Towards Scalable and Self-Supervised AI for Surgical Intelligence and Ultrasound Therapies

Title of the Talk: Towards Scalable and Self-Supervised AI for Surgical Intelligence and Ultrasound Therapies
Speakers: Dr. Vinkle Srivastav
Host Faculty: Dr.Rajesh Kedia
Date: Aug 11, 2025
Time: 11:00 am to 12:00pm
Venue: CSE-LH-1

Abstract:

Artificial Intelligence is rapidly transforming healthcare, but its clinical scalability is constrained by the need for large annotated datasets and computational bottlenecks. In this talk, I will present self-supervised and label-efficient methods that reduce dependence on manual annotations while ensuring robust performance in real-world clinical settings. I will begin with exploring how external calibrated camera views can be used to develop scalable methods for surgical scene understanding. I will share our work on self-supervised techniques for 2D and 3D human pose estimation in the OR, multi-view models for estimating 3D pose and associating identities across views, and methods that recognize team activities during surgery. Next, I will share approaches for internal laparoscopic surgical videos, and share how foundation models can expand the capabilities of AI in surgery. I will highlight our research on multimodal vision-language pre-training for surgical video analysis, which enables zero-shot and few-shot learning, as well as semantic reasoning across diverse surgical procedures. Finally, I will discuss our recent efforts to accelerate and personalize therapeutic interventions. Specifically, I will introduce a deep learning framework for transcranial focused ultrasound simulation that combines patient-specific anatomical information to deliver fast and accurate pressure field predictions. These research directions aim to build intelligent, adaptable, and clinically deployable AI systems for next-generation surgical and therapeutic care.

Speaker Bio: Dr. Vinkle Srivastav is a Research Scientist in the CAMMA group, a collaborative research team between IHU Strasbourg and the University of Strasbourg, with a focus on surgical data science. He completed his Ph.D. at the University of Strasbourg and worked on unsupervised domain adaptation approaches for human pose estimation in the operating room. Prior to this, he developed AI-assisted physical and virtual reality simulators for neurosurgery skills evaluation at the All-India Institute of Medical Science, Delhi. His research interests are surgical activity recognition, vision-language models, self-supervised learning, multi-view 3D human pose estimation, and scientific simulation