Saved in:
Bibliographic Details
Main Authors: Quigley, Keegan, Cha, Miriam, Barua, Josh, Chauhan, Geeticka, Berkowitz, Seth, Horng, Steven, Golland, Polina
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.19635
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912182411722752
author Quigley, Keegan
Cha, Miriam
Barua, Josh
Chauhan, Geeticka
Berkowitz, Seth
Horng, Steven
Golland, Polina
author_facet Quigley, Keegan
Cha, Miriam
Barua, Josh
Chauhan, Geeticka
Berkowitz, Seth
Horng, Steven
Golland, Polina
contents Vision-language pretraining has been shown to produce high-quality visual encoders which transfer efficiently to downstream computer vision tasks. Contrastive learning approaches have increasingly been adopted for medical vision language pretraining (MVLP), yet recent developments in generative AI offer new modeling alternatives. This paper introduces RadTex, a CNN-encoder transformer-decoder architecture optimized for radiology. We explore bidirectional captioning as an alternative MVLP strategy and demonstrate that RadTex's captioning pretraining is competitive with established contrastive methods, achieving a CheXpert macro-AUC of 89.4%. Additionally, RadTex's lightweight text decoder not only generates clinically relevant radiology reports (macro-F1 score of 0.349), but also provides targeted, interactive responses, highlighting the utility of bidirectional captioning in advancing medical image analysis.
format Preprint
id arxiv_https___arxiv_org_abs_2310_19635
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Improving Medical Visual Representations via Radiology Report Generation
Quigley, Keegan
Cha, Miriam
Barua, Josh
Chauhan, Geeticka
Berkowitz, Seth
Horng, Steven
Golland, Polina
Computer Vision and Pattern Recognition
Vision-language pretraining has been shown to produce high-quality visual encoders which transfer efficiently to downstream computer vision tasks. Contrastive learning approaches have increasingly been adopted for medical vision language pretraining (MVLP), yet recent developments in generative AI offer new modeling alternatives. This paper introduces RadTex, a CNN-encoder transformer-decoder architecture optimized for radiology. We explore bidirectional captioning as an alternative MVLP strategy and demonstrate that RadTex's captioning pretraining is competitive with established contrastive methods, achieving a CheXpert macro-AUC of 89.4%. Additionally, RadTex's lightweight text decoder not only generates clinically relevant radiology reports (macro-F1 score of 0.349), but also provides targeted, interactive responses, highlighting the utility of bidirectional captioning in advancing medical image analysis.
title Improving Medical Visual Representations via Radiology Report Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2310.19635