A Reality Check of Vision-Language Pre-training in Radiology: Have We Progressed Using Text?

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Silva-Rodríguez, Julio, Dolz, Jose, Ayed, Ismail Ben
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912313697632256
author Silva-Rodríguez, Julio
Dolz, Jose
Ayed, Ismail Ben
author_facet Silva-Rodríguez, Julio
Dolz, Jose
Ayed, Ismail Ben
contents Vision-language pre-training has recently gained popularity as it allows learning rich feature representations using large-scale data sources. This paradigm has quickly made its way into the medical image analysis community. In particular, there is an impressive amount of recent literature developing vision-language models for radiology. However, the available medical datasets with image-text supervision are scarce, and medical concepts are fine-grained, involving expert knowledge that existing vision-language models struggle to encode. In this paper, we propose to take a prudent step back from the literature and revisit supervised, unimodal pre-training, using fine-grained labels instead. We conduct an extensive comparison demonstrating that unimodal pre-training is highly competitive and better suited to integrating heterogeneous data sources. Our results also question the potential of recent vision-language models for open-vocabulary generalization, which have been evaluated using optimistic experimental settings. Finally, we study novel alternatives to better integrate fine-grained labels and noisy text supervision.
format Preprint
id arxiv_https___arxiv_org_abs_2504_05227
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Reality Check of Vision-Language Pre-training in Radiology: Have We Progressed Using Text?
Silva-Rodríguez, Julio
Dolz, Jose
Ayed, Ismail Ben
Computer Vision and Pattern Recognition
Vision-language pre-training has recently gained popularity as it allows learning rich feature representations using large-scale data sources. This paradigm has quickly made its way into the medical image analysis community. In particular, there is an impressive amount of recent literature developing vision-language models for radiology. However, the available medical datasets with image-text supervision are scarce, and medical concepts are fine-grained, involving expert knowledge that existing vision-language models struggle to encode. In this paper, we propose to take a prudent step back from the literature and revisit supervised, unimodal pre-training, using fine-grained labels instead. We conduct an extensive comparison demonstrating that unimodal pre-training is highly competitive and better suited to integrating heterogeneous data sources. Our results also question the potential of recent vision-language models for open-vocabulary generalization, which have been evaluated using optimistic experimental settings. Finally, we study novel alternatives to better integrate fine-grained labels and noisy text supervision.
title A Reality Check of Vision-Language Pre-training in Radiology: Have We Progressed Using Text?
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.05227