Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore the cutting-edge applications of diffusion models in medical imaging through this comprehensive 59-minute conference talk presented by AI researchers Onkar Kishor Susladkar and Gayatri Deshmukh at Stanford University's MedAI Group Exchange Sessions. Discover two groundbreaking contributions to the field: MedDelina, a scalable and efficient diffusion transformer specifically designed for medical image segmentation that tackles critical challenges including limited labeled data, poor cross-dataset generalization, and the need for precise anatomical delineation, and ViCTr, an innovative two-stage framework for medical image synthesis that maintains anatomical consistency and pathology realism while significantly enhancing segmentation performance. Learn how MedDelina, trained on the extensive ATLAS-8k dataset, demonstrates superior performance across multiple benchmarks including BTCV, AMOS, Cirr600+, and PanSegData in fine-tuning, few-shot, and zero-shot scenarios, achieving high accuracy with reduced data requirements and minimal performance degradation. Understand how ViCTr's compatibility with various diffusion models enables high image fidelity with fewer diffusion steps, resulting in faster inference times for practical medical applications. Gain insights from Susladkar's extensive experience in AI research spanning continual learning, generative modeling, NLP, and computer vision, along with his work on synthetic CT/MRI data generation and video diffusion at Northwestern University's Machine and Hybrid Intelligence Lab, and from Deshmukh's expertise in multimodality and generative AI, including synthetic data generation and controllable image generation. Participate in the interactive discussion and Q&A session that follows the presentation, designed to critically examine key topics in AI and medicine while fostering collaborative learning and idea generation at the intersection of artificial intelligence and healthcare.
