Gondim, Dibson Dibe (UofL)

Research Activities

The field of anatomic pathology centers on disease diagnosis through the expert examination of histopathology slides. These slides are meticulously prepared from biopsy or surgical tissue specimens. Through high-magnification microscopy, pathologists can analyze cellular architecture and characteristics, which often display distinctive patterns in various disease states. Recent technological advances have enabled the digitization of histopathology slides into whole slide images (WSIs). This digital transformation not only enhances collaborative possibilities and educational opportunities but also enables the application of computational approaches, including artificial intelligence, to slide evaluation. These technological capabilities offer promising avenues for improving both diagnostic precision and workflow efficiency. Our research group specializes in developing innovative computational solutions for histopathology analysis, leveraging cutting-edge technology to advance the field of digital pathology.

List all projets

Title: Evaluating Embedding Quality from Pathology Foundation Models for Automated Tumor Classification Principal Investigator: Dibson Dibe Gondim, MD Project Description: Foundation models in pathology represent an advanced application of deep learning, utilizing a teacher-student architecture trained on extensive pathology image datasets. These models generate high-dimensional vector embeddings that can be leveraged for various downstream analytical tasks. Despite their potential, the comparative effectiveness of different foundation models' embeddings for specific pathology applications remains understudied. Our research aims to systematically evaluate the quality and utility of embeddings derived from various pathology foundation models for automated tumor classification. This study will assess multiple foundation models across diverse tumor categories, analyzing their performance in classification tasks and identifying key factors that influence embedding quality and classification accuracy. This investigation will contribute to our understanding of how different architectural choices and training approaches in foundation models affect their utility in practical diagnostic applications. The findings will help guide the selection and optimization of foundation models for specific pathology use cases.

Computational methods

Image preprocessing and analysis will be implemented using Python's robust ecosystem of image processing libraries. PyTorch will be used for deep learning pipeline components.

List all Software

Storage: External S3 bucket Python Development Environment Jupyter Lab/Notebook for interactive development Command-line interface (CLI) for production pipeline Deep Learning Framework PyTorch for neural network implementations Image Processing Libraries scikit-image: Scientific image processing tools PIL (Python Imaging Library): Core image handling OpenCV: Computer vision and image preprocessingTemp page created by http://cilogon.org/serverE/users/118677.