Thyroid diffuse lesion and nodule auxiliary diagnosis system

By combining edge computing and cloud computing in the thyroid ultrasound diagnostic system, using lightweight deep learning models to diagnose diffuse thyroid lesions and nodules, the problem that traditional diagnosis relies on doctors' experience and deep learning models is difficult to run in real time, and improved the problem of the traditional diagnosis being difficult to operate in real time. Diagnostic accuracy and efficiency are suitable for mobile devices and primary care.

CN119049666APending Publication Date: 2024-11-29SHANGHAI SOUNDWISE TECHNOLOGY CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411130633.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Traditional thyroid ultrasound diagnosis relies on doctors’ experience, has strong subjectivity, inconsistent diagnosis results and low efficiency, and the existing deep learning models do not assist in diagnosis of diffuse thyroid lesions and nodules, and are difficult to run in real time on mobile devices.

Method used

Design a diffuse thyroid lesions and nodules assisted diagnosis system, combine edge computing and cloud computing, and use lightweight deep learning models to conduct preliminary diagnosis locally. When conditions are insufficient, feature data will be transmitted to the cloud for comprehensive diagnosis, making full use of the cloud Computing resources.

Benefits of technology

It improves the diagnostic accuracy and consistency of diffuse thyroid lesions and nodules, reduces the work burden of doctors, reduces the dependence on high-experienced doctors, improves the efficiency of medical resource utilization, and is suitable for primary medical care and remote areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119049666A_ABST
    Figure CN119049666A_ABST
Patent Text Reader

Abstract

The invention provides a thyroid diffuse lesion and nodule auxiliary diagnosis system. A local calculation subsystem collects thyroid ultrasonic data; the preliminary diagnosis module is used for extracting first feature data in the ultrasonic data based on a lightweight preliminary recognition diagnosis model and forming a preliminary diagnosis result; the condition evaluation module evaluates a local computing resource condition to obtain a condition evaluation result; when the resource conditions are sufficient, the comprehensive diagnosis module processes the first feature data based on a pre-trained first comprehensive model to form a first comprehensive diagnosis result for thyroid diffuse lesions and nodules and outputs the first comprehensive diagnosis result; when the resource conditions are not sufficient, the data transmission module transmits the data to the cloud; and the cloud end processes the first feature data based on a second comprehensive model to form a second comprehensive diagnosis result about the thyroid diffuse lesion and the nodule and feeds back the second comprehensive diagnosis result. The timeliness of diagnosis is improved, and clinical doctors can conveniently make examination, treatment and follow-up visit suggestions in time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of medical ultrasonic image processing, and in particular to an auxiliary diagnosis system for diffuse thyroid lesions and nodules. Background Art

[0002] Traditional thyroid ultrasound diagnosis mainly relies on the doctor's experience and manual operation. The doctor obtains thyroid images through ultrasound equipment and identifies and diagnoses thyroid lesions and nodules based on the image characteristics. This method has the following shortcomings: it relies on the doctor's experience. The diagnosis results are highly dependent on the doctor's experience and expertise, and the diagnosis results between different doctors may be different; the workload is large: manual analysis of images is time-consuming and labor-intensive, especially when dealing with a large number of cases, which can easily lead to fatigue and misdiagnosis; it is highly subjective: the doctor's subjective judgment may lead to inconsistency and accuracy of the diagnosis results. Because there is sometimes confusion between diffuse thyroid lesions and nodules, different doctors may have different opinions. For example, some doctors believe that it is a diffuse thyroid lesion, while other doctors believe that it is a nodule.

[0003] Computer-aided diagnosis (CAD) technology uses image processing and pattern recognition technology to assist doctors in identifying and diagnosing thyroid lesions and nodules. Although CAD technology has improved the accuracy and efficiency of diagnosis to a certain extent, there are still some limitations: Feature extraction relies on manual design: The feature extraction process of traditional CAD systems relies on features designed by experts and is difficult to adapt to complex and changeable lesion morphology; Classifier performance is limited: The classification performance of traditional machine learning algorithms is limited and it is difficult to handle high-dimensional and nonlinear features. This also limits the distinction between complex diffuse thyroid lesions and thyroid nodules.

[0004] In recent years, deep learning technology has made significant progress in the field of medical image analysis, especially in ultrasound image analysis. Deep learning models can automatically learn features from data, significantly improving the accuracy and robustness of lesion identification. However, there is currently no deep learning model for auxiliary diagnosis of diffuse thyroid lesions and nodules. In addition, the training and reasoning of deep learning models require high-performance computing resources and are difficult to run in real time on mobile devices. With the development of mobile computing and sensing technology, handheld ultrasound devices have gradually become popular. These devices are small in size and highly portable, and can perform rapid ultrasound examinations in environments such as bedside, home or field. However, due to the limitations of computing resources and display screens, handheld ultrasound devices have certain limitations in image processing and result presentation. Summary of the invention

[0005] Based on the above content, the present invention provides a system for auxiliary diagnosis of diffuse thyroid lesions and nodules, aiming to solve technical problems in the prior art such as strong subjectivity in diagnosis of diffuse thyroid lesions and nodules.

[0006] A thyroid diffuse lesion and nodule auxiliary diagnosis system, comprising:

[0007] The local computing subsystem is used for edge computing, including:

[0008] A data acquisition module, used for acquiring first patient data which is thyroid ultrasound data;

[0009] A preprocessing module, connected to the data acquisition module, for preprocessing ultrasound data;

[0010] A preliminary diagnosis module, connected to the preprocessing module, is used to extract the first feature data in the ultrasound data based on a lightweight preliminary recognition diagnosis model and form a preliminary diagnosis result about diffuse thyroid lesions and nodules;

[0011] A condition evaluation module is used to evaluate the local computing resource conditions and obtain condition evaluation results;

[0012] A comprehensive diagnosis module, connected to the condition evaluation module and the preliminary diagnosis module respectively, for executing a comprehensive diagnosis task when the condition evaluation result indicates that the local computing resources are sufficient, and processing the first feature data based on a pre-trained first comprehensive model to form a first comprehensive diagnosis result output on diffuse thyroid lesions and nodules;

[0013] The data transmission module is connected to the condition evaluation module and the preliminary diagnosis module respectively, and is used to transmit the first characteristic data to the cloud when the condition evaluation result is that the local computing resource condition is insufficient;

[0014] The cloud is used to perform comprehensive diagnostic tasks, processing the first feature data based on a pre-trained second comprehensive model to form a second comprehensive diagnostic result about diffuse thyroid lesions and nodules, and feeding it back to the local computing subsystem.

[0015] Further, the preliminary identification and diagnosis model includes a lightweight first feature identification model and a lightweight preliminary diagnosis model;

[0016] The preliminary diagnosis module extracts first feature data from the preprocessed ultrasound data based on the trained first feature recognition model, and processes the first feature data based on the trained preliminary diagnosis model to form a preliminary diagnosis result about diffuse thyroid lesions and nodules.

[0017] Further, the first comprehensive model includes a second feature recognition model and a first comprehensive diagnosis model;

[0018] When the condition evaluation result indicates that the local computing resources are sufficient, the comprehensive diagnosis module is used to:

[0019] Processing the first feature data based on the trained second feature recognition model to extract second feature data of the thyroid gland;

[0020] The second feature data is processed based on a pre-trained first comprehensive diagnosis model to form a first comprehensive diagnosis result output.

[0021] Further, the second comprehensive model includes a third feature recognition model and a second comprehensive diagnosis model;

[0022] When the cloud receives the first characteristic data transmitted by the data transmission module, it is used to:

[0023] Processing the first feature data based on the trained third feature recognition model to extract third feature data of the thyroid gland;

[0024] The third feature data is processed based on a pre-trained second comprehensive diagnosis model to form a second comprehensive diagnosis result, and the second comprehensive diagnosis result is fed back to the local computing subsystem.

[0025] Furthermore, the data collection module is also used to collect existing second patient data related to the patient's thyroid gland, and the types of the second patient data include medical history data and test and examination data;

[0026] The preprocessing module is also used to preprocess the collected second patient data;

[0027] The comprehensive diagnosis module is also connected to the preprocessing module, and when the condition evaluation result indicates that the local computing resource condition is sufficient, the first feature data and the preprocessed second patient data are processed based on the first comprehensive diagnosis model to form a first comprehensive diagnosis result;

[0028] The data transmission module is also connected to the preprocessing module, and when the condition evaluation result is that the local computing resource condition is insufficient, the first feature data and the preprocessed second patient data are transmitted to the cloud;

[0029] The cloud processes the first feature data and the second patient data based on the second comprehensive diagnosis model to form a second comprehensive diagnosis result, and feeds the second comprehensive diagnosis result back to the local computing subsystem.

[0030] Further, the condition evaluation module is used to simultaneously evaluate the local computing resource conditions and the network conditions to obtain the condition evaluation results;

[0031] When the condition evaluation result is that the local computing resource condition is insufficient but the network condition meets the predetermined standard, the data transmission module transmits the first feature data and the pre-processed second patient data to the cloud;

[0032] When the condition evaluation result is that the local computing resource condition is insufficient and the network condition does not meet the predetermined standard, the data transmission module locally caches the first feature data and the pre-processed second patient data, and when the network condition changes to meet the predetermined standard, continues to transmit the unfinished data in a breakpoint resume mode;

[0033] When the network conditions in the cloud do not meet the predetermined standards, the cloud will delay and wait for the network conditions to meet the predetermined standards and then complete the data reception before executing the comprehensive diagnosis task.

[0034] Furthermore, both the first comprehensive diagnostic result and the second comprehensive diagnostic result include auxiliary diagnostic conclusions and auxiliary diagnostic suggestions regarding diffuse thyroid lesions and nodules, and the auxiliary diagnostic suggestions include examination suggestions, treatment suggestions, and follow-up suggestions.

[0035] Furthermore, the comprehensive diagnosis module is further used to perform thyroid cancer risk assessment on the first feature data and the second patient data based on the first comprehensive diagnosis model, and output the assessed thyroid cancer risk probability attached to the first comprehensive diagnosis result;

[0036] The cloud is also used to perform thyroid cancer risk assessment for the first feature data and the second patient data based on the second comprehensive diagnostic model, and feed back the assessed thyroid cancer risk probability in the second comprehensive diagnostic result to the local computing subsystem.

[0037] Furthermore, the local computing subsystem is an edge computing device or is composed of multiple edge computing devices;

[0038] When the local computing subsystem includes multiple edge computing devices, each edge computing device includes a data acquisition module, a preprocessing module, a comprehensive diagnosis module, a preliminary diagnosis module and a data transmission module;

[0039] The primary edge computing device initiates the diagnostic task, and the rest are auxiliary edge computing devices.

[0040] The local computing subsystem allocates diagnostic tasks based on the computing resource conditions of the primary edge computing device and the auxiliary edge computing device.

[0041] Furthermore, the primary edge computing device also includes a data receiving module and a display module;

[0042] The data receiving module is used to receive the second comprehensive diagnosis result fed back by the cloud;

[0043] The display module is connected to the data receiving module, the preliminary diagnosis module and the comprehensive diagnosis module respectively, and is used to display the output preliminary diagnosis result, the first comprehensive diagnosis result and the second comprehensive diagnosis result.

[0044] The beneficial technical effects of the present invention are:

[0045] The present invention trains a deep learning model for the identification and diagnosis of diffuse thyroid lesions and nodules, effectively distinguishes diffuse thyroid lesions and nodules, and improves the accuracy, consistency and reliability of diagnosis. The present invention distributes tasks between edge computing and cloud computing according to the local resource computing conditions. When local resources are sufficient for computing, local resources are used to calculate comprehensive diagnostic tasks to obtain detailed diagnostic results, reducing data transmission resource consumption and time consumption. When local resources are insufficient, the feature data is sent to the cloud for detailed comprehensive diagnosis after lightweight model processing, and cloud computing resources are fully utilized to obtain diagnostic results, overcoming the limitation of insufficient local resources. First, lightweight model diagnosis is performed, and then comprehensive diagnosis is further performed on the basis of preliminary diagnostic feature extraction, reducing the computing resources for comprehensive diagnosis, reducing the amount of data transmission during cloud comprehensive diagnosis, and reducing the workload of doctors through automated diagnosis, so that doctors can spend more time on the treatment of complex cases and patient management, thereby improving the utilization efficiency of medical resources. Through automated diagnosis, diagnostic efficiency is improved, and waiting time and medical expenses of patients are reduced. Especially in primary medical institutions and remote areas, the coverage and quality of medical services can be significantly improved. Reliance on highly experienced doctors is reduced, and the labor cost of medical institutions is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The module schematic diagram of a system for auxiliary diagnosis of diffuse thyroid lesions and nodules of the present invention is shown. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0049] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0050] See also Figure 1 The present invention provides a thyroid diffuse lesions and nodules auxiliary diagnosis system, comprising:

[0051] The local computing subsystem (1) is used for edge computing and includes:

[0052] A data acquisition module (101), used for acquiring first patient data which is thyroid ultrasound data;

[0053] A preprocessing module (102), connected to the data acquisition module (101), for preprocessing ultrasonic data;

[0054] A preliminary diagnosis module (103), connected to the preprocessing module (102), is used to extract first feature data from the ultrasound data based on a lightweight preliminary recognition diagnosis model and form a preliminary diagnosis result on diffuse thyroid lesions and nodules;

[0055] A condition evaluation module (104), used to evaluate the local computing resource conditions and obtain condition evaluation results;

[0056] A comprehensive diagnosis module (105), connected to the condition evaluation module (104) and the preliminary diagnosis module (103), respectively, for executing a comprehensive diagnosis task when the condition evaluation result indicates that the local computing resources are sufficient, and processing the first feature data based on a pre-trained first comprehensive model to form a first comprehensive diagnosis result output on diffuse thyroid lesions and nodules;

[0057] A data transmission module (106), connected to the condition evaluation module (104) and the preliminary diagnosis module (103), respectively, for transmitting the first characteristic data to the cloud (2) when the condition evaluation result indicates that the local computing resource condition is insufficient;

[0058] The cloud (2) is used to perform comprehensive diagnosis tasks, process the first feature data based on a pre-trained second comprehensive model to form a second comprehensive diagnosis result on diffuse thyroid lesions and nodules, and feed it back to the local computing subsystem (1).

[0059] The local computing subsystem is, for example, one or more tablet computers or mobile phones, or directly a handheld ultrasound device. The computing resources GPU and memory are limited, which are insufficient to support the operation of more complex deep learning models (such as CNN, RNN, etc.), but lightweight deep learning models can be deployed. The edge computing device can pre-process ultrasound data, etc., and use lightweight models for preliminary feature extraction and simple preliminary diagnosis.

[0060] By introducing deep learning technology, diffuse thyroid lesions and nodules can be automatically identified and distinguished, reducing the reliance on doctors' experience and subjective judgment, and improving the consistency and accuracy of diagnostic results. The deep learning model is used to automatically extract features from ultrasound images, adapt to complex and changeable lesion morphology, avoid the limitations of manually designed features, and improve the robustness of feature extraction and recognition. Through model optimization and lightweight processing, the computational complexity of the deep learning model is reduced, so that it can run efficiently on mobile devices to meet the needs of real-time diagnosis. Combining the advantages of edge computing and cloud computing, the use of computing resources is optimized to meet the needs of real-time diagnosis. The present invention also distributes tasks between edge computing and cloud computing according to the local resource computing conditions. When local resources are sufficient for calculation, the local resources are used to calculate the comprehensive diagnostic task to obtain detailed diagnostic results, reducing the consumption of data transmission resources and time consumption. When local resources are insufficient, the feature data is sent to the cloud for detailed comprehensive diagnosis after lightweight model processing, making full use of cloud computing resources to obtain diagnostic results, and overcoming the limitations of insufficient local resources.

[0061] Furthermore, lightweight deep learning models such as MobileNet, EfficientNet, etc. Of course, other lightweight model architectures such as SqueezeNet, ShuffleNet, etc. can also be selected and optimized according to specific application scenarios. Lightweight deep learning models can automatically identify diffuse thyroid lesions and nodules while ensuring diagnostic accuracy, reduce the computational complexity and resource requirements of the model, and enable the model to run efficiently on mobile devices. For the training of lightweight models, the cloud uses large-scale annotated data sets for model training, and uses data enhancement, transfer learning and other technologies to improve the generalization ability of the model. Then, the model size is further compressed through quantization, pruning and other technologies to reduce the demand for computing and storage resources.

[0062] MobileNet: A lightweight deep learning model architecture for mobile devices.

[0063] EfficientNet: An efficient deep learning model architecture that achieves high accuracy and low computational complexity through model scaling techniques.

[0064] Edge computing can perform calculations on devices close to the data source, reducing data transmission delays. Cloud computing can use remote servers for large-scale computing and storage, providing powerful computing resources.

[0065] The preprocessing module (102) preprocesses the ultrasound data including image standardization, denoising and image enhancement, thereby improving the quality and consistency of the ultrasound image and providing a good basis for subsequent feature extraction and recognition diagnosis.

[0066] Image normalization: Normalize the pixel values ​​of ultrasound images to a range of 0-1 to eliminate differences caused by different equipment and imaging conditions.

[0067] Denoising: Use Gaussian filtering, mean filtering and other methods to remove noise from the image and improve image clarity.

[0068] Image enhancement: Use contrast enhancement, edge enhancement and other technologies to enhance the characteristics of lesions in the image to facilitate model recognition.

[0069] Further, the preliminary identification and diagnosis model includes a lightweight first feature identification model and a lightweight preliminary diagnosis model;

[0070] The preliminary diagnosis module (103) extracts first feature data from the preprocessed ultrasound data based on the trained first feature recognition model, and processes the first feature data based on the trained preliminary diagnosis model to form a preliminary diagnosis result on diffuse thyroid lesions and nodules.

[0071] Due to the limitation of computing resources on mobile devices, a lightweight model is first used to obtain preliminary diagnostic results and display them, so that doctors can get a preliminary understanding of the diagnostic results in a timely manner. In addition, the first feature data extracted will be displayed after being annotated on the ultrasound data, and doctors can also check the annotated data in a timely manner to understand the patient's condition and whether the patient has thyroid disease. When the second comprehensive diagnostic result calculated on the cloud is fed back, the second comprehensive diagnostic result will be displayed, and doctors can get detailed diagnostic results as a diagnostic aid. Preliminary feature extraction and simple diagnosis are performed using handheld ultrasound equipment or other mobile devices, and complex computing tasks are transmitted to cloud servers through the network for processing, making full use of cloud computing resources and improving diagnostic efficiency.

[0072] The preliminary identification and diagnosis model mainly performs preliminary feature extraction, such as texture features, calculating the grayscale co-occurrence matrix of the image, extracting texture features such as contrast, correlation, uniformity, etc. Another example is edge features. The edge features can extract edge information in the image through the Canny edge detection algorithm to help identify the shape characteristics of lesions and nodules.

[0073] The preliminary identification and diagnosis model uses the MobileNet or SqueezeNet model to perform preliminary classification of images, determine whether there are diffuse thyroid lesions or nodules, locate and mark potential lesion areas, and provide preliminary diagnostic information.

[0074] Specifically, the first feature data includes marking the region of interest, such as the area where the thyroid gland exists, and the region of interest is given to the comprehensive diagnosis module or the cloud. Since there are multiple frames of ultrasound images in the ultrasound video data collected by the handheld ultrasound device, only the regions of interest of these ultrasound images are sent to the comprehensive diagnosis module and the cloud. The comprehensive diagnosis module and the cloud only perform feature extraction and disease diagnosis on the region of interest, which greatly reduces the amount of calculation. In addition, when the data is sent to the cloud, the region of interest data of each frame of the ultrasound video is directly sent to the cloud instead of the entire ultrasound video data, which greatly reduces the amount of data transmission, shortens the data transmission time, and further ensures the timeliness of the diagnosis.

[0075] The preliminary diagnosis results are mainly about the characteristics of the thyroid gland, including diffuse thyroid lesions and nodule characteristics. For example, a preliminary diagnosis result obtained by processing ultrasound data is as follows:

[0076] Characteristics of diffuse thyroid disease: enlarged thyroid gland and uneven echogenicity;

[0077] The characteristics of the nodule are: a 2.0 cm low-echo nodule with unclear boundaries, microcalcification points inside, and rich blood flow signals in the right lobe.

[0078] Based on the preliminary diagnosis results, doctors can see whether the patient has diffuse lesions, whether there are nodules, and the characteristics of the nodules. Based on these characteristics, doctors can preliminarily understand whether the patient has thyroid disease and whether the nodules are benign or malignant.

[0079] Further, the first comprehensive model includes a second feature recognition model and a first comprehensive diagnosis model;

[0080] When the condition evaluation result indicates that the local computing resources are sufficient, the comprehensive diagnosis module (105) is used to:

[0081] Processing the first feature data based on the trained second feature recognition model to extract second feature data of the thyroid gland;

[0082] The second feature data is processed based on a pre-trained first comprehensive diagnosis model to form a first comprehensive diagnosis result output.

[0083] Further, the second comprehensive model includes a third feature recognition model and a second comprehensive diagnosis model;

[0084] When the cloud (2) receives the first characteristic data transmitted by the data transmission module (106), it is used to:

[0085] Processing the first feature data based on the trained third feature recognition model to extract third feature data of the thyroid gland;

[0086] The third feature data is processed based on a pre-trained second comprehensive diagnosis model to form a second comprehensive diagnosis result, and the second comprehensive diagnosis result is fed back to the local computing subsystem (1).

[0087] Specifically, the first comprehensive model and the second comprehensive model use a complex reasoning model, which can be the same trained deep learning model architecture. The first comprehensive model or the second comprehensive model uses, for example, a U-Net model, a ResNet model, or a DenseNet model to perform ultrasound image segmentation, accurately segment the lesion area and nodules, provide detailed segmentation results, and give a comprehensive diagnosis result.

[0088] The cloud (2) can integrate data from multiple local computing subsystems and use the powerful storage and computing capabilities of cloud computing to perform unified data management and batch processing. The cloud can also be used to train various models and perform incremental learning of models in the cloud, using new data to continuously update model parameters, improve the generalization ability of the model, and deploy the optimized model on the edge computing device according to actual conditions.

[0089] In addition, the data transmission module (106) also uploads the preliminary diagnosis result to the cloud (2), and the cloud combines the first feature data and the preliminary diagnosis result to form a second comprehensive diagnosis result based on the second comprehensive diagnosis model. The edge computing device and the cloud collaboratively share the computing task. The edge computing device performs preliminary screening and feature extraction, and the cloud performs in-depth analysis and model reasoning. The calculation is performed based on the data processed by the preliminary diagnosis model, which can reduce computing resources and obtain more accurate diagnosis results.

[0090] Furthermore, the data collection module (101) is also used to collect existing second patient data related to the patient's thyroid gland, and the types of the second patient data include medical history data and test and examination data;

[0091] The preprocessing module (102) is also used to preprocess the collected second patient data;

[0092] The comprehensive diagnosis module (105) is also connected to the preprocessing module (102), and when the condition evaluation result indicates that the local computing resource condition is sufficient, the first feature data and the preprocessed second patient data are processed based on the first comprehensive diagnosis model to form a first comprehensive diagnosis result;

[0093] The data transmission module (106) is also connected to the preprocessing module (102), and when the condition evaluation result is that the local computing resource condition is insufficient, the first feature data and the preprocessed second patient data are transmitted to the cloud (2);

[0094] The cloud (2) processes the first feature data and the second patient data based on the second comprehensive diagnosis model to form a second comprehensive diagnosis result, and feeds the second comprehensive diagnosis result back to the local computing subsystem.

[0095] In general, when patients feel symptoms of thyroid disease, they usually undergo ultrasound scans to obtain ultrasound data. Then, based on the diagnosis results given by this system, if there is a problem, further tests may be performed to further determine the condition, and medical history data will be generated. The medical history data and examination data are not obtained at the same time. Some examination data may come out in subsequent examinations, and some examination data may be obtained during subsequent follow-up visits. Therefore, the system can perform multimodal data fusion based on the existing medical history data and test data combined with the initial scan ultrasound data, pre-process the data from different sources, perform feature fusion, and then conduct joint analysis to provide a more comprehensive diagnostic result. In other words, for the data at different stages of the same patient, a comprehensive diagnostic result for that stage can be given, rather than waiting for all the test results to come out before giving a comprehensive diagnostic result.

[0096] The processing of the second patient data by the preprocessing module (102) includes data cleaning, normalization and format conversion. In addition, the data preprocessing module also includes feature fusion processing of the second patient data that is not ultrasound data, such as constructing a multimodal feature vector to form preprocessed second patient data, and the comprehensive diagnosis module and the cloud further fuse the ultrasound data and the second patient data to form a first comprehensive diagnosis result and a second comprehensive diagnosis result based on the multimodal data, thereby improving the accuracy of the diagnosis.

[0097] Specifically, the medical history data in the second patient data includes:

[0098] Family history: Ask the patient if he or she has a family history of thyroid disease, thyroid cancer, or other endocrine diseases.

[0099] Past medical history: Find out whether the patient has a history of thyroiditis, hyperthyroidism or hypothyroidism.

[0100] Symptoms: Ask the patient in detail about their symptoms, such as fatigue, weight changes, palpitations, neck lumps, and difficulty swallowing.

[0101] Specifically, the test data in the second patient data includes laboratory test data, other imaging test data other than ultrasound examination, fine needle aspiration biopsy (FNA),

[0102] Laboratory test data include:

[0103] Thyroid function tests: including TSH (thyroid stimulating hormone), FT3 (free triiodothyronine), and FT4 (free thyroxine).

[0104] Thyroid autoantibodies: such as TPOAb (thyroid peroxidase antibodies) and TgAb (thyroglobulin antibodies).

[0105] Thyroglobulin (Tg): May be helpful in monitoring thyroid disease activity in some cases.

[0106] Other imaging examination data:

[0107] Radionuclide scanning: to assess thyroid function, especially the functionality of nodules (hot or cold nodules);

[0108] CT / MRI: Consider when invasive thyroid disease is suspected or involvement of surrounding structures is needed.

[0109] Fine needle aspiration biopsy (FNA) is performed on suspicious nodules to clarify the pathological nature, especially for nodules with malignant characteristics (such as microcalcifications, unclear boundaries, and nodules with rich blood flow signals).

[0110] The second patient data that has already been examined for these patients is fused to the ultrasound data for a comprehensive analysis of thyroid disease.

[0111] Furthermore, the condition evaluation module (104) is used to simultaneously evaluate the local computing resource conditions and the network conditions to obtain a condition evaluation result;

[0112] When the condition evaluation result is that the local computing resource condition is insufficient but the network condition meets the predetermined standard, the data transmission module (106) transmits the first characteristic data and the pre-processed second patient data to the cloud (2);

[0113] When the condition evaluation result is that the local computing resource condition is insufficient and the network condition does not meet the predetermined standard, the data transmission module (106) locally caches the first characteristic data and the pre-processed second patient data, and when the network condition changes to meet the predetermined standard, continues to transmit the unfinished data in a breakpoint resume mode;

[0114] When the network condition does not meet the predetermined standard, the cloud (2) delays and waits for the network condition to meet the predetermined standard and then completes data reception before executing the comprehensive diagnosis task.

[0115] The task allocation ratio of edge computing and cloud computing can be adjusted according to network conditions and computing resource allocation, thereby improving resource utilization.

[0116] The data transmission module (106) transmits data according to the set data priority, and regards data that plays an important role in the diagnosis result as key data, such as ultrasound data, and other data as non-key data. The key data is transmitted first, and the non-key data is transmitted later, so as to reduce the amount of data transmitted. When the network conditions in the cloud do not meet the predetermined standards, the cloud will delay the calculation, and calculate after the key data is transmitted to obtain a second comprehensive diagnosis result and feed it back to the local computing system. After the non-key data is also transmitted, the key data and the non-key data are combined to calculate and update the second comprehensive diagnosis result and feed it back to the local computing system, so that the user can know the diagnosis result in time.

[0117] Key data include, for example: the first characteristic data and preliminary diagnosis conclusion obtained by the preliminary diagnosis module, thyroid function tests (TSH, FT3, FT4), family history, past medical history, etc.

[0118] Non-critical data include: other imaging examinations, thyroglobulin (Tg), fine needle aspiration biopsy (FNA), etc.

[0119] Specifically, the non-critical data in the second patient data also includes lifestyle, eating habits, other diseases and medication status.

[0120] If the network conditions are good, the cloud (2) receives the data initially processed by the edge computing device and immediately performs detailed analysis. After completing the detailed analysis, the results are returned to the edge device in real time to ensure the timeliness and accuracy of the diagnosis results.

[0121] For critical data, the data transmission module (106) performs lossless compression transmission to ensure data quality. For non-critical data, the data transmission module performs lossy compression transmission to reduce the data volume to save bandwidth.

[0122] The data transmission module (106) uses an efficient data transmission protocol (such as HTTP / 2, WebSocket) to achieve real-time data synchronization between the edge device and the cloud.

[0123] Furthermore, both the first comprehensive diagnostic result and the second comprehensive diagnostic result include auxiliary diagnostic conclusions and auxiliary diagnostic suggestions regarding diffuse thyroid lesions and nodules, and the auxiliary diagnostic suggestions include examination suggestions, treatment suggestions, and follow-up suggestions.

[0124] The comprehensive diagnosis module and the cloud provide auxiliary diagnosis suggestions based on machine learning models and medical knowledge base, and provide feedback to clinicians for their reference and decision-making.

[0125] The system of the present invention can perform a comprehensive diagnosis at any time based on the patient's existing data about the thyroid gland to obtain diagnostic conclusions and suggestions, thereby improving the timeliness of the diagnosis and facilitating clinicians to make timely inspection, treatment and follow-up suggestions. For example, at the beginning, the patient undergoes a thyroid ultrasound examination, which is most of the time fused with the patient's medical history data. The system obtains the patient's possible malignant nodules based on the thyroid ultrasound data. In order to further determine, the system recommends that the patient undergo thyroid function tests, thyroid fine needle aspiration biopsy, etc. The clinician believes that these tests are indeed necessary for further confirmation based on the diagnostic suggestions in the comprehensive diagnostic results, and then provides the patient with a list of inspection items. The clinician can use the system at any time to fuse the previous ultrasound data (and medical history data) based on the existing inspection results to obtain a new comprehensive diagnostic result. The comprehensive diagnostic result has an updated auxiliary diagnostic suggestion. The clinician further updates the inspection plan, treatment plan and follow-up plan based on the new comprehensive diagnostic result and the auxiliary diagnostic suggestion.

[0126] Treatment recommendations include:

[0127] Drug treatment recommendations: Choose appropriate drug treatment based on the status of thyroid function, such as antithyroid drugs for hyperthyroidism and thyroid hormone replacement therapy for hypothyroidism.

[0128] Surgical treatment recommendations: Consider surgical resection for nodules with malignant features or in cases of suspected thyroid cancer.

[0129] Radioactive iodine therapy is recommended: For patients with hyperthyroidism or certain thyroid cancers, consider radioactive iodine therapy.

[0130] Follow-up recommendations include:

[0131] Arrange regular follow-up visits to monitor thyroid function and nodule changes based on the patient's specific situation;

[0132] Imaging review: Perform ultrasound examinations regularly to monitor changes in diffuse thyroid lesions and nodules.

[0133] Laboratory review: Regularly review thyroid function and autoantibody levels to evaluate treatment efficacy and disease progression.

[0134] Furthermore, the comprehensive diagnosis module (105) is also used to perform thyroid cancer risk assessment on the first feature data and the second patient data based on the first comprehensive diagnosis model, and output the assessed thyroid cancer risk probability attached to the first comprehensive diagnosis result;

[0135] The cloud (2) is also used to perform thyroid cancer risk assessment on the first feature data and the second patient data based on the second comprehensive diagnostic model, and feed back the assessed thyroid cancer risk probability in the second comprehensive diagnostic result to the local computing subsystem.

[0136] This AI system can perform thyroid cancer risk assessment based on multiple parameters (such as nodule characteristics, laboratory results, etc.), further helping patients pay attention to and solve problems.

[0137] Through detailed medical history collection, laboratory tests, imaging tests, and pathological examinations, combined with comprehensive analysis by this AI system, patients with diffuse thyroid lesions and nodules can be more accurately evaluated and managed. The application of this AI system in data integration, pattern recognition, and risk assessment can provide clinicians with valuable diagnostic and treatment recommendations, ultimately helping patients achieve better treatment outcomes and prognosis.

[0138] Furthermore, the local computing subsystem (1) is an edge computing device or is composed of multiple edge computing devices;

[0139] When the local computing subsystem (1) includes a plurality of edge computing devices, each edge computing device includes a data acquisition module (101), a preprocessing module (102), a comprehensive diagnosis module (105), a preliminary diagnosis module (103) and a data transmission module (106);

[0140] The primary edge computing device initiates the diagnostic task, and the rest are auxiliary edge computing devices.

[0141] The local computing subsystem allocates diagnostic tasks based on the computing resource conditions of the primary edge computing device and the auxiliary edge computing device.

[0142] The local computing subsystem detects the resource status of each edge computing device to dynamically allocate and adjust tasks. You can use tools such as Prometheus and Grafana to monitor the resource usage of edge devices. Use tools such as Ping and Traceroute to monitor network conditions.

[0143] Other auxiliary edge computing devices besides the primary edge computing device can be:

[0144] Neighboring edge servers: Additional edge servers located in the same network or physical location as the primary edge computing device that can be used to share computing tasks.

[0145] Edge Gateway: A gateway device that connects multiple edge computing devices and usually has high computing and storage capabilities.

[0146] Edge nodes: small computing devices distributed in different locations, such as smart routers, industrial controllers, etc.

[0147] The local computing subsystem can be a local cluster composed of multiple edge computing devices, such as a micro data center or an edge cluster, which can share computing tasks.

[0148] Specifically, each edge computing device is equipped with a condition evaluation module, which can evaluate computing resource conditions and network conditions by itself.

[0149] Furthermore, the primary edge computing device also includes a data receiving module (107) and a display module (108);

[0150] The data receiving module (107) is used to receive the second comprehensive diagnosis result fed back by the cloud (2);

[0151] The display module (108) is respectively connected to the data receiving module (107), the preliminary diagnosis module (103) and the comprehensive diagnosis module (105), and is used to display the output preliminary diagnosis result, the first comprehensive diagnosis result and the second comprehensive diagnosis result.

[0152] The diagnostic results are presented to clinicians for their reference and decision-making.

[0153] The display screen of handheld ultrasound equipment or other mobile devices is usually limited. Therefore, the diagnosis results are displayed in a reasonable layout on the limited display screen, and the information is displayed in a combination of graphics and text. The display module also supports operations such as zooming in, zooming out, and sliding browsing of the marked ultrasound images and diagnosis results, which is convenient for users to operate and view detailed information. In addition, the display module can use visualization technologies such as color coding and heat maps to intuitively display the location and severity of the lesion. Improve user experience and diagnostic efficiency, so that users can quickly understand and adopt diagnostic results.

[0154] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A system for assisting diagnosis of diffuse thyroid lesions and nodules. It is characterized in that include: The local computing subsystem is used for edge computing, including: A data acquisition module, used for acquiring first patient data which is thyroid ultrasound data; A preprocessing module, connected to the data acquisition module, for preprocessing the ultrasound data; A preliminary diagnosis module, connected to the preprocessing module, for extracting first feature data from the ultrasound data based on a lightweight preliminary recognition diagnosis model and forming a preliminary diagnosis result on diffuse thyroid lesions and nodules; A condition evaluation module is used to evaluate the local computing resource conditions and obtain condition evaluation results; A comprehensive diagnosis module, connected to the condition evaluation module and the preliminary diagnosis module respectively, for executing a comprehensive diagnosis task when the condition evaluation result indicates that the local computing resource condition is sufficient, and processing the first feature data based on a pre-trained first comprehensive model to form a first comprehensive diagnosis result output on diffuse thyroid lesions and nodules; A data transmission module, connected to the condition evaluation module and the preliminary diagnosis module respectively, and configured to transmit the first feature data to the cloud when the condition evaluation result indicates that the local computing resource condition is insufficient; The cloud is used to perform comprehensive diagnosis tasks, process the first feature data based on a pre-trained second comprehensive model to form a second comprehensive diagnosis result about diffuse thyroid lesions and nodules, and feed it back to the local computing subsystem.

2. A thyroid diffuse lesion and nodule auxiliary diagnosis system as claimed in claim 1, It is characterized in that The preliminary identification and diagnosis model includes a lightweight first feature identification model and a lightweight preliminary diagnosis model; The preliminary diagnosis module extracts first feature data from the preprocessed ultrasound data based on the trained first feature recognition model, and processes the first feature data based on the trained preliminary diagnosis model to form a preliminary diagnosis result about diffuse thyroid lesions and nodules.

3. The thyroid diffuse lesions and nodules auxiliary diagnosis system as claimed in claim 1, It is characterized in that The first comprehensive model includes a second feature recognition model and a first comprehensive diagnosis model; When the condition evaluation result indicates that the local computing resource condition is sufficient, the comprehensive diagnosis module is used to: Processing the first feature data based on the trained second feature recognition model to extract second feature data of the thyroid gland; The second feature data is processed based on a pre-trained first comprehensive diagnosis model to form the first comprehensive diagnosis result output.

4. A thyroid diffuse lesion and nodule auxiliary diagnosis system as claimed in claim 3, It is characterized in that The second comprehensive model includes a third feature recognition model and a second comprehensive diagnosis model; When the cloud receives the first characteristic data transmitted by the data transmission module, the cloud is used to: Processing the first feature data based on the trained third feature recognition model to extract third feature data of the thyroid gland; The third feature data is processed based on a pre-trained second comprehensive diagnosis model to form the second comprehensive diagnosis result, and the second comprehensive diagnosis result is fed back to the local computing subsystem.

5. A thyroid diffuse lesion and nodule auxiliary diagnosis system as claimed in claim 4, It is characterized in that The data acquisition module is also used to acquire existing second patient data related to the patient's thyroid gland, and the types of the second patient data include medical history data and test examination data; The preprocessing module is also used to preprocess the collected second patient data; The comprehensive diagnosis module is also connected to the preprocessing module, and when the condition evaluation result indicates that the local computing resource condition is sufficient, the first feature data and the preprocessed second patient data are processed based on the first comprehensive diagnosis model to form the first comprehensive diagnosis result; The data transmission module is also connected to the preprocessing module, and when the condition evaluation result is that the local computing resource condition is insufficient, the first feature data and the preprocessed second patient data are transmitted to the cloud; The cloud processes the first feature data and the second patient data based on the second comprehensive diagnosis model to form the second comprehensive diagnosis result, and feeds back the second comprehensive diagnosis result to the local computing subsystem.

6. A thyroid diffuse lesion and nodule auxiliary diagnosis system as claimed in claim 5, It is characterized in that The condition evaluation module is used to simultaneously evaluate the local computing resource conditions and the network conditions to obtain the condition evaluation result; When the condition evaluation result is that the local computing resource condition is insufficient but the network condition meets the predetermined standard, the data transmission module transmits the first feature data and the pre-processed second patient data to the cloud; When the condition evaluation result is that the local computing resource condition is insufficient and the network condition does not meet the predetermined standard, the data transmission module locally caches the first feature data and the pre-processed second patient data, and when the network condition changes to meet the predetermined standard, continues to transmit the unfinished data in a breakpoint resume mode; When the network condition does not meet the predetermined standard, the cloud will delay and wait for the network condition to meet the predetermined standard and then complete data reception before executing the comprehensive diagnosis task.

7. A thyroid diffuse lesion and nodule auxiliary diagnosis system as claimed in claim 5, It is characterized in that Both the first comprehensive diagnostic result and the second comprehensive diagnostic result include auxiliary diagnostic conclusions and auxiliary diagnostic suggestions about diffuse thyroid lesions and nodules, and the auxiliary diagnostic suggestions include examination suggestions, treatment suggestions, and follow-up suggestions.

8. The system for auxiliary diagnosis of diffuse thyroid lesions and nodules according to claim 5, It is characterized in that The comprehensive diagnosis module is also used to perform thyroid cancer risk assessment on the first feature data and the second patient data based on the first comprehensive diagnosis model, and output the assessed risk probability of thyroid cancer attached to the first comprehensive diagnosis result; The cloud is also used to perform thyroid cancer risk assessment for the first feature data and the second patient data based on the second comprehensive diagnostic model, and feed back the assessed risk probability of thyroid cancer to the local computing subsystem in the second comprehensive diagnostic result.

9. The thyroid diffuse lesions and nodules auxiliary diagnosis system as claimed in claim 1, It is characterized in that The local computing subsystem is an edge computing device or is composed of multiple edge computing devices; When the local computing subsystem includes a plurality of edge computing devices, each of the edge computing devices includes the data acquisition module, the preprocessing module, the comprehensive diagnosis module, the preliminary diagnosis module and the data transmission module; The primary edge computing device initiates the diagnostic task, and the rest are auxiliary edge computing devices. The local computing subsystem allocates diagnostic tasks according to the computing resource conditions of the primary edge computing device and the auxiliary edge computing device.

10. A thyroid diffuse lesion and nodule auxiliary diagnosis system as claimed in claim 9, It is characterized in that The primary edge computing device also includes a data receiving module and a display module; The data receiving module is used to receive the second comprehensive diagnosis result fed back by the cloud; The display module is respectively connected to the data receiving module, the preliminary diagnosis module and the comprehensive diagnosis module, and is used to display the outputted preliminary diagnosis result, the first comprehensive diagnosis result and the second comprehensive diagnosis result.

Citation Information

Cited By

  • Wounded person positioning method and system based on neural network prediction

    CN119920473A

  • Automatic nodule benign and malignant scoring method and system based on thyroid ultrasound image

    CN120070438A