Tandra Sarkar - Digital Health Symposium and Roundtable

Introduction: Screening mammogram is main tool for early diagnosis of breast cancer.Challange remain in big numbers of mammograms for reporting (40 million mammograms performed every year in US alone), Breast cancer incidence and mortality data in 2020 were obtained from GLOBOCAN online database.There were approx 2.3 million new breast cancer cases in 2020,accounting approx 24.5% of all cancers worldwide.

Discussions: Aim of AI is to detect breast cancer early to reduce global burden and improvise this specialised segment of health care industry. Though most AI applications were initially confined to breast cancer diagnosis,now the horizon expanded with potentialities beside cancer detection to decision support,breast density,cancer Risk assessment,work flow applications ,quality assessment,neo adjuvant Chemotherapy response and Image enhancement. AI assessments extensively used in mammography because of BIRADS Lexicon ‘s structured image acquisition and reporting system. However AI applications showing potentiality in ultrasound,both hand held and automated breast ultrasound (AUBS). Besides the FDA approved AI applications,Tele radiology venders are coming up with there own AI solutions with semi supervised learning methods (SSL). This means less data required for mechine learning. Quality assessment needs special mention as this is of utmost importance in good reporting. In USA,breast imaging quality maintained by FDA through EQUIP(enhancing quality using inspiration programme).However breast cancer surveillance consortium found a sensitivity of 86% and specificity of 88.9% in screening mammogram.With AI implementation,there is significant understanding of root cause of poor performance. Regarding risk factor assessment.In India,definite breast screening programs could not be implemented yet because of population and non reachable mammograms fascilities.However AI system can regulate low,moderate and high risk populations and unnecessary biopsies could be avoided. In DCIS, indolent versus significant cancer could be understood with probability percentile and over diagnosis reduced.

Conclusion: Over all future of deep learning using narrow learning technology of AI integrated in breast imaging promises cost effective,high volume workflow which if not independent but will take seat of important co reader and value added adjunct tool.