System and method for detecting a live tissue nodule

Abstract

The application relates to a system and method for detecting nodules of living tissue, the system comprising: an ultrasound image acquisition module generating ultrasound images of living tissue through examination of a patient; a positioning segmentation module in communication connection with the ultrasound image acquisition module, receiving ultrasound images of living tissue generated from the ultrasound image acquisition module, and after preprocessing the ultrasound images, positioning and segmenting nodules on the ultrasound images of living tissue through a Unet++ network and a Medical Transformer network loaded thereon to form one or more images of nodule regions to be examined; an artificial review module artificially judging each of the one or more images of nodule regions to be examined; and an output module outputting results of artificial judgment formed by the artificial review module. The system and method for detecting nodules of living tissue according to the application can be used for auxiliary diagnosis of thyroid cancer, for example.

Description

Technical Field

[0001] This application relates to the processing and analysis of ultrasound images, and in particular to a system and method for detecting nodules in living tissue. Background Technology

[0002] In recent years, the incidence of biopsy nodules, especially thyroid cancer, has been increasing, leading to higher demands for diagnostic accuracy and individualized treatment. Currently, the primary method for detecting thyroid cancer is through ultrasound imaging. Doctors rely on their experience to identify biopsy nodules from the ultrasound images and manually mark the nodules' long and short axes. This method is not only time-consuming and labor-intensive, but its accuracy also heavily depends on the doctor's experience. As the number of patients being examined increases, this will significantly deplete doctors' time and potentially lead to a decline in diagnostic accuracy.

[0003] The key to ultrasound-assisted diagnosis of thyroid cancer lies in locating and segmenting the lesion, thereby enabling focused analysis and judgment. Summary of the Invention

[0004] The technical problem to be solved by this application is to provide a system and method for detecting nodules in living tissue based on computer software processing and manual judgment.

[0005] To address the aforementioned technical problems, according to one aspect of this application, a system for detecting nodules in living tissue is provided, comprising: an ultrasound image acquisition module for generating ultrasound images of living tissue through examination of a patient; a localization and segmentation module communicatively connected to the ultrasound image acquisition module for receiving the ultrasound images of living tissue generated by the ultrasound image acquisition module, and, after preprocessing the ultrasound images, locating and segmenting nodules on the ultrasound images of living tissue using a Unet++ network and a Medical Transformer network loaded thereon, to form one or more nodule region images to be examined; a manual review module for manually judging each of the one or more nodule region images to be examined; and an output module for outputting the results of the manual judgment formed by the manual review module.

[0006] According to embodiments of this application, the system for detecting nodules in living tissue may further include a display module, which can display the process of the positioning and segmentation module locating and segmenting nodules on ultrasound images of living tissue, the process of the manual review module manually judging each of one or more nodule regions to be examined, and the result of the manual judgment output by the output module.

[0007] According to an embodiment of this application, the manual review module may include inputting a conclusive report of the manual judgment and instructing the output module to output the conclusive report of the manual judgment.

[0008] According to embodiments of this application, the system for detecting nodules in living tissue may further include an optimization module, which can receive a conclusive report of human judgment from the output module and continuously optimize the positioning and segmentation module through machine deep learning and training.

[0009] According to embodiments of this application, the communication connection between the positioning and segmentation module and the ultrasound image acquisition module may include one or both of wired and wireless connections.

[0010] According to embodiments of this application, the positioning and segmentation module, the manual review module, and the output module can be integrated into a single PC. Preferably, the positioning and segmentation module, the manual review module, the output module, the display module, and the optimization module can be integrated into a single PC.

[0011] According to embodiments of this application, preprocessing of ultrasound images may include removing parts unrelated to the ultrasound image and adjusting image attributes such as image size and grayscale, with the image size adjusted to 256*256.

[0012] According to an embodiment of this application, the Unet++ network can be a coarse segmentation network that calculates the minimum bounding square of the node. If the size of the bounding square is greater than 80, then it is expanded by 20 pixels; otherwise, it is expanded by 30 pixels to scale the truncated node to a size of 512*512.

[0013] According to embodiments of this application, the Medical Transformer network combines convolutional neural networks and Transformer modules, and may include a local training module, a global training module, and an integration module.

[0014] According to embodiments of this application, the local training module can use a convolution module to downsample the input model twice and then upsample it to obtain local features of the image; the global training module can use a Transformer module to increase the receptive field of the network to obtain global features of the image; the integration module can superimpose global features and local features to obtain the final segmentation result.

[0015] According to embodiments of this application, the system for detecting nodules in living tissue may further include using the Dice coefficient combined with a binary cross-entropy function as a loss function for training in cases of uneven class division.

[0016] According to embodiments of this application, the ultrasound image acquisition module may include a conventional B-ultrasound machine, and the living tissue nodule may include a thyroid nodule.

[0017] According to another aspect of this application, a method for detecting nodules in living tissue is provided using a system for detecting nodules in living tissue. The system for detecting nodules in living tissue includes an ultrasound image acquisition module, a localization and segmentation module, a manual review module, and an output module. The method for detecting nodules in living tissue includes the following steps: generating an ultrasound image of living tissue using the ultrasound image acquisition module through an examination of a patient; receiving the ultrasound image of living tissue generated by the ultrasound image acquisition module using a localization and segmentation module communicatively connected to the ultrasound image acquisition module, and, after preprocessing the ultrasound image, locating and segmenting nodules on the ultrasound image of living tissue using a Unet++ network and a Medical Transformer network loaded thereon to form one or more nodule regions to be examined; manually judging each of the one or more nodule regions to be examined using the manual review module; and outputting the result of the manual judgment formed by the manual review module using the output module.

[0018] According to embodiments of this application, the system for detecting nodules in living tissue may further include a display module, and the method for detecting nodules in living tissue may further include the process of using the display module to display the process of the positioning and segmentation module locating and segmenting nodules on the ultrasound image of the living tissue, the process of the manual review module manually judging each of one or more images of nodule regions to be checked, and the result of the manual judgment output by the output module.

[0019] According to an embodiment of this application, the manual review module may include inputting a conclusive report of the manual judgment and instructing the output module to output the conclusive report of the manual judgment.

[0020] According to embodiments of this application, the system for detecting nodules in living tissue may further include an optimization module, and the method for detecting nodules in living tissue may further include receiving a conclusive report of human judgment from the output module using the optimization module, and continuously optimizing the positioning and segmentation module through machine deep learning and training.

[0021] According to embodiments of this application, the communication connection between the positioning and segmentation module and the ultrasound image acquisition module may include one or both of wired and wireless connections.

[0022] According to embodiments of this application, the positioning and segmentation module, the manual review module, and the output module can be integrated into a single PC. Preferably, the positioning and segmentation module, the manual review module, the output module, the display module, and the optimization module can be integrated into a single PC.

[0023] According to embodiments of this application, preprocessing of ultrasound images may include removing parts unrelated to the ultrasound image and adjusting image attributes such as image size and grayscale, with the image size adjusted to 256*256.

[0024] According to an embodiment of this application, the Unet++ network can be a coarse segmentation network that calculates the minimum bounding square of the node. If the size of the bounding square is greater than 80, then it is expanded by 20 pixels; otherwise, it is expanded by 30 pixels to scale the truncated node to a size of 512*512.

[0025] According to embodiments of this application, the Medical Transformer network combines convolutional neural networks and Transformer modules, and may include a local training module, a global training module, and an integration module.

[0026] According to embodiments of this application, the local training module can use a convolution module to downsample the input model twice and then upsample it to obtain local features of the image; the global training module can use a Transformer module to increase the receptive field of the network to obtain global features of the image; the integration module can superimpose global features and local features to obtain the final segmentation result.

[0027] According to embodiments of this application, the system for detecting nodules in living tissue may further include using the Dice coefficient combined with a binary cross-entropy function as a loss function for training in cases of uneven class division.

[0028] According to embodiments of this application, the ultrasound image acquisition module may include a conventional B-ultrasound machine, and the living tissue nodule may include a thyroid nodule.

[0029] Compared with the prior art, the system and method for detecting nodules in living tissue according to the embodiments of this application can achieve at least the following beneficial effects:

[0030] Because the system and method for detecting nodules in living tissue according to the embodiments of this application employ an ultrasound image acquisition module to generate ultrasound images of living tissue through examination of the patient, and the ultrasound image acquisition module can be a conventional B-ultrasound machine, no changes need to be made to the traditional B-ultrasound machine, thus eliminating the cost of modifying existing equipment.

[0031] The positioning and segmentation module is communicatively connected to the ultrasound image acquisition module, and this communication connection can be wired or wireless. Thus, the processing and analysis components, represented by the positioning and segmentation module, can be placed adjacent to the ultrasound machine, or the two can be remotely connected. In the case of a remote connection, this facilitates telemedicine diagnosis.

[0032] After preprocessing the ultrasound images, the localization and segmentation module uses the Unet++ network and Medical Transformer network loaded on it to locate and segment nodules on the ultrasound images of living tissue, forming one or more images of nodule regions to be investigated. Each of these images is worthy of further analysis, thus ensuring that no potential lesions are missed.

[0033] The manual review module performs a manual assessment of each of the images of one or more nodule regions to be examined. The images of one or more nodule regions to be examined generated by the localization and segmentation module are highly recognizable to the manual review module or the sonographer, thus enabling a relatively accurate assessment. This significantly improves detection efficiency and accuracy.

[0034] The output module outputs the results of the human review module's judgment, as well as a conclusive report. These results and reports can be printed or fed back to the optimization module for continuous training, in-depth learning, and optimization of the software in the positioning and segmentation module.

[0035] The system and method for detecting nodules in living tissue according to embodiments of this application can provide rapid and accurate auxiliary diagnostic results, and therefore have good medical value and great social value. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only relate to some embodiments of this application, and are not intended to limit this application.

[0037] Figure 1 This is a block diagram of a system for detecting nodules in living tissue according to an embodiment of this application.

[0038] Figure 2 This is a segmentation network flowchart of a system for detecting nodules in living tissue according to an embodiment of this application.

[0039] Figure 3 This is a Unet++ network model diagram of a system for detecting nodules in living tissue according to an embodiment of this application.

[0040] Figure 4 This is a MedicalTransformer network model structure diagram of a system for detecting nodules in living tissue according to an embodiment of this application.

[0041] Figure 5 This is a flowchart of a method for detecting nodules in living tissue according to an embodiment of this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the described embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” and similar terms used in the specification and claims of this patent application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a,” and similar terms, do not indicate a limitation of quantity, but rather indicate the presence of at least one.

[0044] The embodiments of this application are described below with reference to the accompanying drawings.

[0045] Figure 1 This is a block diagram of a system for detecting nodules in living tissue according to an embodiment of this application.

[0046] See Figure 1 According to an embodiment of this application, a system for detecting nodules in living tissue is provided, including an ultrasound image acquisition module, a positioning and segmentation module, a manual review module, and an output module, which are described below.

[0047] The ultrasound image acquisition module generates ultrasound images of living tissue through examination of a patient. According to embodiments of this application, the ultrasound image acquisition module may be, for example, a conventional B-mode ultrasound machine. However, embodiments of this application are not limited to this, and the ultrasound image acquisition module may also be any other device that acquires conventional two-dimensional ultrasound images of living tissue at a predetermined location using conventional B-mode ultrasound, such as A-mode ultrasound, M-mode ultrasound, D-mode ultrasound, color Doppler ultrasound, etc.

[0048] For the ultrasound image acquisition module, common examinations include thyroid ultrasound, cervical lymph node ultrasound, female breast ultrasound, abdominal ultrasound, liver ultrasound, and prostate ultrasound. The following description uses thyroid ultrasound as an example.

[0049] The localization and segmentation module is communicatively connected to the ultrasound image acquisition module. It receives ultrasound images of living tissue generated by the ultrasound image acquisition module, and after preprocessing the ultrasound images, it locates and segments the nodules on the ultrasound images of living tissue through the Unet++ network and MedicalTransformer network loaded on it, so as to form one or more images of nodule regions to be investigated.

[0050] Figure 2 This is a segmentation network flowchart of a system for detecting nodules in living tissue according to an embodiment of this application. Figure 3 This is a Unet++ network model diagram of a system for detecting nodules in living tissue according to an embodiment of this application. Figure 4 This is a Medical Transformer network model structure diagram of a system for detecting nodules in living tissue according to an embodiment of this application.

[0051] like Figure 2 and 3 As shown, the uploaded thyroid ultrasound image is first preprocessed to remove parts irrelevant to the ultrasound image and adjust the image size to 256*256. The ultrasound image is then segmented in the first stage using a segmentation model. The network in the first stage is a coarse segmentation network, which aims to obtain the location and approximate shape information of the thyroid nodules so as to narrow the field of interest in the next fine segmentation stage, in order to obtain better segmentation results.

[0052] After obtaining the coarse segmentation result, the minimum bounding square of the nodule is calculated. If the bounding square size is greater than 80, it is expanded by 20 pixels; otherwise, it is expanded by 30 pixels. The truncated nodule is scaled to 512*512 pixels and used as input to the two-stage segmentation network. The two-stage network is a fine segmentation network.

[0053] The coarse segmentation network Unet++ of this application is described in detail below:

[0054] This network employs the Unet++ convolutional neural network, an improved version of the classic segmentation network Unet. Compared to Unet, it fills the gaps in the middle of the Unet network with multiple smaller Unet structures, thus enhancing the network's learning ability. The input image undergoes four convolutional and downsampling processes to capture features at different levels. Subsequently, each level of the network performs multiple upsampling operations until it is restored to its original size. Within each convolutional node, features at the same level are passed to subsequent convolutional modules at the same level through skip-connect operations to enrich the features at each level and avoid gradient vanishing. After network training, 1x1 convolutional modules can be added after X12, X13, X14, and X15 to supervise the output and capture the output of each branch. Depending on the learning capacity, different degrees of pruning can be performed to reduce the final model size and increase its portability without significantly impacting network performance.

[0055] like Figure 4As shown, the fine segmentation network in this application adopts a Medical Transformer network, which is described in detail below:

[0056] This network combines convolutional neural networks (CNNs) and Transformer modules. The model employs a local-global training strategy, with two parallel training networks simultaneously learning and extracting image features. The top module (local training) uses a convolutional layer to downsample the input model twice, followed by upsampling, thus capturing local image features. The bottom module (global training) uses a Transformer module to increase the network's receptive field. It first uses a convolutional layer for feature extraction, then the Transformer module integrates and refines the features. The decoder uses a cascaded upsampling process with four upsampling steps to decode hidden features and output the final segmentation result. Each layer in the decoder uses a skip-connect operation, adding Transformer features and allowing features from different levels to be better blended, improving model accuracy. Finally, the results from the global and local branches are summed to obtain the final segmentation result.

[0057] The training process employed a hybrid approach using Dice coefficient and cross-entropy, which not only focused on the target region but also addressed the issue of class imbalance (more background and smaller nodules).

[0058] In summary:

[0059] According to embodiments of this application, preprocessing of ultrasound images may include removing parts unrelated to the ultrasound image and adjusting image attributes such as image size and grayscale, with the image size adjusted to 256*256.

[0060] According to an embodiment of this application, the Unet++ network can be a coarse segmentation network that calculates the minimum bounding square of the node. If the size of the bounding square is greater than 80, then it is expanded by 20 pixels; otherwise, it is expanded by 30 pixels to scale the truncated node to a size of 512*512.

[0061] According to embodiments of this application, the Medical Transformer network combines convolutional neural networks and Transformer modules, and may include a local training module, a global training module, and an integration module.

[0062] According to embodiments of this application, the local training module can use a convolution module to downsample the input model twice and then upsample it to obtain local features of the image; the global training module can use a Transformer module to increase the receptive field of the network to obtain global features of the image; the integration module can superimpose global features and local features to obtain the final segmentation result.

[0063] Come back and see Figure 1 The output module outputs the results of the manual review by the manual review module.

[0064] According to embodiments of this application, the system for detecting nodules in living tissue may further include a display module, which can display the process of the positioning and segmentation module locating and segmenting nodules on ultrasound images of living tissue, the process of the manual review module manually judging each of one or more nodule regions to be examined, and the result of the manual judgment output by the output module.

[0065] According to an embodiment of this application, the manual review module may include inputting a conclusive report of the manual judgment and instructing the output module to output the conclusive report of the manual judgment.

[0066] According to embodiments of this application, the system for detecting nodules in living tissue may further include an optimization module, which can receive a conclusive report of human judgment from the output module and continuously optimize the positioning and segmentation module through machine deep learning and training.

[0067] According to embodiments of this application, the communication connection between the positioning and segmentation module and the ultrasound image acquisition module may include one or both of wired and wireless connections.

[0068] According to embodiments of this application, the positioning and segmentation module, the manual review module, and the output module can be integrated into a single PC. Preferably, the positioning and segmentation module, the manual review module, the output module, the display module, and the optimization module can be integrated into a single PC.

[0069] According to embodiments of this application, the system for detecting nodules in living tissue may further include using the Dice coefficient combined with a binary cross-entropy function as a loss function for training in cases of uneven class division.

[0070] According to embodiments of this application, the ultrasound image acquisition module may include a conventional B-ultrasound machine, and the living tissue nodule may include a thyroid nodule.

[0071] Figure 5 This is a flowchart of a method for detecting nodules in living tissue according to an embodiment of this application.

[0072] See Figure 3According to embodiments of this application, a method for detecting nodules in living tissue is provided using a system for detecting nodules in living tissue. In the following description, references will be omitted to avoid redundancy. Figure 1-2 The description of a system for detecting nodules in living tissue according to embodiments of this application is repeated.

[0073] The system for detecting nodules in living tissue includes an ultrasound image acquisition module, a localization and segmentation module, a manual review module, and an output module. Preferably, it may also include a display module and an optimization module.

[0074] A method for detecting nodules in living tissue includes the following steps: an ultrasound image acquisition module generates ultrasound images of living tissue through examination of a patient; a positioning and segmentation module, communicatively connected to the ultrasound image acquisition module, receives the ultrasound images of living tissue generated by the ultrasound image acquisition module, and after preprocessing the ultrasound images, locates and segments the nodules on the ultrasound images of living tissue using a Unet++ network and a Medical Transformer network loaded on it to form one or more images of nodule regions to be examined; a manual review module performs manual judgment on each of the one or more images of nodule regions to be examined; and an output module outputs the results of the manual judgment formed by the manual review module.

[0075] According to embodiments of this application, the method for detecting nodules in living tissue may further include a process of locating and segmenting nodules on ultrasound images of living tissue using a display module to display the positioning and segmentation module, a process of manually judging each of one or more nodule regions in an image by a manual review module, and the result of the manual judgment output by an output module.

[0076] According to an embodiment of this application, the manual review module may include inputting a conclusive report of the manual judgment and instructing the output module to output the conclusive report of the manual judgment.

[0077] According to embodiments of this application, the method for detecting nodules in living tissue may further include receiving a conclusive report of human judgment from an output module using an optimization module, and continuously optimizing the positioning and segmentation module through machine deep learning and training.

[0078] According to embodiments of this application, the communication connection between the positioning and segmentation module and the ultrasound image acquisition module may include one or both of wired and wireless connections.

[0079] According to embodiments of this application, the positioning and segmentation module, the manual review module, and the output module can be integrated into a single PC. Preferably, the positioning and segmentation module, the manual review module, the output module, the display module, and the optimization module can be integrated into a single PC.

[0080] According to embodiments of this application, preprocessing of ultrasound images may include removing parts unrelated to the ultrasound image and adjusting image attributes such as image size and grayscale, with the image size adjusted to 256*256.

[0081] According to an embodiment of this application, the Unet++ network can be a coarse segmentation network that calculates the minimum bounding square of the node. If the size of the bounding square is greater than 80, then it is expanded by 20 pixels; otherwise, it is expanded by 30 pixels to scale the truncated node to a size of 512*512.

[0082] According to embodiments of this application, the Medical Transformer network combines convolutional neural networks and Transformer modules, and may include a local training module, a global training module, and an integration module.

[0083] According to embodiments of this application, the local training module can use a convolution module to downsample the input model twice and then upsample it to obtain local features of the image; the global training module can use a Transformer module to increase the receptive field of the network to obtain global features of the image; the integration module can superimpose global features and local features to obtain the final segmentation result.

[0084] According to embodiments of this application, the system for detecting nodules in living tissue may further include using the Dice coefficient combined with a binary cross-entropy function as a loss function for training in cases of uneven class division.

[0085] According to embodiments of this application, the ultrasound image acquisition module may include a conventional B-ultrasound machine, and the living tissue nodule may include a thyroid nodule.

[0086] Compared with the prior art, the system and method for detecting nodules in living tissue according to the embodiments of this application can achieve at least the following beneficial effects:

[0087] Because the system and method for detecting nodules in living tissue according to the embodiments of this application employ an ultrasound image acquisition module to generate ultrasound images of living tissue through examination of the patient, and the ultrasound image acquisition module can be a conventional B-ultrasound machine, no changes need to be made to the traditional B-ultrasound machine, thus eliminating the cost of modifying existing equipment.

[0088] The positioning and segmentation module is communicatively connected to the ultrasound image acquisition module, and this communication connection can be wired or wireless. Thus, the processing and analysis components, represented by the positioning and segmentation module, can be placed adjacent to the ultrasound machine, or the two can be remotely connected. In the case of a remote connection, this facilitates telemedicine diagnosis.

[0089] After preprocessing the ultrasound images, the localization and segmentation module uses the Unet++ network and Medical Transformer network loaded on it to locate and segment nodules on the ultrasound images of living tissue, forming one or more images of nodule regions to be investigated. Each of these images is worthy of further analysis, thus ensuring that no potential lesions are missed.

[0090] The manual review module performs a manual assessment of each of the images of one or more nodule regions to be examined. The images of one or more nodule regions to be examined generated by the localization and segmentation module are highly recognizable to the manual review module or the sonographer, thus enabling a relatively accurate assessment. This significantly improves detection efficiency and accuracy.

[0091] The output module outputs the results of the human review module's judgment, as well as a conclusive report. These results and reports can be printed or fed back to the optimization module for continuous training, in-depth learning, and optimization of the software in the positioning and segmentation module.

[0092] The system and method for detecting nodules in living tissue according to embodiments of this application can provide rapid and accurate auxiliary diagnostic results, and therefore have good medical value and great social value.

[0093] The above description is merely an exemplary embodiment of this application and is not intended to limit the scope of protection of this application. The scope of protection of this application is determined by the appended claims.

Claims

1. A system for detecting nodules in living tissue, comprising: The ultrasound image acquisition module, which is a conventional B-ultrasound or color Doppler ultrasound, generates ultrasound images of living tissues through examination of the patient; The localization and segmentation module is communicatively connected to the ultrasound image acquisition module. It receives ultrasound images of living tissue generated by the ultrasound image acquisition module. After preprocessing the ultrasound images, it locates and segments the nodules on the ultrasound images of the living tissue through the Unet++ network and Medical Transformer network loaded on it, so as to form one or more images of nodule regions to be investigated. The manual review module performs manual judgment on each of the one or more nodule region images to be checked formed on the positioning and segmentation module; The output module outputs the results of the manual judgment generated by the manual review module. as well as The optimization module receives the conclusive report of the human judgment from the output module and continuously optimizes the positioning and segmentation module through machine deep learning and training. as well as The display module shows the process of the positioning and segmentation module locating and segmenting nodules on the ultrasound image of the living tissue, the process of the manual review module manually judging each of one or more nodule regions in the image, and the result of the manual judgment output by the output module. A method for detecting nodules in living tissue includes the following steps: generating ultrasound images of living tissue using an ultrasound image acquisition module through examination of a patient; receiving the ultrasound images of living tissue generated by the ultrasound image acquisition module using a positioning and segmentation module communicatively connected to the ultrasound image acquisition module; and, after preprocessing the ultrasound images, locating and segmenting the nodules on the ultrasound images of living tissue using a Unet++ network and a Medical Transformer network loaded on the image to form one or more images of nodule regions to be examined. The system employs a manual review module to manually assess each of the one or more nodule regions formed by the positioning and segmentation module; an output module outputs the results of the manual assessments; a display module shows the process of the positioning and segmentation module locating and segmenting nodules on the ultrasound image of the living tissue, the process of the manual review module manually assessing each of the one or more nodule regions, and the results of the manual assessments output by the output module; and an optimization module receives the conclusive report of the manual assessments from the output module, and continuously optimizes the positioning and segmentation module and the manual review module through machine deep learning and training. The manual review module includes inputting a conclusive report on the manual judgment and instructing the output module to output the conclusive report on the manual judgment. The Unet++ network is a coarse segmentation network that calculates the minimum bounding square of the nodule. If the size of the bounding square is greater than 80, it is expanded by 20 pixels; otherwise, it is expanded by 30 pixels to scale the truncated nodule to a size of 512*512. The Medical Transformer network employs a combination of convolutional neural networks and Transformer modules, including a local training module, a global training module, and an integration module; and The local training module uses a convolution module to downsample the input model twice and then upsample it to obtain local features of the image; the global training module uses a Transformer module to increase the receptive field of the network to obtain global features of the image; the integration module superimposes the global features and local features to obtain the final segmentation result.

2. The system for detecting a live tissue nodule of claim 1, wherein, The communication connection between the positioning and segmentation module and the ultrasound image acquisition module includes one or both of wired and wireless connections.

3. The system for detecting a live tissue nodule of claim 1, wherein, The positioning and segmentation module, the manual review module, and the output module are integrated into a single PC.

4. The system for detecting a live tissue nodule of claim 1, wherein, The positioning and segmentation module, the manual review module, the output module, the display module, and the optimization module are integrated into one PC.

5. The system for detecting a live tissue nodule of claim 1, wherein, The preprocessing of the ultrasound image includes removing parts that are not related to the ultrasound image and adjusting the image size and grayscale image attributes, with the image size adjusted to 256*256.

6. The system for detecting nodules in living tissue as described in claim 1 further includes using the Dice coefficient combined with a binary cross-entropy function as a loss function for training in cases of uneven class segmentation.

Images