Sophie Ostmeier, MD
Postdoctoral Scholar
Department of Radiology
Stanford School of Medicine
Title: GREEN: Generative Radiology Report Evaluation and Error Notation
Abstract
Evaluating radiology reports is a challenging problem as factual correctness is extremely important due to the need for accurate medical communication about medical images. Existing automatic evaluation metrics either suffer from failing to consider factual correctness (e.g., BLEU and ROUGE) or are limited in their interpretability (e.g., F1CheXpert and F1RadGraph). In this paper, we introduce GREEN (Generative Radiology Report Evaluation and Error Notation), a radiology report generation metric that leverages the natural language understanding of language models to identify and explain clinically significant errors in candidate reports, both quantitatively and qualitatively. Compared to current metrics, GREEN offers: 1) a score aligned with expert preferences, 2) human interpretable explanations of clinically significant errors, enabling feedback loops with end-users, and 3) a lightweight open-source method that reaches the performance of commercial counterparts. We validate our GREEN metric by comparing it to GPT-4, as well as to error counts of 6 experts and preferences of 2 experts. Our method demonstrates not only higher correlation with expert error counts, but simultaneously higher alignment with expert preferences when compared to previous approaches.
Jeong Hoon Lee, PhD
Postdoctoral Researcher
Department of Radiology
Stanford School of Medicine
Title: Leveraging Patch-Level Representation Learning with Vision Transformer for Prostate Cancer Foundation Models
Abstract:
Recent advancements in self-supervised learning (SSL), emerging as an effective approach for imaging foundation models, enable the effective pretraining of AI models across multiple domains without the need for labels. Despite the rapid advancements, their application in medical imaging remains challenging due to the subtle difference between cancer and normal tissue. To address this limitation, in this study, we propose an AI architecture ProViCNet that employs the vision transformer (ViT) based segmentation architecture with patch-level contrastive learning for better feature representation. We validated our model in prostate cancer detection tasks using three types of magnetic resonance imaging (MRI) across multiple centers. To evaluate the performance of feature representation in this model, we performed downstream tasks with respect to Gleason grade score and race prediction. Our model demonstrated significant performance improvements compared to the state-of-the-art segmentation architectures. This study proposes a novel approach to developing foundation models for prostate cancer imaging overcoming SSL limitations.