| MS Seminar

Name of the Speaker: Mr. Krishan Agyakari Raja Babu (EE22S042)
Guide: Dr. Mohanasankar S
Online meeting link: https://meet.google.com/xgk-ngbk-esp
Date/Time: 3rd April 2025 (Thursday), 3.30 PM
Title: Synthetic Data for Fair, Trustworthy and Robust AI in Medical Imaging.

Abstract :

Artificial intelligence is revolutionizing medical imaging, enabling automated diagnosis, treatment planning, and outcome prediction. However, the effectiveness of Medical AI models hinges on access to large, diverse, and well-annotated datasets—an ongoing challenge due to data scarcity, privacy concerns, and annotation costs. Synthetic data, generated using deep generative models, offers a promising solution by supplementing real-world datasets, improving model generalization, mitigating data limitations and facilitating robust evaluation.

Despite its potential, the adoption of synthetic data in medical AI faces significant challenges. This seminar explores these challenges through three key perspectives: (1) Fairness – how synthetic data can introduce spurious correlations, leading to the synthetic simplicity bias; (2) Trust – the skepticism surrounding synthetic datasets in clinical settings and strategies to enhance their reliability; and (3) Robustness – the role of synthetic data as a cost-effective solution for evaluating AI performance under real-world distributional shifts, ensuring generalization across diverse patient populations.

Through empirical studies in medical imaging tasks such as cardiac view classification and brain tumor segmentation, this seminar will examine both the risks and benefits of synthetic data, emphasizing strategies for its responsible integration into medical AI. By addressing these challenges, we provide a comprehensive outlook on leveraging synthetic data to develop fair, trustworthy, and robust AI models for real-world clinical applications.