symptom area, the confidence value, etc. In an embodiment of the present invention, the quantification degree value is positively correlated with (1) the inverse of the distance between the symptom and the physiological tissue, (2) the area, (3) the confidence value, or any combination thereof. The second positive relationship between the physiological tissue and the symptom can be expressed by formula (2), the details of which are not repeated here.

The aforementioned angle-belonging quadrant function can be implemented using fuzzy theory. More specifically, as shown in FIG. 6 , four fuzzy sets can be defined, where the function output is between 0 and 1. In addition, the shape of the angle-belonging quadrant function can be, but is not limited to, a trapezoid, a triangle, or any combination thereof, and the shape can be trained (the shape is trainable).

In the embodiment of this case, when n=1~4 (quadrant), the parameters in formula (1) Represents the confidence value of the symptom detection model 320 in the symptom output. That is, when n=1~4 (quadrant), equation (1) uses the angle membership quadrant function, area parameter and confidence parameter to calculate the quantified value. Confidence parameter The higher the number, the greater the presence of symptoms.

In the present embodiment, when n=5-6 (physiological tissue), formula (2) uses the distance parameter (especially the reciprocal of the distance parameter), the area parameter and the confidence parameter to calculate a quantitative value.

In the present embodiment, the parameters p, q, r, p', q' and r' are real numbers and can be updated during model training, for example but not limited to using backpropagation to find the best solution for the parameters p, q, r, p', q' and r'. In addition, Bayesian Optimization can also be used to find the best combination solution for the parameters p, q, r, p', q' and r', wherein when using Bayesian Optimization, the conditions of p, q, r, p', q' and r' can be but not limited to: 0≦p,q,r,p',q',r'. The parameters p, q, r, p', q' and r' can be updated by backpropagation or Bayesian Optimization.

In one embodiment of this case, the feature integration conversion module 370 converts the symptom detection results 350 and the physiological tissue detection results 360 into feature vectors or scalars, that is, functions .

Specifically, the symptom detection result 350 obtained by the symptom detection model 320 may be a symptom result matrix, as shown in FIG7. In the symptom result matrix, each row element includes: [Lesion, X, Y, W, H, C], wherein Lesion, X, Y, W, H, C represent: the symptom type (Lesion) selected by the symptom detection model prediction box, the X coordinate of the symptom position (X), the Y coordinate of the symptom position (Y), the horizontal length of the symptom (W), the vertical length of the symptom (H), and the confidence value of the symptom (C).

The physiological tissue detection results 360 obtained by the physiological tissue detection model 330 are a physiological tissue result matrix, as shown in Figure 7 . In this physiological tissue result matrix, each column element includes: [Structure, X, Y, W, H, C], where Structure, X, Y, W, H, C respectively represent: the prediction of the physiological tissue detection model Box-selected physiological tissue type (Structure), X coordinate of physiological tissue position (X), Y coordinate of physiological tissue position (Y), horizontal length of physiological tissue (W), vertical length of physiological tissue (H), physiological tissue Confidence value (C).

The feature integration conversion module 370 integrates the symptom result matrix and the physiological tissue result matrix into a symptom-physiological tissue relationship matrix (as shown in FIG. 7). The corresponding relationship and severity between various types of physiological tissues and various types of symptoms in the symptom-physiological tissue relationship matrix can be generated in a quantitative manner. In FIG. 7, it is assumed that there are 5 types of symptoms, namely: _L1 represents hard oozing (HE), _L2 represents hemorrhage (H), _L3 represents soft oozing (SE), _L4 represents neovascularization (NE), and _L5 represents micro-arteriovenous aneurysm (MA).

Figure 8 shows a schematic diagram of merging the output of multiple AI models to input to the machine learning module 380 according to an embodiment of the present case. As shown in Figure 8, the disease classification result 340 of the classification model 310 can be expressed as a one-dimensional confidence value matrix. As mentioned above, the feature integration conversion module 370 integrates the disease symptom detection results 350 (can be expressed as a disease symptom result matrix) and the physiological tissue detection results 360 (can be expressed as a physiological tissue result matrix) into the disease symptom physiological tissue relationship matrix 810. After that, The feature integration conversion module 370 flattens the disease-symptom physiological tissue relationship matrix 810 into a one-dimensional disease-symptom physiological tissue relationship matrix 820. The disease classification result 340 (one-dimensional confidence value matrix) and the one-dimensional disease symptom physiological tissue relationship matrix 820 are input to the machine learning module 380 to perform the machine learning algorithm to obtain the image interpretation result 390.

In an embodiment of this case, the machine learning algorithm is, for example but not limited to, a decision tree. Figure 9 shows a schematic diagram of a machine learning algorithm according to an embodiment of this case. As shown in Figure 9, it is determined at node 910 whether the confidence value F _cls is less than 0.158. If node 910 is false, jump to node 915 to determine that there is DME. If node 910 is true, jump to node 920 to determine whether the confidence value F _cls is less than 0.016. If node 920 is true, jump to node 930 to determine that there is no DME. If node 920 is false, jump to node 925 to determine whether the confidence value F _SL is less than 0.221. If node 925 is false, jump to node 935 to determine that there is DME. If node 925 is true, jump to node 940 to determine that there is no DME.

FIG. 10 shows an example of the image interpretation result 390 displayed on the display unit 230 according to an embodiment of the present invention. As shown in Figure 10, the image interpretation results 390 include: original medical image RMI (for example, but not limited to, a facial image, which further displays disease symptoms and physiological tissues), DR interpretation results (two categories) 1010, DR Interpretation results (five categories) 1020 and DME interpretation results 1030. Of course, Figure 10 is an example showing the image interpretation result 390, and it should be noted that this case is not limited thereto.

Figure 11 shows a flow chart of a medical image analysis method according to an embodiment of the present case. As shown in Figure 11, in step 1105, the original medical image is read. In step 1110, it is determined whether the size of the original medical image is smaller than a predetermined size threshold. If step 1110 is negative, the size of the original medical image is adjusted in step 1115 to be smaller than the predetermined size threshold. If step 1110 is true, then in step 1120, multiple complementary AI models (such as classification models, symptom detection models, physiological tissue detection models, etc.) are used to perform image classification and object detection on the medical images to obtain a first Classification results and multiple object detection results. In step 1125, the object characteristics (such as, but not limited to, disease symptoms, physiological tissue, etc.) are integrated and quantitatively transformed for two of the detection results to obtain a quantitative result. In step 1130, perform machine learning on the quantification result and the first classification result to obtain an image interpretation result, and display the image interpretation result.

The original medical image can be a fundus image (such as Figure 5A) or a non-fundus image (such as Figure 5B), which are within the spirit of this case.

The details of steps 1120, 1125 and 1130 can be as described above and will not be repeated here.

The performance difference between the conventional technology and the embodiment of this case is compared based on the DME disease severity results: sensitivity specificity Accuracy Use a classification model 84.61% 93.46% 93.56% Using symptom detection models and physiological tissue detection models 82.52% 93.36% 92.47% Example of this case 90.91% 94.24% 93.97%

It can be seen from the above table that the embodiment of this case has improved in terms of sensitivity, specificity and accuracy.

The embodiments of this case disclose methods and systems for assisting medical image analysis based on disease symptom detection and physiological tissue detection, especially methods and systems for medical image analysis that combine multi-task AI models.

In the present embodiment, by combining related AI models with different tasks (such as classification models, symptom detection models, physiological tissue detection models, etc.) in a complementary manner, the overall system's performance in lesion severity classification is effectively improved, thereby improving the accuracy of medical image interpretation. In the present embodiment, the classification model is used to assist the inadequacy of the detection model (and vice versa), that is, the functions of these models complement each other, which can improve the shortcomings of using a single model and thus improve the misjudgment rate.

In this embodiment, by combining the extensive interpretation results of the classification model and the results based on pathological analysis, the correlation information between disease symptoms and physiological tissues is obtained, and then the best decision is found through machine learning. Therefore, common problems (such as simply The CNN classification model has displacement invariance and cannot accurately determine the relative positional relationship between disease symptoms and physiological tissue. It is easy to misjudge when the symptoms are small, or under certain circumstances (such as when the disease symptoms completely cover the physiological tissue, the physiological tissue model cannot be identified). causing misjudgment and other problems).

The present invention finds the best output decision rules (prediction rules) through machine learning, which is also applicable to applications other than medical image recognition.

The embodiment of this case can learn the structure distribution contained in the data (data-driven), and can learn high-dimensional and highly abstract data distribution structures; it does not require subjective rules to be set by humans, and is relatively objective; through expert annotation, it can directly learn the experience and knowledge of experts, avoiding the situation where experts cannot describe knowledge and experience through language; and independently adapt to new data.

In summary, the embodiments of the present invention can improve the performance of the lesion screening system; have high interpretability; and these rules can be extended to fuzzy systems in the future and used as fuzzy rules.

In summary, although the present invention has been disclosed above through embodiments, they are not intended to limit the present invention. Those with ordinary knowledge in the technical field to which the present invention belongs can make various modifications and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the appended patent application scope.

100: Original medical image 110: Center of macula R: Diameter of optic disc HE: Hard permeation 111: Optic disc 200: Medical image analysis device 210: Processor 220: Database 230: Display unit RMI: Original medical image 310: Classification model 320: Symptom detection model 330: Physiological tissue detection model 340: Disease classification result 350: Symptom detection result 360: Physiological tissue detection result 370: Feature integration conversion module 380: Machine learning module 390: Image interpretation result 410: Classification model 420: Disease classification result 810: Symptom-physiology-tissue relationship matrix 820: One-dimensional symptom-physiology-tissue relationship matrix 910-940: Nodes 1010: DR interpretation results (two-classification) 1020: DR interpretation results (five-classification) 1030: DME interpretation results 1105-1130: Steps

Image 1 shows diabetic macular edema. Figure 2 shows a functional schematic diagram of a medical image analysis device according to an embodiment of the present invention. Figure 3 shows a schematic diagram of medical image analysis according to an embodiment of this case. Figure 4 shows a schematic diagram of medical image analysis according to another embodiment of the present case. Figure 5A shows a schematic diagram of dividing the fundus image into four quadrants with the optic disc as the center point in one embodiment of this case. Figure 5B shows a display range of a non-diabetic fundus image in another embodiment of this case. Figure 6 shows a schematic diagram of the angle membership quadrant function according to an embodiment of the present case. Figure 7 shows an integrated disease-symptom-physiological-tissue relationship matrix according to an embodiment of this case. Figure 8 shows a schematic diagram of merging the output of multiple AI models to input to the machine learning module according to an embodiment of the present case. Figure 9 shows a schematic diagram of a machine learning algorithm according to an embodiment of this case. Figure 10 shows an example of image interpretation results displayed on the display unit according to an embodiment of the present invention. Figure 11 shows a flow chart of a medical image analysis method according to an embodiment of the present case.

RMI: Raw Medical Image

310:Classification model

320:Symptom Detection Model

330: Physiological tissue detection model

340:Disease classification results

350: Symptom detection results

360: Physiological tissue detection results

370: Feature integration conversion module

380: Machine learning module

390: Image interpretation results

Claims (26)

A medical image analysis method includes: reading an original medical image; using a first classification model, a symptom detection model, and a physiological tissue detection model to perform image classification and object detection on the original medical image to obtain respectively A first classification result, a symptom detection result, and a physiological tissue detection result; a feature integration conversion module integrates and quantifies the object characteristics of the symptom detection result and the physiological tissue detection result, so as to Obtaining a quantified result that is related to the correlation between a disease symptom and a physiological tissue obtained from the disease symptom detection result and the physiological tissue detection result; and using a machine learning module to correlate the quantified result with the physiological tissue detection result The first classification result is subjected to machine learning to obtain and display an image interpretation result; wherein, the feature integration conversion module obtains a scalar quantity or vector through quantitative conversion of the symptom detection result and the physiological tissue detection result; wherein, When the feature integration conversion module performs quantification, for the relationship between multiple quadrants and a disease symptom, the degree of quantification has a first positive relationship with an angle's degree of belonging to a quadrant, an area, a confidence value, or any combination thereof; Among them, when the feature integration conversion module performs quantification, for the relationship between a physiological tissue and the disease symptom, the degree of quantification is respectively related to a reciprocal of a distance between the disease symptom and a physiological tissue, the area, and the confidence value. or any combination thereof exhibits a second positive relationship. The medical image analysis method as described in claim 1 further includes: when a processor determines that a size of the original medical image is not less than a predetermined size threshold, the processor adjusts the size of the original medical image to be less than the predetermined size threshold. The medical image analysis method as described in claim 1, wherein the first classification model is used to analyze the original medical image to obtain the first classification result, and the first classification result is a disease classification result; the symptom detection model analyzes the original medical image to obtain the symptom detection result, and the symptom detection result includes a position, an area, a confidence value and a total number of symptoms of each symptom type for each symptom type; and the physiological tissue detection model analyzes the original medical image to obtain the physiological tissue detection result, and the physiological tissue detection result includes a position, a first length, a second length and a confidence value for each physiological tissue. The medical image analysis method of claim 3, wherein the area of each symptom is calculated by a processor through a vertical length and a horizontal length of a certain bounding box detected by the symptom detection model. The medical image analysis method as described in claim 3, wherein the complementary artificial intelligence models further include a second classification model, the physiological tissue detection result is input into the second classification model, the second classification model selects a target area from the original medical image according to the physiological tissue detection result, analyzes the selected target area, and obtains a second classification result, and the second classification result is input into the machine learning module. The medical image analysis method as described in claim 1, wherein the first positive relationship between the quadrants and the symptom is related to a first operation result of the degree of quadrant belonging of the angle and a first parameter, a second operation result of the area and a second parameter, and a third operation result of the confidence value and a third parameter; and the second positive relationship between the physiological tissue and the symptom is related to a fourth operation result of the reciprocal of the distance between the symptom and the physiological tissue and a fourth parameter, a fifth operation result of the area and a fifth parameter, and a sixth operation result of the confidence value and a sixth parameter. The medical image analysis method as described in claim 6, wherein the first parameter, the second parameter, the third parameter, the fourth parameter, the fifth parameter and the sixth parameter are obtained through data learning. The medical image analysis method as described in claim 7, wherein the first parameter, the second parameter, the third parameter, the fourth parameter, the fifth parameter and the sixth parameter are updated by back propagation or Bayesian optimization. The medical image analysis method as described in claim 1, wherein when the degree of the angle's membership in the quadrant is implemented by a fuzzy theory, a plurality of fuzzy sets are defined by a processor; when the degree of the angle's membership in the quadrant is represented by a function, One of the outputs of the function is between 0 and 1; and the shape of one of the angle membership quadrant functions is a trapezoid, a triangle or any combination thereof, and the shape is trainable. The medical image analysis method as described in claim 1, wherein the symptom detection result is a symptom result matrix, and each column element in the symptom result matrix includes: a type of the symptom, a location of the symptom, A horizontal length, a vertical length of the symptom and a confidence value of the symptom, and the physiological tissue detection result is a physiological tissue result matrix, and each column element of the physiological tissue result matrix includes: a type of the physiological tissue , a position of the physiological tissue, a horizontal length of the physiological tissue, a vertical length of the physiological tissue and a confidence value of the physiological tissue. The medical image analysis method as claimed in claim 10, wherein the feature integration conversion module integrates the disease symptom result matrix and the physiological tissue result matrix into a disease symptom physiological tissue relationship matrix; and the first classification result is a one-dimensional Confidence value matrix, this feature integration conversion module flattens the disease-symptom physiological tissue relationship matrix into a one-dimensional disease-symptom physiological tissue relationship matrix. The medical image analysis method as described in claim 11, wherein the machine learning module performs machine learning on the one-dimensional confidence value matrix and the one-dimensional symptom physiological tissue relationship matrix to obtain the image interpretation result. The medical image analysis method as claimed in claim 12, wherein the image interpretation result includes a medical image with the symptom and the physiological tissue and at least one interpretation result; and the original medical image includes a fundus image or a non-fundus image. image. A medical image analysis device includes a processor; and a display unit coupled to the processor, wherein the processor is configured to: read an original medical image; perform image classification and object detection on the original medical image using a first classification model, a symptom detection model, and a physiological tissue detection model to obtain a first classification result, a symptom detection result, and a physiological tissue detection result, respectively; physiological tissue detection result; integrating and quantifying the object features of the symptom detection result and the physiological tissue detection result by a feature integration conversion module to obtain a quantitative result, which is related to the correlation between a symptom and a physiological tissue obtained from the symptom detection result and the physiological tissue detection result; and integrating and quantifying the object features of the symptom detection result and the physiological tissue detection result by a machine learning module. A classification result is machine-learned to obtain an image recognition result, and the image recognition result is displayed on the display unit; wherein the feature integration conversion module obtains a scalar or vector through quantization conversion of the symptom detection result and the physiological tissue detection result; wherein, when the feature integration conversion module performs quantization, for the relationship between a plurality of quadrants and a symptom, the quantization degree is respectively in a first positive relationship with an angle belonging to a quadrant degree, an area, a confidence value or any combination thereof; wherein, when the feature integration conversion module performs quantization, for the relationship between a physiological tissue and the symptom, the quantization degree is respectively in a second positive relationship with an inverse of a distance between the symptom and a physiological tissue, the area, the confidence value or any combination thereof. The medical image analysis device as claimed in claim 14, wherein the processor is configured to: when determining that a size of the original medical image is not smaller than a predetermined size threshold, adjust the size of the original medical image to be smaller than the size threshold. Established size threshold. The medical image analysis device as claimed in claim 14, wherein the first classification model is used to analyze the original medical image to obtain the first classification result, and the first classification result is a disease classification result; the disease symptom detection A model that analyzes the original medical image to obtain the symptom detection result, which includes a location, an area, a confidence value for each symptom, and a total number of symptoms for each symptom type; and the physiological tissue detection model , analyze the original medical image to obtain the physiological tissue detection result. The physiological tissue detection result includes a position, a first length, a second length and a confidence value of each physiological tissue. A medical image analysis device as described in claim 16, wherein the area of each symptom is calculated by a vertical length and a horizontal length of a certain bounding box detected by the symptom detection model. The medical image analysis device according to claim 17, wherein the complementary artificial intelligence models further include a second classification model, the physiological tissue detection The results are input to the second classification model. The second classification model selects a target area from the original medical image based on the physiological tissue detection result, and analyzes the selected target area to obtain a second classification result. , the second classification result is input to the machine learning module. The medical image analysis device as claimed in claim 14, wherein the first positive relationship between the quadrants and the symptom is related to a first calculation result of the degree of the angle belonging to the quadrant and a first parameter, the area and a first parameter. A second operation result of a second parameter, and a third operation result of the confidence value and a third parameter; and the second positive relationship between the physiological tissue and the disease symptom is related to the relationship between the disease symptom and the physiological tissue. a fourth operation result of the reciprocal of the distance and a fourth parameter, a fifth operation result of the area and a fifth parameter, and a sixth operation result of the confidence value and a sixth parameter. The medical image analysis device as described in claim 19, wherein the first parameter, the second parameter, the third parameter, the fourth parameter, the fifth parameter and the sixth parameter are obtained through data learning. A medical image analysis device as described in claim 20, wherein the first parameter, the second parameter, the third parameter, the fourth parameter, the fifth parameter and the sixth parameter are updated by back propagation or Bayesian optimization. The medical image analysis device as claimed in claim 14, wherein when the degree of the angle belonging to a quadrant is realized using a fuzzy theory, a plurality of fuzzy sets are defined; When the degree of the angle's quadrant membership is represented by a function, one of the outputs of the function is between 0 and 1; and one shape of the angle's quadrant membership function is a trapezoid, a triangle or any combination thereof, and the shape is trained. The medical image analysis device as described in claim 14, wherein the symptom detection result is a symptom result matrix, each row element in the symptom result matrix includes: a type of the symptom, a position of the symptom, a horizontal length of the symptom, a vertical length of the symptom, and a confidence value of the symptom, and the physiological tissue detection result is a physiological tissue result matrix, each row element in the physiological tissue result matrix includes: a type of the physiological tissue, a position of the physiological tissue, a horizontal length of the physiological tissue, a vertical length of the physiological tissue, and a confidence value of the physiological tissue. The medical image analysis device as described in claim 23, wherein the feature integration conversion module integrates the symptom result matrix and the physiological tissue result matrix into a symptom physiological tissue relationship matrix; and the first classification result is a one-dimensional confidence value matrix, and the feature integration conversion module flattens the symptom physiological tissue relationship matrix into a one-dimensional symptom physiological tissue relationship matrix. The medical image analysis device as described in claim 24, wherein the machine learning module performs machine learning on the one-dimensional confidence value matrix and the one-dimensional symptom physiological tissue relationship matrix to obtain the image interpretation result. The medical image analysis device of claim 25, wherein the image interpretation result includes a medical image with the symptom and the physiological tissue and at least one interpretation result; and the original medical image includes a fundus image or a non-fundus image. image.

Images