Stanford community & AIMI affiliates only
Talk 1: Radiologist Perspective on The Promise and Reality of AI in Clinic
The overall penetrance of AI in clinics remains low, despite plethora of approved algorithms in market. Only 33% of radiologists reported using any type of AI in 2020, according to a recent study from the American College of Radiology. While algorithms inconsistent performance was identified asthe primary reason for the slow diffusion of AI, radiologists identified other safety concerns as barriers to adopt AI in practice. In this research talk, I will present an overview of the ethical and safety concerns for AI to fulfill its promises from the radiologist’s perspective. I will share insights from our ongoing research to better understand challenges for radiologists to trust and use AI to improve patient care.
Alaa Youssef is a post-doctoral fellow at the Stanford Center for Artificial Intelligence in Medicine and Imaging. She completed her PhD at the Institute of Medical Science, the University of Toronto, Canada. Her post-doctoral research focuses on understanding the bioethical and clinical safety challenges impeding the use of AI applications in healthcare. Specifically, she examines how radiologists interact with AI-systems and investigates the challenges to integrating AI-systems in clinical settings.
Talk 2: Tailor-made: training cardiac magnetic resonance segmentation models with orders of magnitude less labeled data
Segmentation is a powerful tool for quantitative analysis of medical images. Because manual segmentation can be tedious, be time consuming, and have high inter-observer variability, neural networks (NNs) are an appealing solution for automating the segmentation process. However, most approaches to training segmentation NNs rely on large, labeled training datasets that are costly to curate. In this work, we present a general semi-supervised method for training segmentation networks that reduces the amount of labeled data needed to train segmentation neural networks. Instead, we rely on a small set of labeled data and a large set of unlabeled data for training. We evaluate our method on four cardiac magnetic resonance (CMR) segmentation targets and show that by using only 100 labeled training image slices---up to a 99.4% reduction of labeled data---the proposed model achieves within 1.10% of the Dice coefficient achieved by a network trained with over 16,000 labeled image slices. We use the segmentations predicted by our method to derive cardiac functional biomarkers and find no evidence of statistically significant differences in predicted ejection fraction, end diastolic volume, end systolic volume, stroke volume, or left ventricular mass compared an expert annotator.
Sarah Hooper is a PhD candidate at Stanford University, where she works with Christopher Re and Curtis Langlotz. She is broadly interested in applying machine learning to meet needs in healthcare, with a particular interest in applications that make quality healthcare more accessible. Sarah received her B.S. in Electrical Engineering at Rice University in 2017 and her M.S. in Electrical Engineering at Stanford University in 2020.