Medical Imaging AI

• Multimodal AI Fusion Framework — First-author framework integrating CT imaging and clinical data (XGBoost with weighted loss) to predict 10-year MACE; ROC ≈ 0.88 on a 25,514-patient multimodal dataset.
• Deep Learning Pipeline Beyond CAC — CT-based pipeline predicting MACE beyond calcium scoring with 78.8% accuracy, improving over traditional CAC by 25.7%.
• Probability-Level Fusion — Chest X-rays + clinical data predicting rare outcomes with 92-99% accuracy and PR AUC 61-97% on imbalanced cohorts.
• Active Learning for Label Efficiency — Reduced labeling by >75%, needing only 15.4% (binary) and 23.1% (multi-class) of data to match full-model performance.
• Human-in-the-Loop Eye-Tracking — Co-authored model using gaze from 8 radiologists, improving interpretability and accuracy by up to 13%; collaborations with Emory clinicians (De Cecco, van Assen, Ardeshir-Larijani, Quyyumi, Krupinski).

Publications & Preprints

• Deep Active Learning for Lung Disease Severity Classification from Chest X-rays — Submitted to Journal of Imaging Informatics in Medicine.
• Predicting 10-year MACE Using Multi-Source Modalities with XGBoost — Submitted to Journal of Cardiovascular Computed Tomography.
• Imbalance-Aware Multimodal Fusion on Probability Distributions — Submitted to Radiology: Artificial Intelligence (RSNA).
• Co-author: Explainable ML for Risk Stratification of MACE — AHA Scientific Sessions 2025 (ID: 4366633).
• Co-author: Observer Performance and Eye-Tracking Variations as a Function of AI Output Format — SPIE Medical Imaging (DOI: 10.1117/12.3048588).

View my Google Scholar profile

Integrated AI-Based Discovery & Innovation

• Ensemble Portfolio Optimization — Pioneered LSTM + Deep Q-Learning ensemble, outperforming equal-weighted portfolios by 200%, SPX by 100%, and QQQ by 50%.
• Rebalancing via Reinforcement Learning — Replicated and advanced institutional portfolio rebalancing papers, achieving fast convergence on 512 assets using Linux, PyTorch, and GPU acceleration.
• 100-Dimensional Ellipsoid Sampling — Conducted research on 100-dim sampling with Ellipsoid method under 20 minutes and analyzed stock universe volatility across look-back periods.
• Implied Volatility Curve Modeling — Created robust curves for derivatives, driving profitable trading strategies.
• Generative AI for Trading Decisions — Developed and optimized LLM-based models for trading reasoning.