Snake wound diagnosis device and method

Abstract

The invention discloses a snake wound diagnosis device and method, belongs to the technical field of snake wound diagnosis, and adopts an integrated architecture in which a local edge computing node and a private cloud platform based on FPGA (Field Programmable Gate Array) are coordinated to adapt to a weak network or offline operation environment. The data acquisition module is used for acquiring a wound image, symptom information and treatment region information of a patient; and the preprocessing module is connected with the data acquisition module. According to the snake wound diagnosis device and method, edge computing and a local private cloud architecture are combined, core diagnosis logic and data are both completed locally, dependence on a continuous high-bandwidth network is eliminated, off-line operation is supported, a basic health center with weak network conditions is adapted, blurred images are repaired through FPGA hardware acceleration and an efficient algorithm model in cooperation with a DiffMIC denoising module, and the snake wound diagnosis efficiency is improved. And in combination with the bimodal interactive attention model, the complex scene diagnosis accuracy is improved, the diagnosis time is shortened, and the misdiagnosis rate caused by image blurring is reduced.

Description

Technical Field

[0001] This invention relates to the field of snakebite diagnosis technology, and in particular to a snakebite diagnosis device and method. Background Technology

[0002] Snakebite is a global public health problem. Traditional snakebite diagnosis relies heavily on the experience of specialists, but primary healthcare institutions generally face challenges such as a shortage of specialists, outdated diagnostic equipment, and weak network infrastructure, leading to a high risk of misdiagnosis and missed diagnosis, as well as a long diagnostic cycle.

[0003] In the existing technology, some cloud-based artificial intelligence-assisted diagnostic solutions have begun to emerge. However, these solutions rely on high-speed and stable networks, making them unsuitable for weak network conditions or offline scenarios at the grassroots level. Uploading data to public clouds poses a high risk of privacy leaks and makes it difficult to pass ethical and security reviews. Using general models results in low accuracy in identifying regionally unique snake species such as the Guangxi cobra and the Hunan viper. Furthermore, they struggle to process blurry wound images taken by non-professionals at the grassroots level, leading to a high risk of misdiagnosis. Summary of the Invention

[0004] The technical problem this invention aims to solve is that relying on high-speed and stable networks makes it unsuitable for weak network conditions or offline scenarios at the grassroots level. Uploading data to public clouds poses a high risk of privacy leaks and makes it difficult to pass ethical and security reviews. The accuracy of using general models to identify regionally unique snake species such as the Guangxi cobra and the Hunan viper is low. Furthermore, it is difficult to process blurry wound images taken by non-professionals at the grassroots level, which is prone to misdiagnosis. To address these issues, this invention provides a snakebite diagnosis device and method that can adapt to grassroots conditions, ensure data security, provide rapid and accurate diagnosis, and has regional adaptability for intelligent snakebite diagnosis.

[0005] To solve the above problems, the following technical solutions are provided:

[0006] A device and method for diagnosing snakebites are designed, employing an integrated architecture based on FPGA-based local edge computing nodes and a private cloud platform to adapt to weak network or offline operating environments. The device includes:

[0007] The data acquisition module is used to acquire the patient's wound images, symptom information, and location information of the medical treatment.

[0008] The preprocessing module is connected to the data acquisition module and is used to denoise and enhance the wound image and standardize the symptom information. It has a built-in self-developed DiffMIC denoising unit based on U-Net and attention mechanism to improve the clarity of low-quality or blurry wound images.

[0009] The diagnostic module, which is connected to the preprocessing module, is used to perform diagnostic analysis based on the processed data. It is equipped with a dual-modal interactive attention model that integrates image features and text features.

[0010] A dynamic regional snake species database is connected to the diagnostic module. It stores multi-dimensional snake species characteristic data related to the region and provides the diagnostic module with regional diagnostic basis to improve the accuracy of identifying regional snake species.

[0011] An edge computing unit, which integrates an FPGA acceleration chip, is used to deploy and accelerate the model inference of the diagnostic module locally to achieve rapid diagnosis.

[0012] An encrypted storage module, which employs AES-256 encryption technology, is used to achieve localized encrypted storage and processing of the wound images, symptom information, and diagnostic results.

[0013] An output and interaction module is used to generate and output diagnostic reports and supports interaction with users.

[0014] The device as a whole supports low-cost deployment and low-power operation.

[0015] The above technical solution combines edge computing and local private cloud architecture, with core diagnostic logic and data completed locally, eliminating the dependence on continuous high-bandwidth networks, supporting offline operation, and adapting to primary health care centers with weak network conditions. Through FPGA hardware acceleration and efficient algorithm models, combined with the DiffMIC denoising module to repair blurred images, and combined with a bimodal interactive attention model, it improves the diagnostic accuracy in complex scenarios, reduces diagnostic time, and reduces the misdiagnosis rate caused by image blur.

[0016] Furthermore, the data acquisition module includes:

[0017] The image acquisition unit uses a 1080P high-definition camera that supports a shooting distance of 0.3-1.5m;

[0018] Symptom input unit: The symptom input unit has a pre-set graphical human-computer interaction interface containing 12 core symptom templates and 36 specific options to simplify symptom entry.

[0019] A geographic location unit, which is used to automatically or manually input the location information for medical treatment;

[0020] The illumination compensation unit is an LED fill light with a color temperature of 5500K and adjustable brightness, used to provide a standard lighting environment during image acquisition.

[0021] Furthermore, the preprocessing module also includes:

[0022] An image enhancement unit employs a contrast-limited adaptive histogram equalization (CLAHE) algorithm to optimize the contrast of the wound image.

[0023] The text normalization unit, based on a bidirectional long short-term memory network (BiLSTM) model, is used to automatically segment and encode the symptom information.

[0024] Furthermore, the bimodal interaction attention model includes:

[0025] The image feature extraction submodule is implemented based on the ResNet50 convolutional neural network.

[0026] The text feature extraction submodule is implemented based on the BERT-base pre-trained model.

[0027] The cross-modal attention fusion submodule dynamically fuses feature information from the two modalities by calculating the attention weight matrix between image features and text features, generating a comprehensive feature vector for snake species identification and poisoning type determination.

[0028] Furthermore, the dynamic regional snake species database covers data on at least 58 common venomous snakes from 15 provinces with a high incidence of snakebites. Each data entry includes multiple dimensions such as snake species name, geographical distribution, morphological characteristics, typical wound morphology, poisoning symptoms, toxicity type, recommended first aid measures, and antivenom serum information. The snake species database supports receiving external update packages in a secure manner for quarterly iterations or emergency updates of data and associated model parameters.

[0029] Furthermore, the device supports offline operation of all core diagnostic functions in a network-disconnected environment; the output and interaction module is compatible with PC terminals running Windows 10 and above operating systems, as well as mobile terminals running Android 8.0 or iOS 12.0 and above operating systems.

[0030] Furthermore, a method for diagnosing snakebites, applied to the aforementioned device, the method comprising:

[0031] S1: The data acquisition module is used to acquire images of the patient's wounds, symptom information, and location information of the patient's medical treatment.

[0032] S2: The preprocessing module performs denoising processing based on the DiffMIC algorithm and enhancement processing based on the CLAHE algorithm on the wound image, and at the same time performs standardized encoding based on the BiLSTM model on the symptom information.

[0033] S3: Based on the treatment location information, retrieve the corresponding regional snake species data from the dynamic regional snake species database as prior knowledge;

[0034] S4: The bimodal interactive attention model in the diagnostic module fuses processed image features, text features, and regional snake species data, and the FPGA in the edge computing unit performs accelerated inference to generate a diagnostic result that includes snake species identification, poisoning type, and treatment suggestions.

[0035] S5: Generate a diagnostic report through the output and interaction module, and encrypt and store all relevant data through the encryption storage module;

[0036] S6: Update the parameters of the dynamic regional snake species database and the bimodal interactive attention model by connecting to an external collaborative update system.

[0037] Furthermore, in step S4, the snake species identification results have an overall identification accuracy of no less than 95% for common snake species covered in the dynamic regional snake species database, and an identification accuracy of no less than 92% for regionally unique snake species.

[0038] Furthermore, the update in step S6 specifically includes: the hospital submitting de-identified and anonymized case data, the research institution retraining and labeling the model based on the new data, the enterprise packaging and generating an encrypted update package, and the device importing the update package locally via a USB interface or remotely downloading the update package through a secure network channel to complete the update.

[0039] Furthermore, the diagnostic report generated in step S5 includes at least: snake species name, identification confidence level, type of poisoning, emergency treatment procedure, recommended antivenom serum and referral indications, and supports paper printing and PDF electronic document export.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. The snakebite diagnosis device and method combine edge computing and local private cloud architecture. The core diagnostic logic and data are completed locally, eliminating the dependence on continuous high-bandwidth networks. It supports offline operation and is suitable for primary health care centers with weak network conditions. Through FPGA hardware acceleration and efficient algorithm models, combined with the DiffMIC denoising module to repair blurred images, and combined with the bimodal interactive attention model, it improves the diagnostic accuracy in complex scenarios, reduces diagnostic time, and reduces the misdiagnosis rate caused by image blur.

[0042] 2. In this snakebite diagnosis device and method, all sensitive medical data is encrypted (AES-256) and stored and processed locally on the device, without the need to upload to an external cloud, fundamentally eliminating the risk of data leakage. It is easy to pass hospital ethics review, dispelling the core concerns of primary medical institutions about data privacy. At the same time, it has a built-in dynamic regional snake species database, covering 58 venomous snakes in 15 high-incidence provinces, and has optimized parameters for 23 regional snake species to solve the problem of general model incompatibility and improve regional diagnosis efficiency.

[0043] 3. The device and method for diagnosing snake bites have low deployment costs and low power consumption. They are compatible with older PCs and mobile terminals at the grassroots level, and require little time for medical staff to become proficient, thus alleviating the shortage of specialist doctors at the grassroots level. By building a collaborative ecosystem of "universities-hospitals-enterprises", the device can achieve secure desensitized collection of case data, iterative training of models, and parameter updates, enabling the device to continuously optimize its performance as clinical data accumulates. Attached Figure Description

[0044] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0045] Figure 1 This is a system architecture block diagram of the snakebite diagnosis device of the present invention;

[0046] Figure 2 This is a schematic diagram of the data acquisition and preprocessing process of the present invention;

[0047] Figure 3 This is a schematic diagram illustrating the working principle of the bimodal interactive attention model of the present invention.

[0048] Figure 4 This is a flowchart of the snakebite diagnosis method of the present invention;

[0049] Figure 5 This is a schematic diagram of the external collaborative update mechanism of the device in an embodiment of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] like Figure 1 - Figure 5As shown, this embodiment provides a snakebite diagnosis device and method, which adopts an integrated architecture based on FPGA-based local edge computing nodes and a private cloud platform to adapt to weak network or offline operating environments. The device includes:

[0052] The data acquisition module is used to acquire the patient's wound images, symptom information, and location information of the medical treatment.

[0053] The data acquisition module includes:

[0054] The image acquisition unit uses a 1080P high-definition camera that supports a shooting distance of 0.3-1.5m;

[0055] Symptom input unit: The symptom input unit has a pre-set graphical human-computer interaction interface containing 12 core symptom templates and 36 specific options to simplify symptom entry.

[0056] A geographic location unit, which is used to automatically or manually input the location information for medical treatment;

[0057] The illumination compensation unit is an LED fill light with a color temperature of 5500K and adjustable brightness, used to provide a standard illumination environment during image acquisition.

[0058] The preprocessing module is connected to the data acquisition module and is used to denoise and enhance the wound image and standardize the symptom information. It has a built-in self-developed DiffMIC denoising unit based on U-Net and attention mechanism to improve the clarity of low-quality or blurry wound images.

[0059] The preprocessing module further includes:

[0060] An image enhancement unit employs a contrast-limited adaptive histogram equalization (CLAHE) algorithm to optimize the contrast of the wound image.

[0061] The text normalization unit, based on a bidirectional long short-term memory network (BiLSTM) model, is used to automatically segment and encode the symptom information into words and structured features.

[0062] The diagnostic module, which is connected to the preprocessing module, is used to perform diagnostic analysis based on the processed data. It is equipped with a dual-modal interactive attention model that integrates image features and text features.

[0063] The bimodal interaction attention model includes:

[0064] The image feature extraction submodule is implemented based on the ResNet50 convolutional neural network.

[0065] The text feature extraction submodule is implemented based on the BERT-base pre-trained model.

[0066] The cross-modal attention fusion submodule dynamically fuses feature information from the two modalities by calculating the attention weight matrix between image features and text features, generating a comprehensive feature vector for snake species identification and poisoning type determination.

[0067] A dynamic regional snake species database, connected to the diagnostic module, stores multi-dimensional snake species characteristic data related to specific regions and provides the diagnostic module with regional diagnostic basis to improve the accuracy of identifying regionally unique snake species. The database covers at least 58 common venomous snakes from 15 provinces with high snakebite incidence. Each data entry includes multiple dimensions such as snake species name, geographical distribution, morphological characteristics, typical wound morphology, poisoning symptoms, toxicity type, recommended first aid measures, and antivenom serum information. The database supports securely receiving external update packages for quarterly iterations or emergency updates of data and associated model parameters.

[0068] An edge computing unit, which integrates an FPGA acceleration chip, is used to deploy and accelerate the model inference of the diagnostic module locally to achieve rapid diagnosis.

[0069] An encrypted storage module, which employs AES-256 encryption technology, is used to achieve localized encrypted storage and processing of the wound images, symptom information, and diagnostic results.

[0070] An output and interaction module is used to generate and output diagnostic reports and supports interaction with users.

[0071] The device as a whole supports low-cost deployment and low-power operation. The device supports offline operation of all core diagnostic functions in a network-disconnected environment. The output and interaction module is compatible with PC terminals running Windows 10 and above, as well as mobile terminals running Android 8.0 or iOS 12.0 and above.

[0072] In this embodiment, the method, applied to a snakebite diagnosis device as described above, includes:

[0073] S1: The data acquisition module is used to acquire images of the patient's wounds, symptom information, and location information of the patient's medical treatment.

[0074] S2: The preprocessing module performs denoising processing based on the DiffMIC algorithm and enhancement processing based on the CLAHE algorithm on the wound image, and at the same time performs standardized encoding based on the BiLSTM model on the symptom information.

[0075] S3: Based on the treatment location information, retrieve the corresponding regional snake species data from the dynamic regional snake species database as prior knowledge;

[0076] S4: Through the bimodal interactive attention model in the diagnostic module, the processed image features, text features and the regional snake species data are fused, and the FPGA in the edge computing unit performs accelerated inference to generate a diagnostic result containing snake species identification results, poisoning type and treatment suggestions. The snake species identification result has an overall identification accuracy of not less than 95% for common snake species covered in the dynamic regional snake species database, and an identification accuracy of not less than 92% for regional unique snake species.

[0077] S5: A diagnostic report is generated through the output and interaction module, and all relevant data is encrypted and stored through the encrypted storage module. The diagnostic report includes at least: snake species name, identification confidence level, poisoning type, emergency treatment procedure, recommended antivenom serum and referral indications, and supports paper printing and PDF electronic document export.

[0078] S6: By connecting to an external collaborative update system, the parameters of the dynamic regional snake species database and the bimodal interactive attention model are updated. The update specifically includes: the hospital submits de-identified and anonymized case data, the research institution retrains and labels the model based on the new data, the enterprise encapsulates and generates an encrypted update package, and the device imports the update package locally via USB interface or downloads it remotely via a secure network channel to complete the update.

[0079] This embodiment describes the workflow of the device during a complete diagnostic procedure:

[0080] When a patient visits the facility, medical staff activate the device, take a picture of the wound using the camera, and select the patient's clinical symptoms on the touchscreen. After confirming the location as Hunan province, the device automatically initiates a preprocessing procedure: the DiffMIC unit processes the wound image, and the CLAHE algorithm enhances contrast. The selected symptoms are encoded as feature vectors. Based on the Hunan geographical information, the system loads data on local snake species such as vipers and bamboo pit vipers from a dynamic snake species database. The FPGA-accelerated bimodal model initiates inference: fusing the processed wound image features and symptom text features, and combining this with prior knowledge of the "Hunan region" for comprehensive judgment. Approximately 110 seconds later, the touchscreen displays the diagnosis: "Presumed causative snake species: short-tailed viper (confidence: 96%); poisoning type: primarily hemotoxic; recommendation: immediate injection of anti-viper venom serum and monitoring of coagulation function." Simultaneously, a complete report containing detailed processing procedures is generated. All image, text data, and diagnostic results are stored locally in encrypted form. The anonymized data (removing personal identification information) for this diagnosis can be tagged for subsequent model updates.

[0081] This embodiment describes the collaborative update mechanism of the device. To maintain the device's advanced nature and adaptability, a sustainable collaborative update system involving universities, hospitals, and enterprises is established. At the hospital level: partner hospitals can voluntarily submit de-identified snakebite case data (encrypted) to the regional data center via a secure channel during clinical use. At the university / research institution level: researchers use the new data to incrementally train and optimize the DiffMIC model and bimodal model, adjusting parameters specifically for newly discovered regional snake species characteristics or diagnostic difficulties. At the enterprise level: the trained new model parameters and updated snake species database data are packaged into an encrypted update package with digital signature authentication. Update distribution: primary healthcare institutions can update in two ways: first, automatically download via a secure 5G / broadband channel when network access is available; second, the update package can be distributed offline by higher-level units via physical media such as USB drives. After verifying the signature, the device automatically installs the update, completing one iteration.

[0082] In the description of this invention, it should be understood that the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two elements; they can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0083] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A device for diagnosing snakebite, characterized in that, The device employs an integrated architecture that combines FPGA-based local edge computing nodes with a private cloud platform to adapt to weak network or offline operating environments. The device includes: The data acquisition module is used to acquire the patient's wound images, symptom information, and location information of the medical treatment. The preprocessing module is connected to the data acquisition module and is used to denoise and enhance the wound image and standardize the symptom information. It has a built-in self-developed DiffMIC denoising unit based on U-Net and attention mechanism to improve the clarity of low-quality or blurry wound images. The diagnostic module, which is connected to the preprocessing module, is used to perform diagnostic analysis based on the processed data. It is equipped with a dual-modal interactive attention model that integrates image features and text features. A dynamic regional snake species database is connected to the diagnostic module. It stores multi-dimensional snake species characteristic data related to the region and provides the diagnostic module with regional diagnostic basis to improve the accuracy of identifying regional snake species. An edge computing unit, which integrates an FPGA acceleration chip, is used to deploy and accelerate the model inference of the diagnostic module locally to achieve rapid diagnosis. An encrypted storage module, which employs AES-256 encryption technology, is used to achieve localized encrypted storage and processing of the wound images, symptom information, and diagnostic results. An output and interaction module is used to generate and output diagnostic reports and supports interaction with users. The device as a whole supports low-cost deployment and low-power operation.

2. The snakebite diagnosis device according to claim 1, characterized in that, The data acquisition module includes: The image acquisition unit uses a 1080P high-definition camera that supports a shooting distance of 0.3-1.5m; Symptom input unit: The symptom input unit has a pre-set graphical human-computer interaction interface containing 12 core symptom templates and 36 specific options to simplify symptom entry. A geographic location unit, which is used to automatically or manually input the location information for medical treatment; The illumination compensation unit is an LED fill light with a color temperature of 5500K and adjustable brightness, used to provide a standard lighting environment during image acquisition.

3. The snakebite diagnosis device according to claim 1, characterized in that, The preprocessing module further includes: An image enhancement unit employs a contrast-limited adaptive histogram equalization (CLAHE) algorithm to optimize the contrast of the wound image. The text normalization unit, based on a bidirectional long short-term memory network (BiLSTM) model, is used to automatically segment and encode the symptom information.

4. The snakebite diagnosis device according to claim 1, characterized in that, The bimodal interaction attention model includes: The image feature extraction submodule is implemented based on the ResNet50 convolutional neural network. The text feature extraction submodule is implemented based on the BERT-base pre-trained model. The cross-modal attention fusion submodule dynamically fuses feature information from the two modalities by calculating the attention weight matrix between image features and text features, generating a comprehensive feature vector for snake species identification and poisoning type determination.

5. The snakebite diagnosis device according to claim 1, characterized in that, The dynamic regional snake species database covers data on at least 58 common venomous snakes from 15 provinces with a high incidence of snakebites. Each data entry includes multiple dimensions such as snake species name, geographical distribution, morphological characteristics, typical wound morphology, poisoning symptoms, toxicity type, recommended first aid measures, and antivenom serum information. The snake species database supports receiving external update packages in a secure manner for quarterly iterations or emergency updates of data and associated model parameters.

6. The snakebite diagnosis device and method according to claim 1, characterized in that, The device supports offline operation of all core diagnostic functions in a network-disconnected environment; the output and interaction module is compatible with PC terminals running Windows 10 and above, as well as mobile terminals running Android 8.0 or iOS 12.0 and above.

7. A method for diagnosing snakebite, characterized in that, The method, applied to the snakebite diagnostic device as described in any one of claims 1-6, comprises: S1: The data acquisition module is used to acquire images of the patient's wounds, symptom information, and location information of the patient's medical treatment. S2: The preprocessing module performs denoising processing based on the DiffMIC algorithm and enhancement processing based on the CLAHE algorithm on the wound image, and at the same time performs standardized encoding based on the BiLSTM model on the symptom information. S3: Based on the treatment location information, retrieve the corresponding regional snake species data from the dynamic regional snake species database as prior knowledge; S4: The bimodal interactive attention model in the diagnostic module fuses processed image features, text features, and regional snake species data, and the FPGA in the edge computing unit performs accelerated inference to generate a diagnostic result that includes snake species identification, poisoning type, and treatment suggestions. S5: Generate a diagnostic report through the output and interaction module, and encrypt and store all relevant data through the encryption storage module; S6: Update the parameters of the dynamic regional snake species database and the bimodal interactive attention model by connecting to an external collaborative update system.

8. The method for diagnosing snakebites according to claim 7, characterized in that, In step S4, the snake species identification results shall have an overall identification accuracy of no less than 95% for common snake species covered in the dynamic regional snake species database, and an identification accuracy of no less than 92% for regionally unique snake species.

9. The method for diagnosing snakebites according to claim 7, characterized in that, The update in step S6 specifically includes: the hospital submits de-identified and anonymized case data, the research institution retrains and labels the model based on the new data, the enterprise packages and generates an encrypted update package, and the device imports the update package locally via USB interface or downloads it remotely via a secure network channel to complete the update.

10. The method for diagnosing snakebites according to claim 7, characterized in that, The diagnostic report generated in step S5 includes at least: snake species name, identification confidence level, type of poisoning, emergency treatment procedure, recommended antivenom serum and referral indications, and supports paper printing and PDF electronic document export.

Images