Cohort-Aware Agents for Individualized Lung Cancer Risk Prediction Using a Retrieval-Augmented Model Selection...

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging and Diagnostics
• Computational Biology and Bioinformatics

Background:

• Lung cancer risk prediction models exhibit significant variability across diverse patient populations and clinical settings.
• No single predictive model demonstrates optimal performance for all patient cohorts.
• Accurate, individualized lung cancer risk assessment is crucial for effective screening and early detection.

Purpose of the Study:

• To develop a personalized lung cancer risk prediction agent that dynamically selects the most appropriate model for individual patients.
• To address the challenge of model generalizability by incorporating cohort-specific knowledge.
• To enhance the accuracy and applicability of lung cancer risk prediction in real-world clinical settings.

Main Methods:

• A two-stage agent pipeline combining retrieval and reasoning techniques.
• Cohort retrieval using FAISS-based similarity search across nine diverse real-world cohorts.
• Large Language Model (LLM) prompted with retrieved cohort data and performance metrics to recommend optimal prediction algorithms from a pool of eight models.

Main Results:

• The proposed agent enables dynamic, cohort-aware lung cancer risk prediction tailored to individual patient profiles.
• The pipeline successfully integrates patient CT scans and structured metadata for personalized risk assessment.
• Facilitates flexible and cohort-driven model selection across diverse clinical populations, improving risk prediction applicability.

Conclusions:

• The developed agent offers a practical approach to individualized lung cancer risk assessment by dynamically selecting appropriate prediction models.
• This personalized strategy addresses the limitations of single-model approaches in heterogeneous patient populations.
• The retrieval and reasoning framework paves the way for more precise and adaptable lung cancer screening tools.