Adaptive AI Model for Medical Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI models for classifying medical images are not user-specific, leading to inaccurate classifications and requiring resource-intensive retraining, which can result in decreased accuracy for less difficult cases and increased complexity for users.
Innovation Solution
A method for adapting AI models to specific users by incorporating user-specific data elements during normal workflow, allowing for improved classification performance through automated generation and retraining of models, ensuring better alignment with user experiences and risk tolerance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI models are calibrated during setup at the user location, then the classification accuracy is improved for that user, but the AI becomes static and cannot adapt to changes in user preferences or medical guidelines over time
Solution Approach 1:
The patent implements dynamic model adaptation by enabling continuous retraining of AI models using newly acquired user-specific data. The system transitions from static calibration to dynamic updates, where models are periodically retrained as new data becomes available during normal workflow, allowing the AI to adapt to changing user preferences and medical guidelines while maintaining improved classification accuracy.
2Measurement precision
If extensive retraining is performed to improve classification accuracy, then the model becomes more accurate, but the complexity and resource consumption increase significantly
Solution Approach 1:
The patent applies partial retraining by using only the necessary subset of data and model parameters for updates. Instead of complete retraining, the system performs incremental learning using new user-specific data elements acquired during normal workflow, reducing computational complexity and resource consumption while still improving classification accuracy through targeted model updates.
Solution Approach 2:
The system creates and uses copies of the AI model for training purposes while the original model continues to serve production requests. Multiple model instances allow parallel processing where copy models are trained on new data without interfering with ongoing classifications, reducing the impact on system complexity and maintaining service continuity.
3Productivity
If user-specific data is collected during normal workflow, then the model can be continuously improved, but this requires additional user intervention and time
Solution Approach 1:
The patent implements self-service by automatically acquiring user-specific data elements during normal workflow without requiring explicit user action. The system passively collects data from images classified by the user, automatically generates training data elements, and initiates model retraining processes, eliminating the need for users to manually provide data or intervene in the improvement process while enabling continuous model enhancement.
Solution Approach 2:
The system establishes a feedback loop where classification results from normal workflow automatically feed into the training process. User classifications provide implicit feedback that is captured, stored as training data elements, and used to continuously improve the model, creating a self-reinforcing cycle of improvement that requires no additional user time beyond the normal classification task.
4Measurement precision
If the AI model is made user-specific through calibration, then the classification performance improves for that user, but the process becomes more complex for the user to manage
Solution Approach 1:
The patent makes the AI model self-managing by automatically handling all aspects of user-specific adaptation. The system autonomously acquires user-specific data during normal workflow, processes this data into training elements, performs model retraining, and manages version updates without user intervention. This eliminates the complexity of manual model management while maintaining improved classification performance through continuous user-specific adaptation.
Data Source
AI summary
The invention relates to the field of classifying, using an artificial intelligence, medical images showing a body portion. The invention provides a method for adapting to a specific user a model of an artificial intelligence for classifying images of a body portion, wherein the method is integrated in the user's everyday workflow in a manner that the execution of the method has no or nearly no impact on the user's everyday work. Therefore, user-specific data elements 24 are generated in an automated manner (step S2) during the user's work. A user-specific data element 24 comprises a medical image of the body portion to be classified and a classification 26 (also called label) approved or corrected by the user, wherein the image is taken by the user or a medical imaging system of the user during normal, everyday work. The model is adapted to the user by generating a training data set 27 comprising user-specific data elements 24, by training the artificial intelligence on the training data set 27 for generating an adapted model 2 (step S3), and by replacing a current model 1 of the artificial intelligence used by the user with the adapted model 2 if a replacement criterion is fulfilled (step S5).The invention provides further a method for improving the performance of a system 100 for classifying images of a body portion and a system 100 related to the mentioned methods.