AI/ML-based image analysis for cancer pathology

“Clinically relevant questions addressed using latest in data science”

We are working on the development of Artificial Intelligence Tools based image analysis for identification of aggressive and non-responsive Triple-negative breast cancer patients in India

Clinically relevant research question

Current cutting-edge breast cancer research undertaken to tackle this clinical problem is based on studies done on western populations. With the advent of cutting-edge technologies, aggressive diseases can be identified with biomarker expressions and can be treated aggressively. There is a need to have directed efforts to find associated cellular and molecular markers to identify aggressive subset of TNBC disease in India for better clinical management.

Introduction

Breast cancer burden in India is on the rise, with over 30-35% of breast cancers presented with late stage and aggressive disease. Like western countries, where biomarker-based stratification of subtypes and specific treatment regimens are already in place, in India, there is a need to establish India-specific biomarkers and molecular signatures to stratify breast cancer subtypes. So far, breast cancer has been classified into three molecular subtypes; ER/PR positive, Triple negative (TNBC) and HER2-enriched, on based on the presence of distinct hormone receptor biomarkers (i.e., estrogen receptor; ER, progesterone receptor; PR and Human epidermal growth factor 2; HER2 receptor). Furthermore, studies in western populations have revealed heterogeneity in TNBC in terms of its histological patterns, genetic mutation landscape and molecular expression patterns. Based on molecular expression patterns, TNBC has been proposed to be further classified. Molecular sub-classification for Indian breast cancers and TNBCs has not been attempted yet. Therefore, we propose to identify and classify the molecular subtypes and sub-sub types of Indian breast cancer patients, their response to available therapeutic strategies and the probability of recurrence. This molecular classification of breast cancer will be performed by building robust predictive model using state-of-the-art Artificial Intelligence (AI) and in silico simulation techniques, which will take into consideration large volumes of morphological, radiological, and transcriptomics data. In addition to using well-established molecular markers for each subtype, we plan to use machine-learning techniques on the Indian patient data to identify novel, India-specific molecular markers. 

Immediate goals: 

1.    Develop AI tools to identify and predict novel subtypes of breast cancer 
a.    based on morphological patterns (H&E based) and novel IHC marker expression.   
b.    based on radiological image patterns
c.    based on transcriptomic signatures
2.    Develop AI tools to predict treatment response of breast cancer subtypes
a.    based on novel IHC marker expression 
b.    based on transcriptomic signatures
3.    Develop AI tools to predict breast cancer survival prognosis
a.    based on novel IHC marker expression 
b.    based on transcriptomic signatures

Long term goals:

 
AI tools developed and validated through the proposed work, can serve for better clinical management of breast cancer patients, in terms of 
a.        Early diagnosis of aggressive breast cancers, and aggressive treatment for better prognosis. 
b.        Predicting non-responsive tumors early-on, for alternate and effective treatment option.
c.          Predicting recurrence interval which will aid, timely follow-up and early diagnosis of   recurrent disease.