Back to Projects
in progress
Oyster MSX Detection Pipeline
A Human-in-the-Loop AI system for accelerating the diagnosis of MSX disease in oysters using gigapixel slide images.
About the Project
A comprehensive diagnostic pipeline designed to assist pathologists in identifying MSX disease. Unlike 'black box' AI models, this system functions as an interactive digital co-pilot.
It processes gigapixel Whole-Slide Images (WSI) using a hybrid approach: ResNet20 and YOLO11 for initial detection, followed by the Segment Anything Model (SAM) for precise tissue segmentation. The system prioritizes interpretability, allowing pathologists to validate AI suggestions rather than replacing their judgment.
Engineering Challenges
01Processing Gigapixel Images
Whole-Slide Images are too large to process in memory at once (1TB+ dataset).
Solution: Implemented a tiling strategy and parallel processing pipeline on HPC infrastructure using CUDA.
02Trust & Interpretability
Pathologists needed to trust the AI's decisions.
Solution: Designed an XAI interface that flags potential issues for review, keeping the human expert in the loop.