Image Recognition Model Training Using Self-Labeling for Medical Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity of medical images and limited labeling resources hinder the effective training of image recognition models for automatic diagnosis, resulting in low accuracy and inefficient use of data.
Innovation Solution
A method and apparatus for training an image recognition model that utilizes both labeled and unlabeled medical images across different tasks, effectively reducing the need for extensive labeling and increasing the training data volume, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large quantity of medical images are used to train the image recognition model, then the model prediction accuracy is improved, but the labeling resource requirement increases
Solution Approach 1:
The system performs self-labeling by using the pre-trained image recognition model to automatically generate labels for unlabeled medical images. The model predicts disease types and lesion positions for unlabeled images, and these predictions are used as training labels, eliminating the need for manual medical staff labeling while still expanding the training dataset.
Solution Approach 2:
The system changes the state of unlabeled images by transforming them into labeled images through automated prediction. By applying the image recognition model to generate predicted labels, the parameter state of the images changes from unlabeled to labeled, thereby expanding the effective training data without requiring additional manual labeling resources.
2Adaptability or versatility
If labeled medical images are used for specific task training, then the training set effectiveness is improved, but the data volume for different tasks becomes insufficient
Solution Approach 1:
The system creates a universal training approach where unlabeled medical images can serve multiple tasks simultaneously. By using the image recognition model to generate predictions for disease type classification, lesion position detection, and other tasks on the same unlabeled images, the system enables one dataset to fulfill multiple training purposes, thereby increasing data volume availability across different tasks.
Solution Approach 2:
The system adds a new dimension to the training data by generating multiple types of labels (disease type, lesion position, etc.) from the same unlabeled images through different prediction tasks. This dimensional expansion allows the same image resource to contribute to multiple training objectives simultaneously, solving the data volume insufficiency problem for different tasks.
3Measurement precision
If medical images are manually labeled by medical staff, then the labeling accuracy is improved, but the labeling efficiency decreases
Solution Approach 1:
The system replaces manual medical staff labeling with automated self-labeling using the image recognition model. The model automatically predicts disease types, lesion positions, and other annotations for medical images, eliminating the need for medical staff to manually label images while maintaining consistent labeling standards across the entire dataset.
Solution Approach 2:
The system substitutes the mechanical process of manual labeling by medical staff with an automated computational system. The image recognition model performs the labeling function through algorithmic processing, replacing the human mechanical action of annotation while improving both efficiency and consistency of the labeling process.
Data Source
AI summary
A method for training an image recognition model includes: obtaining training image sets; obtaining a first predicted probability, a second predicted probability, a third predicted probability, and a fourth predicted probability based on the training image sets by using an initial image recognition model; determining a target loss function according to the first predicted probability, the second predicted probability, the third predicted probability, and the fourth predicted probability; and training the initial image recognition model based on the target loss function, to obtain an image recognition model.