284th LG: Machine Learning and Computational Pathology for Breast Cancer Diagnostics

Friday, September 2, 2016 - 12:00
Dr. Humayun Irshad
Since the worldwide acceptance of mammographic screening, breast biopsies to evaluate suspicious lesions are dramatically more common. The proper classification of these lesions could provide critically important diagnostic information to prevent under- and over-treatment and to effectively guide clinical management. Yet manual diagnosis has a major reproducibility problem: there is high inter-observer variability among pathologists. Breast Carcinogenesis is a heterogeneous disease with distinct disease grade and subtypes, driven by distinct sets of molecular abnormalities and morphological and architectural phenotypes. The central research question that I will address is: can the development of novel methods in computational pathology improve diagnosis for pre-invasive lesions and invasive breast cancer? This presentation will cover following five topics. 1) Crowdsourcing image classification and Annotation 2) Deep learning framework for Breast Metastasis Detection 3) Computational Diagnostic Framework for Early Neoplasia 4) Computational Expansion Pathology 5) 3D Computational Framework for Breast Neoplasia These novel computational methods for analyzing breast tissue will enable the identification of new morphological and architectural phenotypes associated with breast cancer etiology, progression, and metastasis, leading to improvements in the ability of pathologists to provide accurate and predictive diagnoses of normal breast tissue, pre-invasive breast lesions, and invasive breast cancer.