Weighted-Ensemble Machine Learning for Ophthalmic Disease Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated systems for detecting ophthalmic diseases from images face challenges due to overfitting when selecting the best-performing machine learning algorithm and assigning weights to ensemble models, which can lead to poor generalization and inaccurate diagnoses.
Innovation Solution
A system that uses an ensemble of hierarchical end-to-end models, where each model is trained and tested, and weights are assigned based on performance, with a weighted average of class scores for classification, to improve generalization and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the best-performing machine learning algorithm is selected based on available test data, then classification accuracy is improved on the test set, but overfitting occurs and generalization performance deteriorates
Solution Approach 1:
The patent combines multiple machine learning algorithms into an ensemble system where predictions from individual algorithms are aggregated. This merging approach prevents overfitting to any single algorithm while maintaining high classification accuracy through diverse model perspectives, directly resolving the contradiction between test set performance and generalization capability
Solution Approach 2:
The patent dynamically adjusts the weights of individual algorithms in the ensemble based on their performance characteristics and the specific characteristics of the input image. This parameter adjustment allows the system to adapt to different scenarios, improving generalization while maintaining high accuracy across various test conditions
2Ease of manufacture
If non-weighted average ensemble approach is used, then implementation simplicity is improved, but generalization performance deteriorates due to blind averaging
Solution Approach 1:
The patent introduces weighted averaging instead of simple uniform averaging, where each algorithm's contribution is modulated by a learned weight parameter. This parameter change maintains relative implementation simplicity while dramatically improving generalization performance by allowing better algorithms to have greater influence on the final prediction
3Device complexity
If equal weights are assigned to all models in ensemble, then computational complexity is reduced, but diagnostic accuracy deteriorates due to giving too much influence to poor-performing models
Solution Approach 1:
The patent changes from equal weighting to differential weighting where each model receives a weight proportional to its performance. This parameter change maintains computational efficiency while improving diagnostic accuracy by reducing the influence of poorly performing models and amplifying the contribution of high-performing models in the ensemble
Data Source
AI summary
Disclosed herein are systems, methods, and devices for classifying ophthalmic images according to disease type, state, and stage. The disclosed invention details systems, methods, and devices to perform the aforementioned classification based on weighted-linkage of an ensemble of machine learning models. In some parts, each model is trained on a training data set and tested on a test dataset. In other parts, the models are ranked based on classification performance, and model weights are assigned based on model rank. To classify an ophthalmic image, that image is presented to each model of the ensemble for classification, yielding a probabilistic classification score—of each model. Using the model weights, a weighted-average of the individual model-generated probabilistic scores is computed and used for the classification.