The computational advantages of deep learning in AI, integrated with digital pathology for microscopy imaging, has led to the emergence of a new field called Computational Pathology (CoPath) that is poised to transform clinical pathology globally.
The field of CoPath is dedicated to the creation of automated tools that address and aid steps in the clinical workflow for cancer diagnostics. With increasing advancements in deep learning, image analytics, and enabling hardware, the research focus in this field has expanded and branched into a broad range of domains.
In this seminar we present our theoretical advancements in deep learning and computer vision algorithms with focused application in CoPath. We investigate this from both data-centric and model-centric approaches to cohesively relate between “data” and “learning-models” so to effectively design, train, and rationalize our algorithmic decisions.
From data-centric viewpoint we discuss our novel approach in designing tissue taxonomies for comprehensive representation of histology landmarks that are identified from primary organ sites. This builds a new foundation of representation learning problem in CoPath where tissues are labelled in multilabel classification problem; dubbed Atlas of Digital Pathology (ADP) database. We discuss several design approaches of multilabel representation learning from ADP and their diagnostics applications in classification, segmentation, and biomarker discovery patterns.
From model-centric viewpoint we discuss our recent theoretical advancements in deep learning by introducing several generalization measures from deep training regimes and augment them in optimization algorithms for interpretable training and searching of deep network architectures for data representation.
We conclude this talk with the objectives of future research plans and how to derive novel AI solutions that can facilitate the transformational changes in clinical pathology for cancer diagnostics.
