Satvika Bharadwaj
BME PhD Defense Presentation

Date: 2026-11-24
Time: 10:00 - 11:30 am
Location / Meeting Link: 2nd Floor, Conference Room 202, Emory Children's Center (ECC), 2015 Uppergate Drive, Atlanta, GA 30322 (across from HSRB) and https://emory.zoom.us/j/98250266880?from=addon

Committee Members:
Anant Madabhushi, PhD (Advisor) ; Farzad Fereidouni, PhD ; Sunil Badve, MD, FRCPath ; Saurabh Sinha, PhD ; Jaydev Desai, PhD ;


Title: Interpretable biomarker discovery using deep phenotyping of histopathology slides to predict disease progression and treatment response

Abstract:
Computational pathology uses machine learning and image analysis to quantify histologic characteristics that may not be fully captured through conventional pathologic examination. These approaches are used to develop imaging biomarkers for disease characterization, prognostic assessment, and prediction of treatment benefit. This thesis investigates deep phenotyping as a framework for developing interpretable and quantitative histologic biomarkers across malignant and nonmalignant diseases. In Aim 1, an automated deep learning framework was developed for static bone histomorphometry in undecalcified bone biopsies. It quantified bone and osteoid areas, bone marrow adiposity, and osteoblast and osteoclast measures. The framework was evaluated against manual histomorphometry and used to characterize bone turnover. In Aim 2, the density and spatial architechture of tumor-infiltrating lymphocytes were investigated for their ability to identify patients with human epidermal growth factor receptor 2-positive (HER2+) breast cancer who derived differential benefit from trastuzumab-based therapy. A computational biomarker was developed and its predictive value was evaluated in the NSABP B-41 randomized clinical trial. In Aim 3, quantitative features of the tumor and its microenvironment were investigated for their ability to identify patients with hormone receptor-positive (HR+) breast cancer at risk of late recurrence (more than five years after diagnosis). A computational prognostic biomarker was developed by integrating nuclear morphology, tubular architecture, mitotic activity, collagen organization, and immune cell characteristics. The biomarker was evaluated across independent cohorts with extended follow-up, and landmark analyses were performed to assess recurrence risk beyond five and ten years. Through these applications, I developed and validated clinically meaningful biomarkers from quantitative histologic measurements to support individualized patient care