Uncertainty Estimation in Cardio Landmark Detection and Heart Disease Diagnosis on Chest X-Ray Images

Dr.S.Suryakumari 1,Gandla Padma Sree 2,Bodigandla Chinna Peeraiah 3,Sarukuru Siva Kumar 4, Chejarla Chetan Sai 5
1Assistant Professor, Dept of Information Technology, SV College of Engineering, Tirupati, India.
2B. Tech, Dept of Information Technology, SV college of Engineering, Tirupati, India.
3B. Tech, Dept of Information Technology, SV college of Engineering, Tirupati, India.
4B. Tech, Dept of Information Technology, SV college of Engineering, Tirupati, India.
5B. Tech, Dept of Information Technology, SV college of Engineering, Tirupati, India.
Email: 1suryakumari.s@svce.edu.in, 2gandlapadmasri0811@gmail.com,
3bodigandlachinnapeeraiah@gmail.com,4sivasarukuru@gmail.com, 5saibrunx0@gmail.com

Abstract-Landmark detection and heart disease diagnosis from chest X-ray images is a complex andtime-consuming process traditionally performedby radiologists,which involves variabilityand subjectivity in labeling. Existing systems typically train models with mean squared error or cross-entropy loss, which do not account well for uncertainty in data or predictions, leading to overconfident and potentiallyless reliable outputs. These systems also rely on single-point estimates without robust modeling of label variability or model confidence, limitinginterpretability and diagnostic reliability. The  proposed system introduces an uncertainty aware framework that explicitly models both aleatoric (data) and epistemic (knowledge) uncertainties via a unified uncertainty-aware negative log-likelihood loss combined with techniques like test-timeaugmentation and deep ensembling.