Blood sample automatic analysis system based on artificial intelligence

Abstract

The invention provides a blood sample automatic analysis system based on artificial intelligence. The blood sample automatic analysis system based on artificial intelligence comprises a sample collection module used for automatically collecting a blood sample of a patient, a data acquisition module comprising a sensor and an imaging device and used for acquiring an image of the blood sample or original data in other forms, and a data preprocessing module, and the noise filtering module is used for carrying out noise filtering, standardization and normalization processing on the original data. According to the blood sample automatic analysis system based on artificial intelligence, by introducing a dynamic adjustment mechanism and a self-adaptive learning algorithm, the system can adjust model parameters in real time, and especially under the condition that data is insufficient or a data set has deviation, the recognition accuracy of rare diseases is greatly improved. The dynamic model training method not only improves the generalization ability of the model, but also effectively relieves the limitation of relying on a large-scale annotation data set.

Description

Technical Field

[0001] The present invention relates to the technical field of blood sample analysis, and in particular to an automatic blood sample analysis system based on artificial intelligence. Background Art

[0002] The AI-based automatic blood sample analysis system usually includes the following main modules: sample collection, data preprocessing, feature extraction, model training and reasoning, and result output. The system first collects blood samples from the patient through automated equipment and uses sensors or imaging devices to obtain images or other forms of raw data of the samples. Next, the data preprocessing module processes these raw data, including steps such as noise filtering, standardization, and normalization to ensure the quality and consistency of the data. The system enters the feature extraction stage and uses deep learning or other machine learning algorithms to extract key features from the preprocessed data, such as cell morphology, color distribution, and other biomarkers. These features will be fed as input into a pre-trained AI model, which is trained based on a large amount of labeled data and can identify various diseases or abnormalities. Finally, the output results of the model will be interpreted and further analyzed, and presented to doctors or researchers in the form of charts or reports.

[0003] The AI-based automated blood sample analysis system has significant advantages in accuracy and efficiency, but there are some drawbacks. The accuracy of the system is highly dependent on the quality and diversity of the training data. If the training data is insufficient or the data set is biased, the model may not correctly identify certain rare or complex diseases in practical applications. The "black box" nature of the AI ​​system makes the results difficult to interpret, and medical staff may not fully understand or trust the diagnosis given by the system. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides an automatic blood sample analysis system based on artificial intelligence, which solves the problem that the accuracy of the system is highly dependent on the quality and diversity of training data; the "black box" nature of the artificial intelligence system makes the results difficult to interpret, and medical staff may not be able to fully understand or trust the diagnostic results given by the system.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an automatic blood sample analysis system based on artificial intelligence, comprising:

[0008] The sample collection module is used to automatically collect blood samples from patients. In order to effectively distinguish different cell types in the blood, the data acquisition module should be equipped with a multispectral imaging device, which can distinguish the morphology and physical and chemical properties of red blood cells, white blood cells and platelets through spectra of different wavelengths;

[0009] Data acquisition module, including sensors and imaging devices, used to obtain images or other forms of raw data of blood samples, cell classification algorithms:

[0010] Red blood cells: Since red blood cells have a relatively regular shape in images, algorithms based on morphological analysis are usually used for identification and counting. Convolutional neural networks can be trained to recognize the typical biconcave shape of red blood cells and use Gaussian filters to remove noise.

[0011] White blood cells: White blood cells include multiple subtypes with different morphologies and functions. To address this, CNN can be combined with regional convolutional networks to accurately locate and classify different white blood cells. The attention mechanism can be further used to highlight cell features with diagnostic value, such as nuclear morphology and granular structure.

[0012] Platelets: Platelets are small and easy to aggregate, making them difficult to identify. Clustering-based algorithms and deep learning models can be combined to improve the accuracy of platelet identification.

[0013] The data preprocessing module is used to perform noise filtering, standardization and normalization on the raw data, wherein a Gaussian filter is used to perform noise filtering on the image data:

[0014]

[0015] And the image is normalized by the standardization formula:

[0016]

[0017] The feature extraction module uses a deep learning-based algorithm to extract key features from the preprocessed data, where a convolutional neural network (CNN) is used to extract features:

[0018] f l =σ(W l *f l-1 +b l )

[0019] And combined with the attention mechanism to weight key features:

[0020]

[0021] Model training and reasoning module: The model is trained through a large-scale and diverse annotated data set, and a dynamic adjustment mechanism is used to improve the recognition accuracy of rare diseases. The pre-training parameters are fine-tuned using the transfer learning method:

[0022]

[0023] And perform hyperparameter tuning via Bayesian optimization:

[0024]

[0025] The explainability analysis module uses Grad-CAM to generate heat maps, associate diagnostic results with key features, and generate visual explanation images or reports:

[0026]

[0027] The result output module is used to present the analysis results to medical staff in the form of charts or reports;

[0028] Database module, used to store blood sample data, feature information, training models, and analysis results, and supports real-time update and management;

[0029] The security and privacy protection module ensures the security of patient data during storage and processing based on encryption technology and access control mechanisms. Homomorphic encryption technology is used to implement data calculation in an encrypted state:

[0030] E(x+y)=E(x)×E(y),E(x·y)=E(x) y

[0031] Among them, the system also includes a dynamic model training method based on an adaptive learning algorithm, which adjusts model parameters in real time by introducing new data samples to improve the ability to identify rare diseases in cases of insufficient data or data set bias.

[0032] Preferably, the data acquisition module further includes a multimodal data acquisition unit for simultaneously acquiring data in various forms of the blood sample, and the multimodal data is subjected to feature fusion processing to improve the feature extraction efficiency and accuracy of the feature extraction module.

[0033] Preferably, the data preprocessing module includes a submodule based on a noise suppression algorithm for filtering environmental noise and system noise in sample data, and the feature extraction module uses a convolutional neural network to extract features from image data and performs weighted processing on key features through an attention mechanism.

[0034] Preferably, the model training and reasoning module adopts a transfer learning method to apply the pre-trained model to a new data set to improve the generalization ability of the model under small sample data. The dynamic model training method includes a hyperparameter adjustment mechanism based on Bayesian optimization to improve the recognition accuracy of the model with optimal configuration.

[0035] Preferably, the hyperparameter adjustment mechanism dynamically updates the hyperparameters according to the performance indicators of the model to cope with the diversity of different sample data, and the interpretability analysis module uses a visualization method based on Grad-CAM to display the decision-making process of the model in the form of a heat map.

[0036] Preferably, the explanation report generated by the explainability analysis module includes a credibility score of the model decision and a detailed description of key features, and the result output module has a customizable function, allowing medical personnel to adjust the format and content of the result display according to needs.

[0037] Preferably, the database module supports a federated learning framework, allowing multi-center data collaborative training without sharing patient data, wherein a federated averaging algorithm is used to aggregate model parameters:

[0038]

[0039] Preferably, the security and privacy protection module uses homomorphic encryption technology to calculate and analyze encrypted data without decryption, and the system integrates a self-diagnosis module to monitor the operating status of the system and automatically trigger a troubleshooting procedure when potential problems are found.

[0040] (III) Beneficial effects

[0041] The present invention provides an automatic blood sample analysis system based on artificial intelligence. It has the following beneficial effects:

[0042] By introducing a dynamic adjustment mechanism and adaptive learning algorithm, the system can adjust model parameters in real time, especially when data is insufficient or the data set is biased, greatly improving the accuracy of identifying rare diseases. This dynamic model training method not only improves the generalization ability of the model, but also effectively alleviates the limitations of relying on large-scale annotated data sets. At the same time, the system fine-tunes the model and optimizes hyperparameters through transfer learning and Bayesian optimization, so that the recognition accuracy of the model can be maintained at a high level even in a small sample data environment.

[0043] The present invention also has unique innovations in terms of system interpretability and data security. The interpretability analysis module uses Grad-CAM to generate heat maps, presents the decision-making process of the model in a visual way, and provides a credibility score and a detailed description of key features in the form of an explanation report, which solves the "black box" problem of the artificial intelligence system and enhances the medical staff's trust in the diagnosis results. The system uses homomorphic encryption technology and a federated learning framework to ensure data calculation and multi-center collaborative training without decrypting the data, which protects the privacy and security of patient data, while improving the reliability of the system and its feasibility in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the internal structure of the present invention. DETAILED DESCRIPTION

[0045] 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.

[0046] like Figure 1 As shown, an embodiment of the present invention provides an automatic blood sample analysis system based on artificial intelligence, including a sample collection module for automatically collecting blood samples from patients;

[0047] The data acquisition module includes a sensor and an imaging device for acquiring images of blood samples or other forms of raw data. The data acquisition module also includes a multimodal data acquisition unit for simultaneously acquiring multiple forms of data of the blood sample. The multimodal data is processed by feature fusion to improve the feature extraction efficiency and accuracy of the feature extraction module.

[0048] The data preprocessing module is used to perform noise filtering, standardization and normalization on the original data. A Gaussian filter is used to perform noise filtering on the image data:

[0049]

[0050] And the image is normalized by the standardization formula:

[0051]

[0052] The data preprocessing module includes a submodule based on the noise suppression algorithm, which is used to filter the environmental noise and system noise in the sample data. The feature extraction module uses a convolutional neural network to extract features from the image data and performs weighted processing on key features through the attention mechanism.

[0053] The feature extraction module uses a deep learning-based algorithm to extract key features from the preprocessed data, where a convolutional neural network (CNN) is used to extract features:

[0054] f l =σ(W l *f l-1 +b l )

[0055] And combined with the attention mechanism to weight key features:

[0056]

[0057] Model training and reasoning module: The model is trained through a large-scale and diverse annotated data set, and a dynamic adjustment mechanism is used to improve the recognition accuracy of rare diseases. The pre-training parameters are fine-tuned using the transfer learning method:

[0058]

[0059] And perform hyperparameter tuning via Bayesian optimization:

[0060]

[0061] The model training and reasoning module uses transfer learning methods to apply pre-trained models to new data sets to improve the generalization ability of the model under small sample data. The dynamic model training method includes a hyperparameter adjustment mechanism based on Bayesian optimization to improve the recognition accuracy of the model with optimal configuration.

[0062] The explainability analysis module uses Grad-CAM to generate heat maps, associate diagnostic results with key features, and generate visual explanation images or reports:

[0063]

[0064] The explanation report generated by the interpretability analysis module includes the credibility score of the model decision and a detailed description of key features. The result output module has a customizable function, allowing medical staff to adjust the format and content of the result display according to their needs.

[0065] The result output module is used to present the analysis results to medical staff in the form of charts or reports;

[0066] The database module is used to store blood sample data, feature information, training models, and analysis results, and supports real-time updates and management. The hyperparameter adjustment mechanism dynamically updates hyperparameters based on the performance indicators of the model to cope with the diversity of different sample data. The interpretability analysis module uses a visualization method based on Grad-CAM to display the decision-making process of the model in the form of a heat map. The database module supports the federated learning framework, allowing multi-center data collaborative training without sharing patient data, in which the federated average algorithm is used to aggregate model parameters:

[0067]

[0068] The security and privacy protection module ensures the security of patient data during storage and processing based on encryption technology and access control mechanisms. Homomorphic encryption technology is used to implement data calculation in an encrypted state:

[0069] E(x+y)=E(x)×E(y),E(x·y)=E(x) y

[0070] Among them, the system also includes a dynamic model training method based on an adaptive learning algorithm. By introducing new data samples, the model parameters are adjusted in real time to improve the ability to identify rare diseases in cases of insufficient data or data set bias. The security and privacy protection module uses homomorphic encryption technology to calculate and analyze encrypted data without decryption. The system integrates a self-diagnosis module to monitor the system's operating status and automatically trigger a troubleshooting procedure when potential problems are found.

[0071] The sample collection module efficiently collects blood samples from patients through automated equipment, ensuring the integrity and quality of the samples and providing a reliable foundation for subsequent data processing.

[0072] The data acquisition module is responsible for capturing various forms of raw data from blood samples. This module not only includes standard sensors and imaging devices, but also integrates a multimodal data acquisition unit for simultaneously acquiring multiple data forms such as optical images, fluorescence images, and spectral data. By fusion processing of multimodal data, the efficiency and accuracy of the feature extraction module are greatly improved, thus providing richer and more diverse inputs for subsequent analysis.

[0073] The data preprocessing module performs necessary processing on the acquired raw data to remove noise and unify the format and scale of the data. Specifically, this module uses a Gaussian filter to filter the noise of the image data, thereby reducing environmental interference and system noise in the image. Standardization and normalization processing adjusts the data to a uniform range through corresponding formulas so that the subsequent machine learning model can better process the data. In addition, this module also includes a submodule based on a noise suppression algorithm to further filter the environmental noise and system noise in the data to ensure high-quality data input.

[0074] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based automatic blood sample analysis system, characterized in that: include: A sample collection module, used for automatically collecting blood samples from patients; A data acquisition module, including sensors and imaging devices, for acquiring images or other forms of raw data of blood samples; The data preprocessing module is used to perform noise filtering, standardization and normalization on the raw data, wherein a Gaussian filter is used to perform noise filtering on the image data: And the image is normalized by the standardization formula: The feature extraction module uses a deep learning-based algorithm to extract key features from the preprocessed data, where a convolutional neural network (CNN) is used to extract features: f l =σ(W l *f l-1 +b l ) And combined with the attention mechanism to weight key features: Model training and reasoning module: The model is trained through a large-scale and diverse annotated data set, and a dynamic adjustment mechanism is used to improve the recognition accuracy of rare diseases. The pre-training parameters are fine-tuned using the transfer learning method: And perform hyperparameter tuning via Bayesian optimization: The explainability analysis module uses Grad-CAM to generate heat maps, associate diagnostic results with key features, and generate visual explanation images or reports: The result output module is used to present the analysis results to medical staff in the form of charts or reports; Database module, used to store blood sample data, feature information, training models, and analysis results, and supports real-time update and management; The security and privacy protection module ensures the security of patient data during storage and processing based on encryption technology and access control mechanisms. Homomorphic encryption technology is used to implement data calculation in an encrypted state: E(x+y)=E(x)×E(y),E(x·y)=E(x) y Among them, the system also includes a dynamic model training method based on an adaptive learning algorithm, which adjusts model parameters in real time by introducing new data samples to improve the ability to identify rare diseases in cases of insufficient data or data set bias.

2. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The data acquisition module also includes a multimodal data acquisition unit for simultaneously acquiring multiple forms of data of the blood sample. The multimodal data is subjected to feature fusion processing to improve the feature extraction efficiency and accuracy of the feature extraction module.

3. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The data preprocessing module includes a submodule based on a noise suppression algorithm for filtering environmental noise and system noise in sample data. The feature extraction module uses a convolutional neural network to extract features from image data and performs weighted processing on key features through an attention mechanism.

4. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The model training and reasoning module adopts a transfer learning method to apply the pre-trained model to a new data set to improve the generalization ability of the model under small sample data. The dynamic model training method includes a hyperparameter adjustment mechanism based on Bayesian optimization to improve the recognition accuracy of the model with optimal configuration.

5. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The hyperparameter adjustment mechanism dynamically updates the hyperparameters according to the performance indicators of the model to cope with the diversity of different sample data. The interpretability analysis module uses a visualization method based on Grad-CAM to display the decision-making process of the model in the form of a heat map.

6. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The explanation report generated by the explainability analysis module includes the credibility score of the model decision and a detailed description of key features. The result output module has a customizable function, allowing medical staff to adjust the format and content of the result display according to needs.

7. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The database module supports the federated learning framework, allowing multi-center data collaborative training without sharing patient data, where the federated averaging algorithm is used to aggregate model parameters:

8. The blood sample automatic analysis system based on artificial intelligence according to claim 1, characterized in that: The security and privacy protection module uses homomorphic encryption technology to calculate and analyze encrypted data without decryption. The system integrates a self-diagnosis module to monitor the operating status of the system and automatically trigger a troubleshooting procedure when potential problems are found.

Images