Binod Bhattarai & Yueming Jin - CMIC/WEISS joint seminar series

Speaker: Binod Bhattarai

Title: Histogram of Oriented Gradients as Pseudo Label to train deep semantic segmentation network

Abstract:

Semantic segmentation is an active research problem in medical image analysis. Recently, deep learning algorithms are becoming effective in tackling such problems. However, to train such models, a large amount of annotated examples are required.  Annotating such a large data set is expensive, time-consuming, and needs experts. Recently, learning frameworks such as self-supervised learning algorithms solving pre-text tasks using pseudo labels are getting popular to generalize the deep architectures better.  In this talk, I present one of our latest MICCAI challenge-winning work to train a semantic segmentation network exploiting the Histogram of Oriented Gradients as pseudo labels for pre-text tasks.  

Keywords: Endoscopic image analysis, semantic segmentation,  semi-supervised learning, Self-supervised learning

Speaker: Yueming Jin

Title: Dynamic Surgical Video Analysis for Intelligent Robotic Surgery

Abstract:

In modern healthcare, the operating room has undergone tremendous transformations evolving into a highly complicated and technologically rich environment. Such transformations innovate the surgery procedure and greatly enhance the patient safety. To better tackle this new scenario, the computer-assisted and robotic-assisted systems have been gradually developed to provide surgeons with the detailed contextual support.

Automatic surgical visual perception has become a crucial component when developing these systems. The exploding amount of surgical videos collected in nowadays clinical centers offer enormous opportunities, by developing a new-generation of data analytics techniques for improving these assisted systems and even revolutionizing healthcare industry. In the meanwhile, the momentum in cutting-edge AI systems is towards representation learning and pattern recognition via data-driven approaches.

In this talk, I will present a series of deep learning methods towards interdisciplinary researches at artificial intelligence for medical image analysis and surgical visual perception, for improving surgical workflow recognition, instrument detection and segmentation, and anatomy tissue semantic parsing. The proposed methods cover a wide range of deep learning topics including design of network architectures, novel learning strategies, multi-task learning, semi-supervision, etc. The challenges, up-to-date progresses and promising future directions of intelligent surgery will also be discussed.