Candidate Correlation Classification in Medical Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional classification methods in medical imaging assume independent and identically distributed data, which is often violated in real-world applications, leading to inaccurate classification of candidates in medical images due to ignored correlations among features and labels.
Innovation Solution
A method that determines internal correlations and differences among candidates within a medical image, using both probabilistic and mathematical programming approaches to classify candidates by accounting for their locations and descriptive features, thereby enhancing classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional classification methods assume independent and identically distributed data, then the classification process is simple and fast, but the classification accuracy deteriorates due to ignored correlations among candidates
Solution Approach 1:
The patent combines multiple candidates from the same patient into a single composite candidate by merging their features and labels. This merging process captures the correlations among candidates while simplifying the overall classification task. The composite candidate integrates information from multiple individual candidates, allowing the classifier to benefit from correlated data without requiring complex multi-candidate classification algorithms.
Solution Approach 2:
The patent segments the classification problem by first grouping candidates into patient-specific subsets, then creating composite candidates for each subset. This segmentation approach breaks down the complex problem of classifying multiple correlated candidates into simpler sub-problems of creating and classifying composite candidates, thereby improving accuracy without proportionally increasing complexity.
2Reliability
If conventional methods classify candidates independently one at a time, then the computational process is efficient, but the sensitivity and specificity of disease detection deteriorate
Solution Approach 1:
By merging multiple candidates into a single composite candidate that represents all candidates from a patient, the method improves disease detection reliability. The composite candidate aggregates evidence from multiple observations, reducing false positives and false negatives. This merging occurs before classification, maintaining computational efficiency while improving reliability through better utilization of correlated data.
3Measurement precision
If the standard assumption of independent data is violated to account for correlations, then classification accuracy improves, but the complexity of the classification system increases
Solution Approach 1:
Instead of modifying the classification algorithm to handle correlated data directly (which would increase complexity), the patent inverts the approach by transforming the data representation. It creates composite candidates that inherently encode the correlations, allowing standard independent classification algorithms to achieve better accuracy. This inversion shifts the complexity from the classification algorithm to the data preprocessing stage.
Data Source
AI summary
A method and system correlate candidate information and provide batch classification of a number of related candidates. The batch of candidates may be identified from a single data set. There may be internal correlations and/or differences among the candidates. The candidates may be classified taking into consideration the internal correlations and/or differences. The locations and descriptive features of a batch of candidates may be determined. In turn, the locations and/or descriptive features determined may used to enhance the accuracy of the classification of some or all of the candidates within the batch. In one embodiment, the single data set analyzed is associated with an internal image of patient and the distance between candidates is accounted for. Two different algorithms may each simultaneously classify all of the samples within a batch, one being based upon probabilistic analysis and the other upon a mathematical programming approach. Alternate algorithms may be used.