Relevance Feedback for Patient Cohort Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for identifying patient cohorts in medical tasks, such as clinical trials and diagnosis, face challenges due to the complexity of large feature spaces and the inability of unsupervised feature selection techniques to account for clinical relevance, leading to inefficient cohort selection.
Innovation Solution
A system that combines unsupervised feature selection with clinician-provided relevance feedback, allowing for iterative adjustment of feature sets and weights to improve cohort identification, leveraging the clinician's holistic understanding of patient similarities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If unsupervised feature selection techniques are used to automatically select features for patient clustering, then the automation extent increases, but the clinical relevance and accuracy of cohort identification deteriorates
Solution Approach 1:
The system implements feedback by displaying clustered patients to clinicians for review, receiving their feedback on whether patients are appropriately clustered, and using this feedback to iteratively adjust the clustering algorithm and feature weights, thereby improving clinical relevance while maintaining automation
Solution Approach 2:
The system dynamically adjusts feature weights and clustering parameters based on clinician feedback throughout the iterative process, allowing the system to adapt and improve its performance over time rather than using static automated selection
2Quantity of substance
If all available patient features are used for clustering, then the completeness of patient profiling increases, but the device complexity and computational burden increases
Solution Approach 1:
The system extracts and selects only the most relevant features from the large set of available patient features through automated feature selection and iterative refinement, removing unnecessary features to reduce complexity while maintaining clustering quality
Solution Approach 2:
The system segments the large feature space into relevant and irrelevant features through iterative clustering and clinician feedback, processing only the relevant subset for final cohort identification
3Measurement precision
If manual feature selection by clinicians is performed, then the clinical relevance improves, but the productivity and efficiency of cohort identification deteriorates
Solution Approach 1:
The system performs partial manual action by involving clinicians only in the feedback review stage rather than requiring them to manually select all features, combining automated feature selection with selective clinical oversight to balance efficiency and relevance
Solution Approach 2:
The system performs self-service through automated feature selection and initial clustering, reducing the burden on clinicians to manually process all features while still incorporating their expertise through feedback
Data Source
AI summary
In patient cohort identification, clustering (30) of patients is performed using a patient comparison metric dependent on a set of features (24). Information is displayed on sample patients who are similar or dissimilar to a query patient according to the clustering. User inputted comparison values are received comparing the sample patients with the query patient. The set of features and/or feature weights are adjusted to generate an adjusted patient comparison metric having improved agreement with the user inputted comparison values. The clustering is repeated using the adjusted patient comparison metric. A patient cohort is identified from a cluster (34) containing the query patient produced by the last clustering repetition. The information on the sample patients may be shown by simultaneously displaying two or more graphical modality representations (70, 72, 74) each plotting the sample patients and the query patient against two or more features of the modality.