Jointly Trained Image Normalization for Low-Label Disease Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models require large amounts of labeled training data to ensure accuracy, which is time-consuming to obtain and often fails to capture real-world variability, leading to low prediction accuracy due to insufficient training data.
Innovation Solution
Jointly train a normalization model with a prediction model to normalize input data, removing irrelevant variations and highlighting relevant features, using a loss function that includes idempotence and entropy terms to minimize training data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large amounts of labeled training data are used to train machine learning models, then prediction accuracy is improved, but time consumption and cost increase
Solution Approach 1:
The normalization model is trained in advance using unlabeled data to learn the distribution characteristics of the data. This preliminary normalization process transforms the input data before it reaches the prediction model, allowing the prediction model to be trained with fewer labeled samples while still achieving high accuracy by receiving pre-processed, distribution-matched input data.
2Measurement precision
If large amounts of labeled training data are used to train machine learning models, then prediction accuracy is improved, but the complexity of data collection and labeling increases
Solution Approach 1:
The normalization model serves as an intermediary between the raw input data and the prediction model. It processes the input data to match the training distribution, acting as a bridge that allows the prediction model to operate effectively with fewer labeled examples. This intermediary layer handles the complexity of distribution matching separately from the prediction task.
3Quantity of substance
If traditional training methods are used without data normalization, then training data requirements are high, but computational resources are consumed
Solution Approach 1:
The training process is segmented into two distinct phases: (1) training the normalization model on unlabeled data to learn distribution characteristics, and (2) training the prediction model on normalized data with fewer labeled samples. This segmentation allows each model to specialize in its respective task, reducing the overall computational burden and labeled data requirements while maintaining prediction accuracy.
Data Source
AI summary
Systems and methods for using a prediction model jointly with a normalization model to provide prediction results are provided. One example method includes receiving an input image of a tissue sample of a patient and generating a normalized image by applying a normalization model on the input image. The normalization model is configured to generate normalized data using input data for a prediction model, and the prediction model is configured to generate prediction results using normalized data generated by the normalization model. The normalization model and the prediction model are jointly trained. The method further includes generating a prediction of disease severity for the patient by applying the prediction model on the normalized image.