Ensemble Embedding for Multimodal Biomedical Data Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dimensionality reduction techniques for biomedical data analysis face challenges such as sensitivity to noise and parameter choice, which affect the accuracy of classification and clustering, especially in multi-modal data fusion for disease diagnosis.
Innovation Solution
The ensemble embedding method generates and combines multiple uncorrelated embeddings using linear or non-linear dimensionality reduction techniques, selecting strong embeddings based on embedding strength and propagating pair-wise relationships to construct a stable lower-dimensional representation that preserves class-discriminatory information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single dimensionality reduction projection is applied to high-dimensional biomedical data, then the data can be reduced to lower dimensional space, but the result is sensitive to noise and parameter choice, reducing classification accuracy
Solution Approach 1:
The patent segments the single dimensionality reduction process into multiple independent projection steps, each generating a separate embedding. By dividing the data into multiple projections with different parameters and combining them, the system reduces sensitivity to individual parameter choices and noise in any single projection.
Solution Approach 2:
The patent merges multiple independent embeddings into a single ensemble embedding that combines the strengths of individual projections. This combination approach integrates information from multiple dimensionality reduction perspectives, improving overall classification accuracy while reducing sensitivity to noise and parameter selection in any single embedding.
2Reliability
If multiple independent embeddings are generated and combined, then the robustness of classification improves and sensitivity to noise is reduced, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by generating a limited number of independent embeddings (not exhaustively all possible projections) and selecting those that contribute most to classification performance. This approach achieves robustness through multiple embeddings while controlling computational complexity by avoiding unnecessary projections.
3Loss of information
If non-linear dimensionality reduction techniques are used to preserve local adjacencies, then class-discriminatory information is better retained, but the methods are more sensitive to parameter choice and computational cost increases
Solution Approach 1:
The patent segments the non-linear dimensionality reduction process into multiple independent projections, each using different parameter settings. By applying non-linear techniques to multiple subsets of data with varying parameters and then combining the results, the system preserves class-discriminatory information while reducing sensitivity to any single parameter choice.
Solution Approach 2:
The patent merges multiple non-linear embeddings into an ensemble that combines their respective strengths in preserving local adjacencies and class discrimination. This combination reduces the computational burden and parameter sensitivity associated with any single non-linear projection while maintaining the benefits of non-linear dimensionality reduction.
Data Source
AI summary
The present invention provides a system and method for analysis of multimodal imaging and non-imaging biomedical data, using a multi-parametric data representation and integration framework. The present invention makes use of (1) dimensionality reduction to account for differing dimensionalities and scale in multimodal biomedical data, and (2) a supervised ensemble of embeddings to accurately capture maximum available class information from the data.