Agentic Vision-Language Models for Biomedical Anomaly Detection and Differential Biomedical Image Analysis

Overview

We aim at applying advanced computer vision techniques to tackle three challenges in biomedical visual analysis.
Automatically discover differences between two sets of medical images, Automatic lab-scene reasoning to identify anomalies and risks, and Reveal protein locations inside a living cell at nanometer resolution and provide biological insights from it.

We research on three topics in total:

• Comparative Decoding: Describing Differences in Image Sets
• Bio-lab Video Anomaly Reasoning: video segmentation, object tracking, VLM CoT reasoning, agentic framework
• Agentic VLMs for Biomedical Images: perception audit, quantitation via visual self-feedback & agentic spatial reasoning

---