Towards Generalized and Robust Histology Image Segmentation – Azadeh Fakhrzadeh (Extra Seminar, Online)

Developing robust segmentation methods that can adapt to new datasets remains a critical challenge in histology image analysis. Histology images exhibit substantial variability, even among samples from the same organ, due to differences in staining protocols, color intensity, laboratory procedures, and tissue quality. These variations alter morphological and textural characteristics, often causing models trained on one dataset to perform poorly when applied to another. Convolutional neural networks (CNNs) have demonstrated strong performance in histology image segmentation but suffer from limited generalizability due to their task-specific nature, often requiring retraining for new datasets. Recently, there has been increasing interest in developing models that leverage large-scale pretraining on diverse datasets to acquire generalized representations, thereby enabling effective adaptation to specific downstream tasks. This paradigm has emerged as a promising new research direction in the field of image segmentation. In this talk, I will present a comparative evaluation of two state-of-the-art pretrained large-scale models, one general-purpose and one histopathology-specifi, against a CNN baseline for histology instance segmentation. Our analysis focuses on two key tasks: (1) germinal epithelium segmentation in testicular histology images, and (2) gland segmentation in colorectal tissue. Through systematic benchmarking, I will highlight insights for developing robust and generalizable histology segmentation methods.

About the speaker: Azadeh Fakhrzadeh is an Assistant Professor in the Information Technology Research Department at the Iranian Research Institute for Information Science and Technology in Tehran and the Director of the Artificial Intelligence and Data Analysis (AIDA) Lab. She holds a Ph.D. in Computerized Image Processing from Uppsala University and an M.Sc. in Electrical and Computer Engineering from Toronto Metropolitan University. Her research focuses on AI-driven medical image analysis, histology image processing, OCR systems, and vision transformer and language models.

More Info About Azadeh Fakhrzadeh