I am a research scholar specializing in medical image analysis and deep learning, with a particular focus on dermatological image processing. My research interests include skin lesion segmentation, controllable image synthesis, and uncertainty quantification in medical imaging. I work at the intersection of computer vision and healthcare, developing novel methods to improve diagnostic accuracy and clinical workflows.
We propose PSAM, a novel framework that enhances SAM's segmentation capability for dermatological images through uncertainty-guided point selection and pseudo-box constraints. Our approach addresses domain shift challenges in medical imaging by leveraging the model's own predictive uncertainty to strategically place complementary prompts.
Key Achievements: Achieves best Dice score (0.901) on HAM10000 validation set, outperforming nnU-Net by +1.2%, with a 20-fold speed improvement over Medical SAM (0.127-0.134 seconds per image).
We present LF-VAR, a novel approach for controllable skin image synthesis that addresses the limitations of existing methods in generating high-quality images with precise control over lesion location and type. Our method leverages a lesion-focused vector autoregression model to generate realistic skin images for training deep learning models.
Publication: Accepted at MICCAI 2025