Background
The integration of deep learning and spatial transcriptomics is transforming digital pathology, particularly in cancer diagnostics and prognosis. AI-powered pathology assistants, including multimodal large language models (LLMs), have demonstrated the ability to interpret histopathological images and generate clinical insights. However, current models primarily focus on morphological features and do not leverage spatially resolved gene expression data, limiting their diagnostic accuracy.
Aim
This project aims to develop an AI copilot that integrates spatial transcriptomics with histopathological image analysis for melanoma diagnosis. By incorporating molecular information into AI models, this system will provide pathologists with more accurate, explainable, and high-resolution insights into high-risk regions of melanoma tissues.
Approach
This project will involve preprocessing and aligning H&E histopathology images with spatial transcriptomics data, developing deep learning models using contrastive learning to integrate multimodal features, and fine-tuning large language models (LLMs) for clinical pathology queries. An interactive AI-powered web interface will be developed for real-time pathology assistance using inhouse melanoma datasets.
Project Potential
This project offers a unique opportunity to work at the intersection of AI, spatial omics, and digital pathology, contributing to the development of next-generation clinical diagnostic tools. Students will gain hands-on experience with deep learning, contrastive learning, and multimodal AI models. The outcomes of this research could lead to improved melanoma diagnostics, enhanced AI-assisted pathology workflows, and potential clinical translation of AI-powered decision support systems.