AI Disease Subclassification With Two-Stage ICD-10 Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical regulatory and legal frameworks require extensive clinical trials and mapping to ICD-10 codes for autonomous AI systems, making it difficult to use clinically relevant disease categories without going through rigorous regulatory hurdles.
Innovation Solution
A two-stage AI system is developed, comprising a diagnostic model for clinical diagnosis and a sub-classification model to map outputs to hierarchical coding systems like ICD-10, allowing for clinically relevant disease subclassification without requiring new studies, using machine learning to generate sub-classifications within existing frameworks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a fully-autonomous AI diagnosis system is used that diagnoses in a dichotomous manner (severity less than vs severity at least), then clinically relevant disease categories can be diagnosed, but the output does not map to ICD-10 codes requiring immense regulatory hurdles and huge expense
Solution Approach 1:
The AI system is segmented into two distinct models: a diagnostic model that provides clinically relevant dichotomous diagnoses, and a sub-classification model that maps outputs to ICD-10 codes. This segmentation allows each model to specialize in its specific function, resolving the contradiction between clinical relevance and regulatory compliance.
Solution Approach 2:
The sub-classification model acts as an intermediary between the diagnostic model's clinically relevant output and the ICD-10 coding system. It translates the dichotomous clinical diagnoses into hierarchical ICD-10 codes without requiring the diagnostic model itself to be retrained or validated against ICD-10, thus reducing regulatory hurdles.
2Reliability
If mapping to ICD-10 is required for regulatory compliance, then hierarchical coding framework is satisfied, but clinically less relevant or even not relevant diagnoses are output
Solution Approach 1:
The system separates the clinical diagnosis function from the coding function. The diagnostic model maintains clinical relevance by outputting dichotomous diagnoses, while the sub-classification model handles ICD-10 mapping for regulatory compliance. This ensures both reliability for regulation and ease of operation for clinical use.
Solution Approach 2:
The solution adds another dimension to the output by creating a hierarchical structure where the primary diagnostic output remains clinically relevant, and a secondary ICD-10 classification layer is added for regulatory purposes. This dimensional addition satisfies both clinical and regulatory requirements simultaneously.
3Reliability
If a new AI system is built to diagnose various ICD-10 categories for DRD and DME, then ICD-10 mapping is achieved, but it would not be clinically very relevant and would require much larger N for confirmatory studies
Solution Approach 1:
The sub-classification model is trained preliminarily on existing data to learn the mapping between clinical diagnoses and ICD-10 codes. This preliminary training allows the system to achieve ICD-10 mapping accuracy without requiring extensive new clinical trials, as the mapping relationships can be learned from existing labeled data.
Solution Approach 2:
Instead of building a completely new AI system for ICD-10 diagnosis, the solution copies the output structure of the diagnostic model and applies a transformation layer (sub-classification model) that maps to ICD-10 codes. This copying approach maintains clinical relevance while achieving regulatory mapping without requiring much larger N for confirmatory studies.
Data Source
AI summary
A fully autonomous system is used to subclassify a disease in a patient. For example, a tool receives one or more images of a body part of a patient, inputs the one or more images into a diagnostic model, and receives, as output from the diagnostic model, a diagnosis for the patient. The tool determines whether the diagnosis is positive for a given disease, and, responsive to determining that the diagnosis is positive, inputs a representation of the one or more images into a diagnosis subclassification model. The tool determines, based on output from the diagnosis subclassification model, a subclassification for the diagnosis, and outputs a control signal based on the subclassification.