阅读说明：本技术 用于检测活体组织结节的系统和方法 (System and method for detecting in vivo tissue nodules ) 是由 罗全勇 沈晨天 王韧 丁雪海 于 2021-09-14 设计创作，主要内容包括：本申请涉及一种用于检测活体组织结节的系统和方法,所述系统包括：超声图获取模块,通过对患者的检查生成活体组织的超声图像；定位分割模块,与超声图获取模块通信连接,接收来自超声图获取模块生成的活体组织的超声图像,并且在对超声图像进行预处理后,通过其上加载的Unet++网络和Medical Transformer网络对活体组织的超声图像上的结节进行定位和分割,以形成一个或多个待查结节区域图像；人工审核模块,对一个或多个待查结节区域图像的每一个进行人工判断；以及输出模块,输出人工审核模块形成的人工判断的结果。根据本申请的用于检测活体组织结节的系统和方法例如可用于甲状腺癌的辅助诊断。(The present application relates to a system and method for detecting a nodule in living tissue, the system comprising: an ultrasound image acquisition module that generates an ultrasound image of a living tissue through examination of a patient; the positioning segmentation module is in communication connection with the ultrasonic image acquisition module, receives the ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positions and segments nodules on the ultrasonic image of the living tissue through a Unet + + network and a Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed so as to form one or more images of the nodule area to be examined; the manual auditing module is used for manually judging each of the one or more to-be-checked node area images; and the output module is used for outputting the result of the manual judgment formed by the manual checking module. The system and method for detecting a nodule in living tissue according to the present application may be used, for example, for the assisted diagnosis of thyroid cancer.)

1. A system for detecting a nodule in living tissue, comprising:

an ultrasound image acquisition module that generates an ultrasound image of a living tissue through examination of a patient;

the positioning segmentation module is in communication connection with the ultrasonic image acquisition module, receives an ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positions and segments a nodule on the ultrasonic image of the living tissue through a Unet + + network and a Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed so as to form one or more images of a nodule area to be examined;

the manual checking module is used for manually judging each of one or more to-be-checked node area images formed on the positioning segmentation module; and

and the output module is used for outputting the result of the manual judgment formed by the manual auditing module.

2. The system for detecting a nodule in living tissue of claim 1, further comprising a display module displaying the process of the localization and segmentation module locating and segmenting the nodule on the ultrasound image of the living tissue, the process of the manual review module manually judging each of the one or more images of the nodule area to be examined, and the result of the manual judgment output by the output module.

3. The system for detecting a nodule in living tissue of claim 2, wherein the manual review module includes inputting a conclusive report on the manual determination and instructing the output module to output the conclusive report on the manual determination.

4. A system for detecting in vivo tissue nodules as in claim 3, further comprising an optimization module that receives the human judged conclusive report from the output module and continuously optimizes the localized segmentation module through machine in-depth learning and training.

5. The system for detecting a nodule in living tissue of claim 1, wherein the communicative connection of the localization segmentation module to the sonogram acquisition module includes one or both of a wired connection and a wireless connection.

6. The system for detecting a nodule in living tissue of claim 1, wherein the localized segmentation module, the manual review module and the output module are integrated in one PC.

7. The system for detecting a nodule in living tissue of claim 4, wherein the localized segmentation module, the manual review module, the output module, the display module and the optimization module are integrated in one PC.

8. The system for detecting a nodule in living tissue of claim 1, wherein the pre-processing of the ultrasound image includes removing portions not associated with the ultrasound image and adjusting image attributes such as image size and gray scale to 256 x 256.

9. The system for detecting a nodule in living tissue of claim 1, wherein the Unet + + network is a coarsely segmented network, computing the smallest circumscribed square of the nodule, and if the circumscribed square size is greater than 80, then extrapolating 20 pixels, otherwise extrapolating 30 pixels to scale the truncated nodule to 512 x 512 size.

10. The system for detecting a nodule in living tissue of claim 1, wherein the Medical transducer network employs a convolutional neural network in combination with a transducer module, including a local training module, a global training module, and an integration module.

11. The system for detecting a nodule in living tissue of claim 10, wherein the local training module uses a convolution module to downsample the input model twice and then upsample to obtain local features of the image; the global training module uses a Transformer module to increase the receptive field of the network so as to obtain the global characteristics of the image; and the integration module superposes the global features and the local features to obtain a final segmentation result.

12. A system for detecting a nodule in living tissue as recited in claim 1, further comprising using a Dice coefficient in combination with a binary cross-entropy function as a trained loss function for cases of class inhomogeneity in the segmentation.

13. A system for detecting a biopsy nodule as set forth in claim 1, wherein the sonogram acquisition module includes a conventional B-mode ultrasound machine and the biopsy nodule includes a thyroid nodule.

14. A method for detecting a nodule of living tissue by a system for detecting a nodule of living tissue, the system comprising an sonogram acquisition module, a localization segmentation module, a manual review module, and an output module, the method comprising the steps of: generating an ultrasound image of the living tissue by examination of the patient using an ultrasound image acquisition module; adopting a positioning segmentation module which is in communication connection with the ultrasonic image acquisition module to receive an ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positioning and segmenting nodules on the ultrasonic image of the living tissue through a Unet + + network and a Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed so as to form one or more images of a nodule area to be examined; adopting a manual checking module to manually judge each of one or more to-be-checked node area images formed on the positioning segmentation module; and outputting the result of the manual judgment formed by the manual auditing module by adopting an output module.

15. The method for detecting a nodule in living tissue of claim 14, wherein the system further comprises a display module, the method further comprising displaying with the display module the procedure of the localization and segmentation module for localization and segmentation of the nodule on the ultrasound image of living tissue, the procedure of the manual review module for manual determination of each of the one or more images of the nodule region to be reviewed, and the result of the manual determination output by the output module.

16. The method for detecting a nodule in living tissue of claim 15, wherein the manual review module includes inputting a conclusive report on the manual determination and instructing the output module to output the conclusive report on the manual determination.

17. The method for detecting a nodule in living tissue as set forth in claim 16, wherein the system further includes an optimization module, the method further including employing the optimization module to receive a conclusive report of the manual determination from the output module and continuously optimizing the location segmentation module and the manual review module through machine in-depth learning and training.

18. The method for detecting a nodule in living tissue of claim 14, wherein the communicative connection of the localization segmentation module to the sonogram acquisition module includes one or both of a wired connection and a wireless connection.

19. The method for detecting a nodule in living tissue of claim 14, wherein the localized segmentation module, the manual review module and the output module are integrated in one PC.

20. The method for detecting a nodule in living tissue of claim 17, wherein the localized segmentation module, the manual review module, the output module, the display module and the optimization module are integrated in one PC.

21. The method for detecting a nodule in living tissue of claim 14, wherein the preprocessing of the ultrasound image includes removing portions not associated with the ultrasound image and resizing the image to 256 x 256 image attributes, such as image size and gray scale.

22. The method for detecting a nodule in living tissue of claim 14, wherein the Unet + + network is a coarsely segmented network, the smallest circumscribed square of the nodule is calculated, 20 pixels are extrapolated if the circumscribed square size is greater than 80, and 30 pixels are extrapolated otherwise to scale the truncated nodule to 512 x 512 size.

23. The method for detecting a nodule in living tissue of claim 14, wherein the Medical fransformer network employs a convolutional neural network in combination with a fransformer module, including a local training module, a global training module, and an integration module.

24. The method for detecting a nodule in living tissue of claim 23, wherein the local training module uses a convolution module to downsample the input model twice and then upsample to obtain local features of the image; the global training module uses a Transformer module to increase the receptive field of the network so as to obtain the global characteristics of the image; and the integration module superposes the global features and the local features to obtain a final segmentation result.

25. The method for detecting a nodule in living tissue of claim 14, further comprising using a Dice coefficient in combination with a binary cross-entropy function as a trained loss function for the case of class inhomogeneity in the segmentation.

26. A method for detecting a biopsy nodule as set forth in claim 14, wherein the sonogram acquisition module includes a conventional B-mode ultrasound machine and the biopsy nodule includes a thyroid nodule.

Technical Field

The present application relates to the processing and analysis of ultrasound images, and more particularly to a system and method for detecting nodules in living tissue.

Background

In recent years, the incidence of nodules in living tissues, particularly thyroid cancer, tends to increase, and higher requirements on the diagnosis accuracy and treatment individuation are provided. At present, the detection means of thyroid cancer is mainly to perform B-ultrasonic imaging by doctors, judge the nodules of the living tissues from the B-ultrasonic imaging according to self experience, and manually mark the long and short diameters of the nodules. The method is time-consuming and labor-consuming, the accuracy degree of the method greatly depends on the judgment of the experience of doctors, and the experience of the doctors is greatly consumed along with the increasing number of detected people, so that the diagnosis accuracy rate is possibly reduced.

The key point of the B-ultrasonic image aided diagnosis of thyroid cancer is to position and segment the focus so as to analyze and judge the focus position.

Disclosure of Invention

The technical problem to be solved by the application is to provide a system and a method for detecting a biopsy nodule based on computer software processing and manual judgment.

To solve the above technical problem, according to an aspect of the present application, there is provided a system for detecting a nodule of living tissue, including: an ultrasound image acquisition module that generates an ultrasound image of a living tissue through examination of a patient; the positioning segmentation module is in communication connection with the ultrasonic image acquisition module, receives the ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positions and segments nodules on the ultrasonic image of the living tissue through a Unet + + network and a Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed so as to form one or more images of the nodule area to be examined; the manual auditing module is used for manually judging each of the one or more to-be-checked node area images; and the output module is used for outputting the result of the manual judgment formed by the manual checking module.

According to an embodiment of the application, the system for detecting a nodule of living tissue may further include a display module, and the display module may display a process of positioning and segmenting the nodule on the ultrasound image of the living tissue by the positioning and segmenting module, a process of manually judging each of one or more images of the nodule area to be checked by the manual auditing module, and a result of the manual judgment output by the output module.

According to an embodiment of the 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.

According to an embodiment of the present application, the system for detecting a nodule in living tissue may further comprise an optimization module that may receive a conclusive report of human judgment from the output module and continuously optimize the localization segmentation module through machine in-depth learning and training.

According to embodiments of the application, the communicative connection of the localization segmentation module to the sonogram acquisition module may include one or both of a wired connection and a wireless connection.

According to the embodiment of the application, the positioning and dividing module, the manual auditing module and the output module can be integrated in one PC. Preferably, the positioning and dividing module, the manual review module, the output module, the display module and the optimization module can be integrated into one PC.

According to an embodiment of the present application, the preprocessing of the ultrasound image may include removing portions unrelated to the ultrasound image and adjusting image attributes such as image size and gray scale to 256 × 256.

According to an embodiment of the present application, the Unet + + network may be a coarsely partitioned network, computing the smallest circumscribed square of the nodule, and if the circumscribed square size is greater than 80, then scaling 20 pixels out, otherwise scaling 30 pixels out to scale the truncated nodule to 512 × 512 size.

According to the embodiment of the application, the Medical Transformer network adopts a convolutional neural network and a Transformer module which are combined, and can comprise a local training module, a global training module and an integration module.

According to the embodiment of the application, the local training module can use a convolution module to perform up-sampling after performing down-sampling twice on an input model so as to obtain the local features of the image; the global training module can use a Transformer module to increase the receptive field of the network so as to obtain the global characteristics of the image; the integration module can superpose the global features and the local features to obtain a final segmentation result.

According to an embodiment of the present application, the system for detecting a nodule in living tissue may further comprise using a Dice coefficient in combination with a binary cross-entropy function as a trained loss function for cases of class inhomogeneity in the segmentation.

According to embodiments of the application, the ultrasound image acquisition module may comprise a conventional B-ultrasound machine, and the biopsy nodule may comprise a thyroid nodule.

According to another aspect of the present application, there is provided a method for detecting a nodule of living tissue by a system for detecting a nodule of living tissue, the system for detecting a nodule of living tissue comprising an ultrasound image acquisition module, a localization segmentation module, a manual review module and an output module, the method for detecting a nodule of living tissue comprising the steps of: generating an ultrasound image of the living tissue by examination of the patient using an ultrasound image acquisition module; adopting a positioning segmentation module which is in communication connection with the ultrasonic image acquisition module to receive an ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positioning and segmenting nodules on the ultrasonic image of the living tissue through a Unet + + network and a Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed to form one or more images of the nodule area to be examined; adopting a manual auditing module to carry out manual judgment on each of one or more to-be-checked node area images; and outputting the result of the manual judgment formed by the manual auditing module by adopting an output module.

According to an embodiment of the present application, the system for detecting a nodule of living tissue may further include a display module, and the method for detecting a nodule of living tissue may further include displaying, by the display module, a process of positioning and segmenting the nodule on the ultrasound image of living tissue by the positioning and segmenting module, a process of manually judging each of the one or more images of the nodule area to be examined by the manual review module, and a result of the manual judgment output by the output module.

According to an embodiment of the 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.

According to an embodiment of the present application, the system for detecting a nodule of living tissue may further comprise an optimization module, and the method for detecting a nodule of living tissue may further comprise receiving a conclusive report of manual judgment from the output module with the optimization module, and continuously optimizing the localization segmentation module through machine in-depth learning and training.

According to embodiments of the application, the communicative connection of the localization segmentation module to the sonogram acquisition module may include one or both of a wired connection and a wireless connection.

According to the embodiment of the application, the positioning and dividing module, the manual auditing module and the output module can be integrated in one PC. Preferably, the positioning and dividing module, the manual review module, the output module, the display module and the optimization module can be integrated into one PC.

According to an embodiment of the present application, the preprocessing of the ultrasound image may include removing portions unrelated to the ultrasound image and adjusting image attributes such as image size and gray scale to 256 × 256.

According to an embodiment of the present application, the Unet + + network may be a coarsely partitioned network, computing the smallest circumscribed square of the nodule, and if the circumscribed square size is greater than 80, then scaling 20 pixels out, otherwise scaling 30 pixels out to scale the truncated nodule to 512 × 512 size.

According to the embodiment of the application, the Medical Transformer network adopts a convolutional neural network and a Transformer module which are combined, and can comprise a local training module, a global training module and an integration module.

According to the embodiment of the application, the local training module can use a convolution module to perform up-sampling after performing down-sampling twice on an input model so as to obtain the local features of the image; the global training module can use a Transformer module to increase the receptive field of the network so as to obtain the global characteristics of the image; the integration module can superpose the global features and the local features to obtain a final segmentation result.

According to an embodiment of the present application, the system for detecting a nodule in living tissue may further comprise using a Dice coefficient in combination with a binary cross-entropy function as a trained loss function for cases of class inhomogeneity in the segmentation.

According to embodiments of the application, the ultrasound image acquisition module may comprise a conventional B-ultrasound machine, and the biopsy nodule may comprise a thyroid nodule.

Compared with the prior art, the system and the method for detecting the nodule of the living tissue according to the embodiment of the application can realize at least the following beneficial effects:

because the ultrasonic image acquisition module is adopted in the system and the method for detecting the nodes of the living tissues, and the ultrasonic image of the living tissues is generated through the examination of the patient, the ultrasonic image acquisition module can be a conventional B-ultrasonic machine, so that no change is needed to be made on the conventional B-ultrasonic machine, and the modification cost of the existing equipment is saved.

The positioning segmentation module is in communication connection with the sonogram acquisition module, and the communication connection can be a wired connection or a wireless connection. Thus, the processing and analysis portion, represented by the positioning and segmentation module, may be located adjacent to the ultrasound machine, or may be remotely connected thereto. In case of remote connection, remote medical diagnosis is facilitated.

After the ultrasonic image is preprocessed, the locating and segmenting module locates and segments nodules on the ultrasonic image of the living tissue through the Unet + + network and the Medical transducer network loaded on the ultrasonic image so as to form one or more images of the nodule area to be checked. Each of the one or more images of the nodal region to be investigated is an image which is worth further analysis and therefore does not miss any possible lesion.

The manual auditing module carries out manual judgment on each of the one or more to-be-checked node area images. The one or more images of the nodule area to be examined generated by the positioning segmentation module have better identification for a manual examination module or an ultrasonic doctor, so that relatively accurate judgment can be given. This greatly improves detection efficiency and accuracy.

The output module outputs the result of the manual judgment formed by the manual auditing module, and also can output a conclusion report of the manual judgment. The results and reports can be printed into book-side reports and can be fed back to the optimization module, so that the software on the positioning and dividing module can be continuously trained, deeply learned and optimized.

The system and the method for detecting the in-vivo tissue nodule according to the embodiment of the application can provide a quick and accurate auxiliary diagnosis result, so that the system and the method have good medical value and social value.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments will be briefly introduced below, and it is apparent that the drawings in the following description only relate to some embodiments of the present application and are not limiting on the present application.

FIG. 1 is a block diagram of a system for detecting a nodule in living tissue in accordance with an embodiment of the present application.

FIG. 2 is a segmentation network flow diagram of a system for detecting a nodule in living tissue according to an embodiment of the present application.

Fig. 3 is a diagram of a net + + network model for a system for detecting in vivo tissue nodules according to an embodiment of the present application.

Fig. 4 is a diagram of a Medical transducer network model architecture for a system for detecting a nodule in living tissue according to an embodiment of the present application.

FIG. 5 is a flow chart of a method for detecting a nodule in living tissue according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings of the embodiments of the present application. It should be apparent that the described embodiments are only some of the embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the application without any inventive step, are within the scope of protection of the application.

Unless defined otherwise, 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 belongs. The use of "first," "second," and similar terms in the description and claims of this patent application do not denote any order, quantity, or importance, but rather the terms are used to distinguish one element from another. Also, the use of the terms "a" or "an" and the like do not denote a limitation of quantity, but rather denote the presence of at least one.

Embodiments of the present application are described below with reference to the drawings.

FIG. 1 is a block diagram of a system for detecting a nodule in living tissue in accordance with an embodiment of the present application.

Referring to FIG. 1, a system for detecting a nodule in living tissue is provided according to an embodiment of the present application, including an ultrasound image acquisition module, a localization segmentation module, a manual review module, and an output module, which are now described in detail below.

The ultrasound image acquisition module generates an ultrasound image of the living tissue through an examination of the patient. According to an embodiment of the application, the sonogram acquisition module may be, for example, a conventional B-ultrasound machine. However, the embodiments of the present application are not limited thereto, but the ultrasound image acquisition module may also be any other device that acquires a conventional two-dimensional ultrasound image of a predetermined position living tissue by conventional B-mode ultrasound, for example, a-mode ultrasound, M-mode ultrasound, D-mode ultrasound, color ultrasound, or the like.

Common examinations for the ultrasound acquisition module are thyroid, cervical lymph node, female breast, abdominal B-ultrasound, liver B-ultrasound, prostate B-ultrasound, etc. The following description will be given only by way of example of thyroid B-mode ultrasound.

The positioning segmentation module is in communication connection with the ultrasonic image acquisition module, receives the ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positions and segments nodules on the ultrasonic image of the living tissue through the Unet + + network and the Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed, so as to form one or more images of the nodule area to be examined.

FIG. 2 is a segmentation network flow diagram of a system for detecting a nodule in living tissue according to an embodiment of the present application. Fig. 3 is a diagram of a net + + network model for a system for detecting in vivo tissue nodules according to an embodiment of the present application. Fig. 4 is a diagram of a Medical transducer network model architecture for a system for detecting a nodule in living tissue according to an embodiment of the present application.

As shown in fig. 2 and 3, the uploaded thyroid B-mode ultrasound image is preprocessed to remove the portions irrelevant to the ultrasound image, the size of the image is adjusted to 256 × 256, the B-mode ultrasound image is segmented in the first stage by using a segmentation model, and the network in the first stage is a rough segmentation network, so as to obtain the position and approximate shape information of the thyroid nodule, so that the interesting field of view is reduced in the next fine segmentation, and a better segmentation effect is expected to be obtained.

After the coarse segmentation result is obtained, the minimum circumscribed square of the nodule is calculated, if the size of the circumscribed square is larger than 80, 20 pixels are externally expanded, otherwise 30 pixels are externally expanded. The nodules are truncated and scaled to 512 x 512 size as input to the two-stage segmentation network. The two-phase network is a finely divided network.

The roughly divided network uet + + network of the present application is described specifically as follows:

the network adopts a convolutional neural network Unet + + network, the model is an improved version of a segmented classical network Unet, and compared with the Unet, the network fills a vacant part in the middle of the Unet network with a plurality of small Unet structures, so that the learning capacity of the network can be improved. The image of the input model is subjected to four convolution and down-sampling processes to obtain different hierarchical features of the input image. And then, each level of network performs up-sampling for multiple times until the original size is restored. In the convolution nodes of each stage, the characteristics of the same stage are transmitted to a subsequent convolution module of the same stage through skip-connect operation, so that the characteristics of each stage are enriched, and the condition that the gradient disappears is avoided. After the network training is finished, a convolution module of 1X1 can be added after X12, X13, X14 and X15, and the output condition of each branch is monitored to obtain the output of each branch. According to the learning ability, pruning can be carried out to different degrees, the size of the final model is reduced under the condition that the network performance is not influenced too much, and the portability of the model is improved.

As shown in fig. 4, the subdivision network of the present application employs a Medical Transformer network, which is described in detail as follows:

the network adopts a convolution neural network and a Transformer module to be combined. The model adopts a local-global training strategy, and two parallel training networks simultaneously learn and extract picture characteristics. The local training module uses a convolution module to perform downsampling twice on the input model and then perform upsampling, so that the local features of the image can be obtained. The lower training model is a global training model, and a Transformer model is used for increasing the receptive field of the network. The method uses a convolution layer to extract features, then uses a transform module to integrate the features in the image and refine the features, a decoder part uses a cascaded upsampler and has four upsampling processes for decoding hidden features to output a final segmentation result, each layer in the decoder adopts skip-connect operation, the features of the transform are increased, so that the features of different layers can be better mixed together, and the accuracy of the model is improved. And finally, adding the results of the global branch and the local branch to obtain a final segmentation result.

A method of mixing the dice coefficient and the cross entropy is adopted in training, so that the target area can be concentrated, and the problem of unbalanced category (multiple backgrounds and small nodules) is solved.

In summary, the following steps:

according to an embodiment of the present application, the preprocessing of the ultrasound image may include removing portions unrelated to the ultrasound image and adjusting image attributes such as image size and gray scale to 256 × 256.

According to an embodiment of the present application, the Unet + + network may be a coarsely partitioned network, computing the smallest circumscribed square of the nodule, and if the circumscribed square size is greater than 80, then scaling 20 pixels out, otherwise scaling 30 pixels out to scale the truncated nodule to 512 × 512 size.

According to the embodiment of the application, the Medical Transformer network adopts a convolutional neural network and a Transformer module which are combined, and can comprise a local training module, a global training module and an integration module.

According to the embodiment of the application, the local training module can use a convolution module to perform up-sampling after performing down-sampling twice on an input model so as to obtain the local features of the image; the global training module can use a Transformer module to increase the receptive field of the network so as to obtain the global characteristics of the image; the integration module can superpose the global features and the local features to obtain a final segmentation result.

Referring back to fig. 1, the output module outputs the result of the manual judgment formed by the manual review module.

According to an embodiment of the application, the system for detecting a nodule of living tissue may further include a display module, and the display module may display a process of positioning and segmenting the nodule on the ultrasound image of the living tissue by the positioning and segmenting module, a process of manually judging each of one or more images of the nodule area to be checked by the manual auditing module, and a result of the manual judgment output by the output module.

According to an embodiment of the 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.

According to an embodiment of the present application, the system for detecting a nodule in living tissue may further comprise an optimization module that may receive a conclusive report of human judgment from the output module and continuously optimize the localization segmentation module through machine in-depth learning and training.

According to embodiments of the application, the communicative connection of the localization segmentation module to the sonogram acquisition module may include one or both of a wired connection and a wireless connection.

According to the embodiment of the application, the positioning and dividing module, the manual auditing module and the output module can be integrated in one PC. Preferably, the positioning and dividing module, the manual review module, the output module, the display module and the optimization module can be integrated into one PC.

According to an embodiment of the present application, the system for detecting a nodule in living tissue may further comprise using a Dice coefficient in combination with a binary cross-entropy function as a trained loss function for cases of class inhomogeneity in the segmentation.

According to embodiments of the application, the ultrasound image acquisition module may comprise a conventional B-ultrasound machine, and the biopsy nodule may comprise a thyroid nodule.

FIG. 5 is a flow chart of a method for detecting a nodule in living tissue according to an embodiment of the present application.

Referring to FIG. 3, a method for detecting a nodule of living tissue by a system for detecting a nodule of living tissue is provided according to an embodiment of the present application. In the following description, in order to avoid redundancy, a description overlapping with the system for detecting a nodule of living tissue according to an embodiment of the present application described with reference to fig. 1 to 2 will be omitted.

The system for detecting the in-vivo tissue nodule comprises an ultrasonic image acquisition module, a positioning segmentation module, a manual auditing module and an output module, and preferably also comprises a display module and an optimization module.

A method for detecting a nodule in living tissue comprising the steps of: generating an ultrasound image of the living tissue by examination of the patient using an ultrasound image acquisition module; adopting a positioning segmentation module which is in communication connection with the ultrasonic image acquisition module to receive an ultrasonic image of the living tissue generated by the ultrasonic image acquisition module, and positioning and segmenting nodules on the ultrasonic image of the living tissue through a Unet + + network and a Medical transducer network loaded on the ultrasonic image after the ultrasonic image is preprocessed to form one or more images of the nodule area to be examined; adopting a manual auditing module to carry out manual judgment on each of one or more to-be-checked node area images; and outputting the result of the manual judgment formed by the manual auditing module by adopting an output module.

According to an embodiment of the application, the method for detecting a nodule of living tissue may further include displaying, by using the display module, a process of positioning and segmenting the nodule on the ultrasound image of the living tissue by the positioning and segmenting module, a process of manually judging each of the one or more images of the nodule area to be checked by the manual review module, and a result of the manual judgment output by the output module.

According to an embodiment of the 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.

According to an embodiment of the present application, the method for detecting a nodule in living tissue may further comprise receiving a conclusive report of human judgment from the output module with the optimization module, and continuously optimizing the localization segmentation module through machine in-depth learning and training.

According to embodiments of the application, the communicative connection of the localization segmentation module to the sonogram acquisition module may include one or both of a wired connection and a wireless connection.

According to the embodiment of the application, the positioning and dividing module, the manual auditing module and the output module can be integrated in one PC. Preferably, the positioning and dividing module, the manual review module, the output module, the display module and the optimization module can be integrated into one PC.

According to an embodiment of the present application, the preprocessing of the ultrasound image may include removing portions unrelated to the ultrasound image and adjusting image attributes such as image size and gray scale to 256 × 256.

According to an embodiment of the present application, the Unet + + network may be a coarsely partitioned network, computing the smallest circumscribed square of the nodule, and if the circumscribed square size is greater than 80, then scaling 20 pixels out, otherwise scaling 30 pixels out to scale the truncated nodule to 512 × 512 size.

According to the embodiment of the application, the Medical Transformer network adopts a convolutional neural network and a Transformer module which are combined, and can comprise a local training module, a global training module and an integration module.

According to the embodiment of the application, the local training module can use a convolution module to perform up-sampling after performing down-sampling twice on an input model so as to obtain the local features of the image; the global training module can use a Transformer module to increase the receptive field of the network so as to obtain the global characteristics of the image; the integration module can superpose the global features and the local features to obtain a final segmentation result.

According to an embodiment of the present application, the system for detecting a nodule in living tissue may further comprise using a Dice coefficient in combination with a binary cross-entropy function as a trained loss function for cases of class inhomogeneity in the segmentation.

According to embodiments of the application, the ultrasound image acquisition module may comprise a conventional B-ultrasound machine, and the biopsy nodule may comprise a thyroid nodule.

Compared with the prior art, the system and the method for detecting the nodule of the living tissue according to the embodiment of the application can realize at least the following beneficial effects:

because the ultrasonic image acquisition module is adopted in the system and the method for detecting the nodes of the living tissues, and the ultrasonic image of the living tissues is generated through the examination of the patient, the ultrasonic image acquisition module can be a conventional B-ultrasonic machine, so that no change is needed to be made on the conventional B-ultrasonic machine, and the modification cost of the existing equipment is saved.

The positioning segmentation module is in communication connection with the sonogram acquisition module, and the communication connection can be a wired connection or a wireless connection. Thus, the processing and analysis portion, represented by the positioning and segmentation module, may be located adjacent to the ultrasound machine, or may be remotely connected thereto. In case of remote connection, remote medical diagnosis is facilitated.

After the ultrasonic image is preprocessed, the locating and segmenting module locates and segments nodules on the ultrasonic image of the living tissue through the Unet + + network and the Medical transducer network loaded on the ultrasonic image so as to form one or more images of the nodule area to be checked. Each of the one or more images of the nodal region to be investigated is an image which is worth further analysis and therefore does not miss any possible lesion.

The manual auditing module carries out manual judgment on each of the one or more to-be-checked node area images. The one or more images of the nodule area to be examined generated by the positioning segmentation module have better identification for a manual examination module or an ultrasonic doctor, so that relatively accurate judgment can be given. This greatly improves detection efficiency and accuracy.

The output module outputs the result of the manual judgment formed by the manual auditing module, and also can output a conclusion report of the manual judgment. The results and reports can be printed into book-side reports and can be fed back to the optimization module, so that the software on the positioning and dividing module can be continuously trained, deeply learned and optimized.

The system and the method for detecting the in-vivo tissue nodule according to the embodiment of the application can provide a quick and accurate auxiliary diagnosis result, so that the system and the method have good medical value and social value.

The above description is only exemplary of the present application and is not intended to limit the scope of the present application, which is defined by the appended claims.