# The Role of Local Alignment and Uniformity in Image-Text Contrastive Learning on Medical Images

Source: [The Role of Local Alignment and Uniformity in Image-Text Contrastive Learning on Medical Images](<https://www.sota2.com/research/paper/68fa6ea91e209131cea4131a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T01:02:50.787Z

Authors: Philip M\\"uller, Georgios Kaissis, Daniel Rueckert

Published: 2022-11-14T10:32:51.000Z

## Abstract

Image-text contrastive learning has proven effective for pretraining medical image models. When targeting localized downstream tasks like semantic segmentation or object detection, additional local contrastive losses that align image regions with sentences have shown promising results. We study how local contrastive losses are related to global (per-sample) contrastive losses and which effects they have on localized medical downstream tasks. Based on a theoretical comparison, we propose to remove some components of local losses and replace others by a novel distribution prior which enforces uniformity of representations within each sample. We empirically study this approach on chest X-ray tasks and find it to be very effective, outperforming methods without local losses on 12 of 18 tasks.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2211.07254>)

[PDF](<https://arxiv.org/pdf/2211.07254.pdf>)