Intelligent medical image reading method, device, equipment and storage medium

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses an intelligent medical image reading method, device, equipment and storage medium, which are used to increase the types of images for artificial intelligence medical image reading. The intelligent medical image reading method includes: obtaining a target image; calling a preset medical image reading model to detect the image type of the target image to obtain a target image type; if the target image type is a type of image, performing lesion recognition on the target image, and if there is a lesion area in the target image, marking the lesion area in the target image and sending the marked image to a medical staff terminal; if the target image type is a type of image, performing preprocessing on the target image to obtain a preprocessed type of image; performing lesion recognition on the preprocessed type of image, and if there is a lesion area in the target image, marking the lesion area in the target image and sending the marked image to a medical staff terminal.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to an intelligent medical image reading method, device, equipment and storage medium. Background Art

[0002] With the economic development and social progress, the quality of people's life has improved, and various diseases have gradually become younger. At the same time, with the convenience of seeing a doctor, many people go to the hospital to see a doctor. Since some diseases require professional medical equipment to collect images of human internal tissues or body surfaces for lesion recognition, a large number of images of human internal tissues or body surfaces are generated. A large number of images of human internal tissues or body surfaces need to be manually read by doctors in the hospital, resulting in a large workload of manual image reading and easy misdiagnosis.

[0003] With the development of artificial intelligence (AI) technology, more and more artificial intelligence technologies have been applied in various fields. Existing artificial intelligence medical image reading can perform lesion recognition on images of human internal tissues or body surfaces collected by professional medical equipment, so as to obtain the lesion area.

[0004] Existing artificial intelligence medical image reading is to perform lesion recognition on images of human internal tissues or body surfaces collected by professional medical equipment in a single way, and the types of images that can be recognized are few. Summary of the Invention

[0005] The present invention provides an intelligent medical image reading method, device, equipment and storage medium for increasing the types of images for artificial intelligence medical image reading.

[0006] In the first aspect of the present invention, an intelligent medical image reading method is provided, including: obtaining a target image, where the target image is a type of image or a type of image to be recognized, the type of image is an in-vivo image or a body surface image collected by a professional medical equipment, and the type of image is a body surface image collected by a non-professional medical equipment; calling a preset medical image reading model to perform image type detection on the target image to obtain a target image type; if the target image type is the type of image, performing lesion recognition on the target image, and if there is a lesion area in the target image, marking the lesion area in the target image and sending the marked image to a medical staff terminal; if the target image type is the type of image, performing preprocessing on the target image to obtain a preprocessed type of image, where the preprocessed type of image is an image on which the preset medical image reading model can perform lesion recognition; performing lesion recognition on the preprocessed type of image, and if there is a lesion area in the target image, marking the lesion area in the target image and sending the marked image to a medical staff terminal.

[0007] In a feasible implementation manner, the step of detecting the image type of the target image by invoking a preset medical image reading model to obtain the target image type includes: performing image noise analysis on the target image by using the preset medical image reading model to obtain an image noise analysis result; if the image noise analysis result indicates that Poisson noise exists in the target image, determining the target image type corresponding to the target image as a type-one image; if the image noise analysis result indicates that Gaussian noise exists in the target image, determining the target image type corresponding to the target image as a type-two image.

[0008] In a feasible implementation manner, if the target image type is the type-one image, then performing lesion recognition on the target image. If a lesion area exists in the target image, marking the lesion area in the target image and sending the marked image to the medical staff terminal, including: if the target image type is the type-one image, performing segmentation processing on the target image to obtain a target image area; if the pixel value of each pixel point in the target image area is greater than or equal to a preset pixel value, determining that a lesion area exists in the target image; marking the lesion area in the target image and sending the marked image to the medical staff terminal.

[0009] In a feasible implementation manner, after performing segmentation processing on the target image to obtain a target image area when the target image type is the type-one image, the method further includes: if the pixel value of each pixel point in the target image area is less than the preset pixel value, determining that no lesion area exists in the target image and generating a reminder message, and sending the reminder message to the medical staff terminal, where the reminder message is used to indicate a review of the lesion recognition result of the target image.

[0010] In a feasible implementation manner, if the target image type is the type-two image, then performing preprocessing on the target image to obtain a preprocessed type-two image, including: if the target image type is the type-two image, performing noise reduction processing on the target image to obtain a noise-reduced target image; performing grayscale processing on the noise-reduced target image to obtain a grayscale target image; performing binarization processing on the grayscale target image to obtain a preprocessed type-two image.

[0011] In a feasible implementation manner, for the identification of lesions in the preprocessed second-class images, if there is a lesion area in the target image, then mark the lesion area in the target image and send the marked image to the medical staff terminal, including: segmenting the preprocessed second-class images into image blocks to obtain a plurality of image blocks; performing convolution on each of the plurality of image blocks one by one according to a specified order to obtain the convolution value of each corresponding image block; if the convolution value of an image block is greater than or equal to a preset convolution value, then determine that there is a lesion area in the corresponding image block, and add the image block with the lesion area to a set of lesion image blocks, where the set of lesion image blocks includes a plurality of lesion image blocks; mark the lesion area in each lesion image block of the plurality of lesion image blocks to generate a marked target image, and send the marked target image to the medical staff terminal.

[0012] In a feasible implementation manner, before acquiring the target image, it further includes: acquiring the first-class images and second-class images of historical patients, where there are lesion areas in the first-class images and second-class images; extracting the first-class images to obtain the target lesion areas in the first-class images, and marking the target lesion areas to obtain marked first-class images; extracting the second-class images to obtain the target lesion areas in the second-class images, and marking the target lesion areas to obtain marked second-class images; performing model training according to the marked first-class images and the marked second-class images to generate a preset medical image reading model.

[0013] The second aspect of the present invention provides an intelligent medical image reading device, including: an acquisition module, configured to acquire a target image, where the target image is a first-class image or a second-class image to be identified, the first-class image is an in-vivo image or a surface image collected by a professional medical device, and the second-class image is a surface image collected by a non-professional medical device; a detection module, configured to call a preset medical image reading model to perform image type detection on the target image to obtain a target image type; a first identification and sending module, configured to, if the target image type is the first-class image, perform lesion identification on the target image, and if there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal; a preprocessing module, configured to, if the target image type is the second-class image, perform preprocessing on the target image to obtain a preprocessed second-class image, where the preprocessed second-class image is an image on which the preset medical image reading model can perform lesion identification; a second identification and sending module, configured to perform lesion identification on the preprocessed second-class image, and if there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal.

[0014] In a feasible implementation manner, the detection module is specifically configured to: perform image noise analysis on the target image through a preset medical image reading model to obtain an image noise analysis result; if the image noise analysis result indicates that Poisson noise exists in the target image, determine the target image type corresponding to the target image as a type-one image; if the image noise analysis result indicates that Gaussian noise exists in the target image, determine the target image type corresponding to the target image as a type-two image.

[0015] In a feasible implementation manner, the first recognition and sending module includes: a segmentation unit, configured to, if the target image type is the type-one image, perform segmentation processing on the target image to obtain a target image region; a determination unit, configured to, if the pixel value of each pixel point in the target image region is greater than or equal to a preset pixel value, determine that a lesion region exists in the target image; a marking unit, configured to mark the lesion region in the target image and send the marked image to a medical staff terminal.

[0016] In a feasible implementation manner, the first recognition and sending module further includes: a reminder unit, configured to, if the pixel value of each pixel point in the target image region is less than the preset pixel value, determine that no lesion region exists in the target image, generate a reminder message, and send the reminder message to the medical staff terminal, where the reminder message is used to indicate a review of the lesion recognition result of the target image.

[0017] In a feasible implementation manner, the preprocessing module is specifically configured to: if the target image type is the type-two image, perform noise reduction processing on the target image to obtain a noise-reduced target image; perform grayscale processing on the noise-reduced target image to obtain a grayscale target image; perform binarization processing on the grayscale target image to obtain a preprocessed type-two image.

[0018] In a feasible implementation manner, the second recognition and sending module is specifically configured to: perform image block segmentation on the preprocessed type-two image to obtain a plurality of image blocks; perform convolution on each of the plurality of image blocks one by one according to a specified order to obtain the convolution value of each corresponding image block; if the convolution value of an image block is greater than or equal to a preset convolution value, determine that a lesion region exists in the corresponding image block, and add the image block with the lesion region to a lesion image block set, where the lesion image block set includes a plurality of lesion image blocks; mark the lesion region in each lesion image block of the plurality of lesion image blocks to generate a marked target image, and send the marked target image to the medical staff terminal.

[0019] In a feasible implementation manner, the intelligent medical image reading device further includes: a generation module, configured to obtain a first type of image and a second type of image of a historical patient, wherein there are lesion regions in the first type of image and the second type of image; extract the first type of image to obtain a target lesion region in the first type of image, and mark the target lesion region to obtain a marked first type of image; extract the second type of image to obtain a target lesion region in the second type of image, and mark the target lesion region to obtain a marked second type of image; perform model training according to the marked first type of image and the marked second type of image to generate a preset medical image reading model.

[0020] The third aspect of the present invention provides an intelligent medical image reading device, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the intelligent medical image reading device to execute the above-mentioned intelligent medical image reading method.

[0021] The fourth aspect of the present invention provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on a computer, the computer is enabled to execute the above-mentioned intelligent medical image reading method.

[0022] In the technical solution provided by the present invention, a target image is obtained, where the target image is a type-one image or a type-two image to be recognized. The type-one image is an in-vivo image or a surface image collected by a professional medical device, and the type-two image is a surface image collected by a non-professional medical device. A pre-set medical image reading model is called to perform image type detection on the target image to obtain the target image type. If the target image type is a type-one image, lesion recognition is performed on the target image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal. If the target image type is a type-two image, preprocessing is performed on the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which the pre-set medical image reading model can perform lesion recognition. Lesion recognition is performed on the preprocessed type-two image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal. In the embodiment of the present invention, by obtaining the target image, performing image type detection on the target image, if the target image is an in-vivo image or a surface image collected by a professional medical device, directly performing lesion recognition on the target image, if the target image is a surface image collected by a non-professional medical device, performing preprocessing on the target image, the preprocessed target image is an image on which the pre-set medical image reading model can perform lesion recognition, and then performing lesion recognition on the preprocessed target image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal, increasing the types of images for artificial intelligence medical image reading. Description of the Drawings

[0023] Figure 1 It is a schematic diagram of an embodiment of the intelligent medical image reading method in the embodiment of the present invention;

[0024] Figure 2 It is a schematic diagram of another embodiment of the intelligent medical image reading method in the embodiment of the present invention;

[0025] Figure 3 It is a schematic diagram of an embodiment of the intelligent medical image reading device in the embodiment of the present invention;

[0026] Figure 4 It is a schematic diagram of another embodiment of the intelligent medical image reading device in the embodiment of the present invention;

[0027] Figure 5 It is a schematic diagram of an embodiment of the intelligent medical image reading device in the embodiment of the present invention. Detailed Embodiments

[0028] The present invention provides an intelligent medical image reading method, device, equipment and storage medium for increasing the types of images for artificial intelligence medical image reading.

[0029] In the description and claims of the present invention and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that such data used can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that comprises a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0030] Embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Among them, Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.

[0031] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0032] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Please refer to Figure 1 , an embodiment of the intelligent medical image reading method in the embodiments of the present invention includes:

[0033] 101. Obtain a target image, where the target image is a type of image or a type of image to be recognized. The type of image is an in-vivo image or a surface image collected by a professional medical device, and the type of image is a surface image collected by a non-professional medical device;

[0034] It can be understood that the execution subject of the present invention can be an intelligent medical image reading device or a terminal, and specifically, it is not limited here. Embodiments of the present invention will be described by taking the intelligent medical image reading device as the execution subject as an example.

[0035] In this embodiment, the target image can be obtained through various data. For example, the intelligent medical image reading device obtains the target image through the online patient consultation data of the Internet hospital, or obtains the target image through the physical examination report of the patient's health hut or the physical examination report of the health space station, or obtains the target image through the registration graphic data of the patient participating in the health lecture.

[0036] 102. Call a preset medical image reading model to detect the image type of the target image to obtain the target image type;

[0037] In this embodiment, the internal or surface images collected by professional medical devices have a relatively single spectrum through the elimination of light scattering, that is, the noise distribution of the internal or surface images collected by professional medical devices is approximately regarded as a Poisson distribution. Most of the internal or surface images collected by professional medical devices are single-channel grayscale images, and all the information contained in the internal or surface images collected by professional medical devices has potential utilization value. For example, human tissues have a high degree of similarity, and a slight change in the internal or surface images collected by professional medical devices may represent a diseased tissue. Due to the existence of light scattering, the surface images collected by non-professional medical devices have a relatively wide frequency spectrum, that is, the noise distribution of the surface images collected by non-professional medical devices is approximately regarded as a Gaussian distribution.

[0038] 103. If the target image type is a type-one image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal;

[0039] In this embodiment, the target area and non-target area usually exist simultaneously in the target image, and the non-target area usually interferes with the recognition of the target area. Therefore, the target area can be segmented out, and the non-target area can be filtered out to obtain the target area. For the lesion recognition of the target image, it can be through the recognition method at the pixel level, and the position where the pixel points with pixel values within the preset range in the target image are located is estimated as the lesion area. The preset range of pixel values can be set according to historical experience. It is also possible to input the target image into a segmentation model trained to convergence, and determine the lesion area in the target image through this segmentation model.

[0040] 104. If the target image type is a type-two image, perform preprocessing on the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which the preset medical image reading model can perform lesion recognition;

[0041] In this embodiment, the surface images collected by non-professional medical devices are preprocessed by an intelligent medical image reading device, and the preprocessed images can be used for lesion recognition by the preset medical image reading model. Among them, the preprocessing process includes noise reduction, grayscale conversion, and binarization. Noise reduction is used to reduce the Gaussian noise of the surface images collected by non-professional medical devices to Poisson noise, grayscale conversion is used to convert the surface images collected by non-professional medical devices from three-channel images to single-channel images, and binarization is used to select the target area and exclude the non-target area.

[0042] 105. Identify lesions in the pre-processed Class II images. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal.

[0043] In this embodiment, a convolutional neural network can be used to identify lesions in the pre-processed Class II images, or a Vision Transformer (Transformer) can be used to identify lesions in the pre-processed Class II images. The convolutional neural network includes various neural networks, such as Region-based Convolutional Neural Network (Region-CNN), Spatial Pyramid Pooling Convolutional Neural Network (SPP-Net), You Only Look Once (Yolo) object detection convolutional neural network, Single Shot MultiBox Detector (SSD) object detection convolutional neural network, and other convolutional neural networks.

[0044] In an embodiment of the present invention, a target image is obtained, where the target image is a Class I image or a Class II image to be recognized. The Class I image is an in-vivo image or a surface image collected by a professional medical device, and the Class II image is a surface image collected by a non-professional medical device. Call a pre-set medical image reading model to perform image type detection on the target image to obtain the target image type. If the target image type is a Class I image, perform lesion identification on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal. If the target image type is a Class II image, perform pre-processing on the target image to obtain a pre-processed Class II image, where the pre-processed Class II image is an image on which the pre-set medical image reading model can perform lesion identification. Perform lesion identification on the pre-processed Class II image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal, increasing the types of images for artificial intelligence medical image reading.

[0045] Please refer to Figure 2 , another embodiment of the intelligent medical image reading method in an embodiment of the present invention includes:

[0046] 201. Generate a pre-set medical image reading model based on the Class I images and Class II images of historical patients, where there are lesion areas in the Class I images and Class II images.

[0047] Specifically, (1) the intelligent medical image reading device obtains the Class I images and Class II images of historical patients, where there are lesion areas in the Class I images and Class II images; (2) extract the Class I images to obtain the target lesion areas in the Class I images, and mark the target lesion areas to obtain the marked Class I images; (3) extract the Class II images to obtain the target lesion areas in the Class II images, and mark the target lesion areas to obtain the marked Class II images; (4) perform model training based on the marked Class I images and marked Class II images to generate a pre-set medical image reading model.

[0048] For example, an intelligent medical image reading device obtains a first type of image and a second type of image of a historical patient. The first type of image is an in-vivo image or a surface image collected by a professional medical device. The in-vivo image collected by the professional medical device can be a magnetic resonance imaging (MRI) image or an X-ray image. The surface image collected by the professional medical device can be an image collected by a medical microscope imaging device. The second type of image is a surface image collected by a non-professional medical device. The surface image collected by the non-professional medical device can be a human face image or a human leg image. Among them, there are lesion areas in the first type of image and the second type of image of the historical patient; extract the first type of image to obtain the target lesion area in the first type of image. If the target lesion area is a lung tumor, mark the lung tumor to obtain a marked lung tumor image. If the target lesion area is a surface hemangioma, mark the surface hemangioma to obtain a marked surface hemangioma image; extract the second type of image to obtain the target lesion area in the second type of image. If the target lesion area is facial acne, mark the facial acne to obtain a marked facial acne image. If the target lesion area is herpes zoster, mark the herpes zoster to obtain a marked herpes zoster image; perform model training based on the marked lung tumor image or the marked surface hemangioma image and the marked facial acne image or the marked herpes zoster image to generate a preset medical image reading model.

[0049] 202. Obtain a target image, where the target image is a first type of image or a second type of image to be recognized. The first type of image is an in-vivo image or a surface image collected by a professional medical device, and the second type of image is a surface image collected by a non-professional medical device.

[0050] In this embodiment, the target image can be obtained through various data. For example, the intelligent medical image reading device obtains the target image through the online patient consultation data of the Internet hospital, or through the physical examination report of the patient's health hut or the physical examination report of the health space station, or through the registration graphic data of the patient participating in the health lecture.

[0051] In this embodiment, the professional medical device includes various devices, such as magnetic resonance imaging (MRI) image devices, computed tomography (CT) image devices, X-ray image devices, ultrasonic image devices, positron emission tomography (PET) image devices, endoscopic examination devices, medical microscope imaging devices, and other professional medical devices. The non-professional medical device includes various devices, such as mobile phones, cameras, scanners, and other non-professional medical devices.

[0052] 203. Invoke the preset medical image reading model to perform image type detection on the target image to obtain the target image type.

[0053] In this embodiment, the internal or surface images collected by professional medical devices have a relatively single spectrum by eliminating light scattering, that is, the noise distribution of the internal or surface images collected by professional medical devices is approximated as a Poisson distribution. Most of the internal or surface images collected by professional medical devices are single-channel grayscale images, and all the information contained in the internal or surface images collected by professional medical devices has potential utilization value. For example, human tissues have a high degree of similarity, and a slight change in the internal or surface images collected by professional medical devices may represent a diseased tissue. Due to the existence of light scattering, the surface images collected by non-professional medical devices have a relatively wide spectrum, that is, the noise distribution of the surface images collected by non-professional medical devices is approximated as a Gaussian distribution.

[0054] Specifically, (1) The intelligent medical image reading device performs image noise analysis on the target image through a pre-set medical image reading model to obtain an image noise analysis result; (2) If the image noise analysis result shows that the target image has Poisson noise, the target image type corresponding to the target image is determined as a type-one image; (3) If the image noise analysis result shows that the target image has Gaussian noise, the target image type corresponding to the target image is determined as a type-two image.

[0055] For example, the intelligent medical image reading device performs image noise analysis on the target image through a pre-set medical image reading model to obtain an image noise analysis result. If the image noise analysis result shows that the target image has Poisson noise, the target image type corresponding to the target image is determined as a type-one image, that is, the target image is an internal or surface image collected by a professional medical device; if the image noise analysis result shows that the target image has Gaussian noise, the target image type corresponding to the target image is determined as a type-two image, that is, the target image is a surface image collected by a non-professional medical device.

[0056] 204. If the target image type is a type-one image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal;

[0057] In this embodiment, there are usually a target area and a non-target area in the target image. The non-target area usually interferes with the recognition of the target area. Therefore, the target area can be segmented out, and the non-target area can be filtered out to obtain the target area. For the lesion recognition of the target image, it can be through a pixel-level recognition method. The position of the pixel points in the target image whose pixel values are within a preset range is estimated as the lesion area, and the preset range of the pixel values can be set according to historical experience. It is also possible to input the target image into a segmentation model trained to convergence, and determine the lesion area in the target image through this segmentation model.

[0058] Specifically, (1) if the target image type is a type-one image, the intelligent medical image reading device performs segmentation processing on the target image to obtain a target image area; (2) if the pixel value of each pixel point in the target image area is greater than or equal to a preset pixel value, it is determined that there is a lesion area in the target image; (3) the lesion area in the target image is marked, and the marked image is sent to the medical staff terminal.

[0059] After step (2), it further includes: if the pixel value of each pixel point in the target image area is less than the preset pixel value, it is determined that there is no lesion area in the target image, and a reminder message is generated. The intelligent medical image reading device sends the reminder message to the medical staff terminal, and the reminder message is used to indicate a review of the lesion recognition result of the target image.

[0060] For example, the preset pixel value is 100. If the target image is a lung image, the intelligent medical image reading device performs segmentation processing on the lung image to obtain the target image area of the lung image; if the pixel value of each pixel point in the target image area is greater than or equal to 100, it is determined that there is a lesion area in the lung image, the lesion area in the lung image is marked, and the marked lung image is sent to the medical staff terminal. If the pixel value of each pixel point in the target image area is less than 100, it is determined that there is no lesion area in the lung image, and a reminder message is generated. The intelligent medical image reading device sends the reminder message to the medical staff terminal, and the reminder message is used to indicate a review of the lesion recognition result of the lung image.

[0061] 205. If the target image type is a type-two image, preprocess the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which a preset medical image reading model can perform lesion recognition;

[0062] In this embodiment, the intelligent medical image reading device preprocesses the body surface image collected by a non-professional medical device, and only after preprocessing can the image be used for lesion recognition by a preset medical image reading model. The preprocessing process includes noise reduction, grayscale conversion, and binarization. Noise reduction is used to reduce the Gaussian noise of the body surface image collected by the non-professional medical device to Poisson noise, grayscale conversion is used to convert the body surface image collected by the non-professional medical device from a three-channel image to a single-channel image, and binarization is used to select the target area and exclude the non-target area.

[0063] Specifically, (1) if the target image type is a type-two image, the intelligent medical image reading device performs noise reduction processing on the target image to obtain a noise-reduced target image; (2) perform grayscale conversion processing on the noise-reduced target image to obtain a grayscale target image; (3) perform binarization processing on the grayscale target image to obtain a preprocessed type-two image.

[0064] For example, if the target image is a skin image, the intelligent medical image reading device performs noise reduction processing on the skin image to obtain a noise-reduced skin image, that is, converts the Gaussian noise of the skin image into Poisson noise; performs grayscale processing on the noise-reduced skin image to obtain a grayscale skin image, that is, converts the three-channel skin image into a single-channel skin image; performs binarization processing on the grayscale skin image to obtain a preprocessed skin image, where the grayscale value of the pixel points in the preprocessed skin image is 0 or 255.

[0065] 206. Perform lesion recognition on the preprocessed two-category image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal.

[0066] In this embodiment, lesion recognition can be performed on the preprocessed two-category image through a convolutional neural network, or lesion recognition can be performed on the preprocessed two-category image through a vision transformer. The convolutional neural network includes various neural networks, for example, Region-CNN (Region Convolutional Neural Network), SPP-Net (Spatial Pyramid Pooling Convolutional Neural Network), Yolo (Object Detection Convolutional Neural Network), SSD (Object Detection Convolutional Neural Network), and other convolutional neural networks.

[0067] Specifically, (1) The intelligent medical image reading device divides the preprocessed two-category image into image blocks to obtain multiple image blocks; (2) Convolve each of the multiple image blocks one by one according to the specified order to obtain the convolution value of each corresponding image block; (3) If the convolution value of the image block is greater than or equal to the preset convolution value, it is determined that there is a lesion area in the corresponding image block, and the image block with the lesion area is added to the lesion image block set, and the lesion image block set includes multiple lesion image blocks; (4) Mark the lesion area in each lesion image block of the multiple lesion image blocks to generate a marked target image, and send the marked target image to the medical staff terminal.

[0068] After step (2), it further includes: If the convolution value of each image block is less than the preset convolution value, it is determined that there is no lesion area in the target image, then a reminder message for reviewing the lesion recognition result of the target image is generated, and the reminder message is sent to the medical staff terminal.

[0069] For example, an intelligent medical image reading device segments the preprocessed type-II image into image blocks, obtaining multiple image blocks, including image block No. 1, image block No. 2, image block No. 3, and image block No. 4; convolves each of the multiple image blocks one by one in the order from 1 to 4, obtaining a convolution value of 0.6 for image block No. 1, a convolution value of 0.4 for image block No. 2, a convolution value of 0.7 for image block No. 3, and a convolution value of 0.5 for image block No. 4; if the preset convolution value is 0.5, then the convolution value of image block No. 1 is greater than the preset convolution value, that is, it is determined that there is a lesion area in image block No. 1, and image block No. 1 is added to the set of lesion image blocks. Then, the convolution value of image block No. 3 is greater than the preset convolution value, that is, it is determined that there is a lesion area in image block No. 3, and image block No. 3 is added to the set of lesion image blocks. Then, the convolution value of image block No. 4 is equal to the preset convolution value, that is, it is determined that there is a lesion area in image block No. 4, and image block No. 4 is added to the set of lesion image blocks. The set of lesion image blocks includes image block No. 1, image block No. 3, and image block No. 4; marks the lesion area in each of the three lesion image blocks to generate a marked target image, and sends the marked target image to the medical staff terminal. Or convolves each of the multiple image blocks one by one in the order from 1 to 4, obtaining a convolution value of 0.3 for image block No. 1, a convolution value of 0.4 for image block No. 2, a convolution value of 0.3 for image block No. 3, and a convolution value of 0.2 for image block No. 4; if the preset convolution value is 0.5, then the convolution value of each image block is less than the preset convolution value, that is, it is determined that there is no lesion area in the target image, then a reminder message for reviewing the lesion recognition result of the target image is generated and sent to the medical staff terminal.

[0070] In an embodiment of the present invention, a target image is obtained, where the target image is a type-I image or a type-II image to be recognized. The type-I image is an in-vivo image or a surface image collected by a professional medical device, and the type-II image is a surface image collected by a non-professional medical device; a preset medical image reading model is called to perform image type detection on the target image to obtain the target image type; if the target image type is a type-I image, then lesion recognition is performed on the target image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal; if the target image type is a type-II image, then the target image is preprocessed to obtain a preprocessed type-II image, where the preprocessed type-II image is an image on which the preset medical image reading model can perform lesion recognition; lesion recognition is performed on the preprocessed type-II image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal, increasing the types of images for artificial intelligence medical image reading.

[0071] The intelligent medical image reading method in the embodiment of the present invention is described above. Next, the intelligent medical image reading device in the embodiment of the present invention is described. Please refer toFigure 3 , in one embodiment of the intelligent medical image reading device in the embodiments of the present invention, it includes:

[0072] An acquisition module 301, configured to acquire a target image, where the target image is a type-one image or a type-two image to be recognized. The type-one image is an in-vivo image or a surface image collected by a professional medical device, and the type-two image is a surface image collected by a non-professional medical device;

[0073] A detection module 302, configured to call a preset medical image reading model to perform image type detection on the target image to obtain the target image type;

[0074] A first recognition and sending module 303, configured to, if the target image type is a type-one image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal;

[0075] A preprocessing module 304, configured to, if the target image type is a type-two image, perform preprocessing on the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which the preset medical image reading model can perform lesion recognition;

[0076] A second recognition and sending module 305, configured to perform lesion recognition on the preprocessed type-two image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal.

[0077] In the embodiments of the present invention, a target image is acquired, where the target image is a type-one image or a type-two image to be recognized. The type-one image is an in-vivo image or a surface image collected by a professional medical device, and the type-two image is a surface image collected by a non-professional medical device; a preset medical image reading model is called to perform image type detection on the target image to obtain the target image type; if the target image type is a type-one image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal; if the target image type is a type-two image, perform preprocessing on the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which the preset medical image reading model can perform lesion recognition; perform lesion recognition on the preprocessed type-two image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal, which increases the types of images for artificial intelligence medical image reading.

[0078] Please refer to Figure 4 , another embodiment of the intelligent medical image reading device in the embodiments of the present invention includes:

[0079] An acquisition module 301 for acquiring a target image, where the target image is a type-one image or a type-two image to be recognized, the type-one image is an in-vivo image or a surface image collected by a professional medical device, and the type-two image is a surface image collected by a non-professional medical device;

[0080] A detection module 302 for calling a pre-set medical image reading model to perform image type detection on the target image to obtain the target image type;

[0081] A first recognition and sending module 303 for, if the target image type is a type-one image, performing lesion recognition on the target image, and if there is a lesion area in the target image, marking the lesion area in the target image and sending the marked image to a medical staff terminal;

[0082] A preprocessing module 304 for, if the target image type is a type-two image, performing preprocessing on the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which the pre-set medical image reading model can perform lesion recognition;

[0083] A second recognition and sending module 305 for performing lesion recognition on the preprocessed type-two image, and if there is a lesion area in the target image, marking the lesion area in the target image and sending the marked image to a medical staff terminal.

[0084] Optionally, the detection module 302 can also be specifically used for:

[0085] Performing image noise analysis on the target image through the pre-set medical image reading model to obtain an image noise analysis result;

[0086] If the image noise analysis result indicates that the target image has Poisson noise, determining the target image type corresponding to the target image as a type-one image;

[0087] If the image noise analysis result indicates that the target image has Gaussian noise, determining the target image type corresponding to the target image as a type-two image.

[0088] Optionally, the first recognition and sending module 303 includes:

[0089] A segmentation unit 3031 for, if the target image type is a type-one image, performing segmentation processing on the target image to obtain a target image area;

[0090] A determination unit 3032 for, if the pixel value of each pixel point in the target image area is greater than or equal to a pre-set pixel value, determining that there is a lesion area in the target image;

[0091] A marking unit 3033 for marking the lesion area in the target image and sending the marked image to a medical staff terminal.

[0092] Optionally, the first recognition and sending module 303 further includes:

[0093] A reminder unit 3034, configured to determine that there is no lesion area in the target image if the pixel value of each pixel point in the target image area is less than a preset pixel value, generate a reminder message, and send the reminder message to the medical staff terminal, where the reminder message is used to indicate a review of the lesion recognition result of the target image.

[0094] Optionally, the preprocessing module 304 may further specifically be configured to:

[0095] If the target image type is a type II image, perform noise reduction processing on the target image to obtain a noise-reduced target image;

[0096] Perform grayscale processing on the noise-reduced target image to obtain a grayscale target image;

[0097] Perform binarization processing on the grayscale target image to obtain a preprocessed type II image.

[0098] Optionally, the second recognition and sending module 305 may further specifically be configured to:

[0099] Perform image block segmentation on the preprocessed type II image to obtain a plurality of image blocks;

[0100] Perform convolution on each of the plurality of image blocks one by one according to a specified order to obtain the convolution value of each corresponding image block;

[0101] If the convolution value of the image block is greater than or equal to a preset convolution value, determine that there is a lesion area in the corresponding image block, and add the image block with the lesion area to a lesion image block set, where the lesion image block set includes a plurality of lesion image blocks;

[0102] Mark the lesion area in each lesion image block of the plurality of lesion image blocks to generate a marked target image, and send the marked target image to the medical staff terminal.

[0103] Optionally, the intelligent medical image reading device further includes:

[0104] A generation module 306, configured to obtain a type I image and a type II image of a historical patient, where there is a lesion area in the type I image and the type II image;

[0105] Extract the type I image to obtain a target lesion area in the type I image, and mark the target lesion area to obtain a marked type I image;

[0106] Extract the type II image to obtain a target lesion area in the type II image, and mark the target lesion area to obtain a marked type II image;

[0107] Model training is performed based on a marked first type of image and a marked second type of image to generate a preset medical image reading model.

[0108] In an embodiment of the present invention, a target image is obtained, where the target image is a first type of image or a second type of image to be recognized. The first type of image is an in-vivo image or a surface image collected by a professional medical device, and the second type of image is a surface image collected by a non-professional medical device. The preset medical image reading model is called to perform image type detection on the target image to obtain the target image type. If the target image type is a first type of image, lesion recognition is performed on the target image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal. If the target image type is a second type of image, preprocessing is performed on the target image to obtain a preprocessed second type of image, where the preprocessed second type of image is an image on which the preset medical image reading model can perform lesion recognition. Lesion recognition is performed on the preprocessed second type of image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal, increasing the types of images for artificial intelligence medical image reading.

[0109] Above Figure 3 And Figure 4 The intelligent medical image reading device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. Next, the intelligent medical image reading device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0110] Figure 5 FIG. is a schematic structural diagram of an intelligent medical image reading device provided by an embodiment of the present invention. The intelligent medical image reading device 500 may vary greatly due to configuration or performance differences, and may include one or more processors (central processing units, CPUs) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 (for example, one or more mass storage devices) for storing application programs 533 or data 532. Among them, the memory 520 and the storage media 530 may be transient storage or persistent storage. The program stored in the storage media 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the intelligent medical image reading device 500. Further, the processor 510 may be configured to communicate with the storage media 530 and execute a series of instruction operations in the storage media 530 on the intelligent medical image reading device 500.

[0111] The intelligent medical film reading device 500 may further include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input / output interfaces 560, and / or one or more operating systems 531, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 5 The structure of the illustrated intelligent medical film reading device does not constitute a limitation on the intelligent medical film reading device, and it may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0112] The present invention also provides an intelligent medical film reading device. The computer device includes a memory and a processor. When computer-readable instructions stored in the memory are executed by the processor, the processor is caused to execute the steps of the intelligent medical film reading method in the above-mentioned various embodiments.

[0113] The present invention also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the steps of the intelligent medical film reading method.

[0114] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0115] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0116] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent medical image reading method, characterized in that, The intelligent medical image reading method includes: Obtain a target image, where the target image is a type-one image or a type-two image to be recognized. The type-one image is an in-vivo image or a surface image collected by a professional medical device, and the type-two image is a surface image collected by a non-professional medical device; Call a pre-set medical image reading model to perform image type detection on the target image to obtain a target image type; If the target image type is the type-one image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal; for the lesion recognition of the target image, through a pixel-level recognition method, estimate the position of the pixel points whose pixel values are within a preset range in the target image as the lesion area, and the preset range of pixel values is set according to historical experience; or input the target image into a segmentation model trained to convergence, and determine the lesion area in the target image through the segmentation model; If the target image type is the type-two image, perform preprocessing on the target image to obtain a preprocessed type-two image, where the preprocessed type-two image is an image on which the pre-set medical image reading model can perform lesion recognition; Perform lesion recognition on the preprocessed type-two image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal; The calling of the pre-set medical image reading model to perform image type detection on the target image to obtain a target image type includes: Perform image noise analysis on the target image through the pre-set medical image reading model to obtain an image noise analysis result; If the image noise analysis result is that the target image has Poisson noise, determine the target image type corresponding to the target image as a type-one image; If the image noise analysis result is that the target image has Gaussian noise, determine the target image type corresponding to the target image as a type-two image.

2. The intelligent medical image reading method according to claim 1, wherein The "if the target image type is the type-one image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal" includes: If the target image type is the type-one image, perform segmentation processing on the target image to obtain a target image area; If the pixel value of each pixel point in the target image area is greater than or equal to the pre-set pixel value, determine that there is a lesion area in the target image; Mark the lesion area in the target image and send the marked image to the medical staff terminal.

3. The intelligent medical image reading method according to claim 2, wherein, After the "if the target image type is the type-one image, perform segmentation processing on the target image to obtain a target image area", it also includes: If the pixel value of each pixel point in the target image area is less than the preset pixel value, it is determined that there is no lesion area in the target image, and a reminder message is generated and sent to the medical staff terminal. The reminder message is used to indicate a review of the lesion recognition result of the target image.

4. The intelligent medical film reading method according to claim 1, wherein If the target image type is the second type of image, preprocessing is performed on the target image to obtain a preprocessed second type of image, including: If the target image type is the second type of image, noise reduction processing is performed on the target image to obtain a noise-reduced target image; Gray-scale processing is performed on the noise-reduced target image to obtain a gray-scale target image; Binary processing is performed on the gray-scale target image to obtain a preprocessed second type of image.

5. The intelligent medical image reading method according to claim 1, wherein Performing lesion recognition on the preprocessed second type of image. If there is a lesion area in the target image, the lesion area is marked in the target image and the marked image is sent to the medical staff terminal, including: Segment the preprocessed second type of image into image blocks to obtain a plurality of image blocks; Perform convolution on each of the plurality of image blocks one by one according to a specified order to obtain the convolution value of each corresponding image block; If the convolution value of an image block is greater than or equal to the preset convolution value, it is determined that there is a lesion area in the corresponding image block, and the image block with the lesion area is added to the lesion image block set. The lesion image block set includes a plurality of lesion image blocks; Mark the lesion area in each lesion image block of the plurality of lesion image blocks to generate a marked target image, and send the marked target image to the medical staff terminal.

6. The intelligent medical film reading method according to any one of claims 1-5, characterized in that Before obtaining the target image, it further includes: Obtain the first type of image and the second type of image of historical patients, where there are lesion areas in the first type of image and the second type of image; Extract the first type of image to obtain the target lesion area in the first type of image, and mark the target lesion area to obtain a marked first type of image; Extract the second type of image to obtain the target lesion area in the second type of image, and mark the target lesion area to obtain a marked second type of image; Perform model training according to the marked first type of image and the marked second type of image to generate a preset medical image reading model.

7. An intelligent medical film reading device, characterized in that, The intelligent medical image reading device includes: An acquisition module for acquiring a target image, where the target image is a first type of image or a second type of image to be recognized. The first type of image is an in-vivo image or a surface image collected by a professional medical device, and the second type of image is a surface image collected by a non-professional medical device; A detection module for calling a preset medical image reading model to perform image type detection on the target image to obtain a target image type; The first recognition and transmission module is configured to, if the target image type is the first type of image, perform lesion recognition on the target image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal. For the lesion recognition of the target image, through the recognition method at the pixel level, the position of the pixel points in the target image whose pixel values are within a preset range is estimated as the lesion area, and the preset range of the pixel values is set according to historical experience; or input the target image into a segmentation model trained to convergence, and determine the lesion area in the target image through the segmentation model. The preprocessing module is configured to, if the target image type is the second type of image, perform preprocessing on the target image to obtain a preprocessed second type of image, where the preprocessed second type of image is an image on which the preset medical image reading model can perform lesion recognition. The second recognition and transmission module is configured to perform lesion recognition on the preprocessed second type of image. If there is a lesion area in the target image, mark the lesion area in the target image and send the marked image to the medical staff terminal. The acquisition module is further configured to perform image noise analysis on the target image through a preset medical image reading model to obtain an image noise analysis result. If the image noise analysis result is that the target image has Poisson noise, the target image type corresponding to the target image is determined as the first type of image. If the image noise analysis result is that the target image has Gaussian noise, the target image type corresponding to the target image is determined as the second type of image.

8. An intelligent medical film reading device, characterized in that, The intelligent medical image reading device includes: a memory and at least one processor, and instructions are stored in the memory. The at least one processor calls the instructions in the memory to enable the intelligent medical image reading device to execute the intelligent medical image reading method according to any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, the intelligent medical image reading method according to any one of claims 1-6 is implemented.

Images