An Empirical Analysis for Zero-Shot Multi-Label Classification on COVID-19 CT Scans and Uncurated Reports

TL;DR

This paper evaluates zero-shot multi-label classification of COVID-19 CT scans using contrastive visual language models, highlighting their potential to assist radiologists in detailed lung analysis from unstructured data.

Contribution

It introduces an empirical analysis of zero-shot models for fine-grained, multi-label COVID-19 CT scan classification, addressing a gap in medical multimodal pretraining research.

Findings

Zero-shot models can assist in detecting pulmonary embolisms.

Models effectively identify ground glass opacities and consolidations.

The study highlights challenges and future directions in medical image analysis.

Abstract

The pandemic resulted in vast repositories of unstructured data, including radiology reports, due to increased medical examinations. Previous research on automated diagnosis of COVID-19 primarily focuses on X-ray images, despite their lower precision compared to computed tomography (CT) scans. In this work, we leverage unstructured data from a hospital and harness the fine-grained details offered by CT scans to perform zero-shot multi-label classification based on contrastive visual language learning. In collaboration with human experts, we investigate the effectiveness of multiple zero-shot models that aid radiologists in detecting pulmonary embolisms and identifying intricate lung details like ground glass opacities and consolidations. Our empirical analysis provides an overview of the possible solutions to target such fine-grained tasks, so far overlooked in the medical multimodal pretraining literature. Our investigation promises future advancements in the medical image analysis community by addressing some challenges associated with unstructured data and fine-grained multi-label classification.