Specialised or Generic? Tokenization Choices for Radiology Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Warr, Hermione, Xu, Wentian, Anthony, Harry, Ibrahim, Yasin, McGowan, Daniel, Kamnitsas, Konstantinos
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916896563003392
author Warr, Hermione
Xu, Wentian
Anthony, Harry
Ibrahim, Yasin
McGowan, Daniel
Kamnitsas, Konstantinos
author_facet Warr, Hermione
Xu, Wentian
Anthony, Harry
Ibrahim, Yasin
McGowan, Daniel
Kamnitsas, Konstantinos
contents The vocabulary used by language models (LM) - defined by the tokenizer - plays a key role in text generation quality. However, its impact remains under-explored in radiology. In this work, we address this gap by systematically comparing general, medical, and domain-specific tokenizers on the task of radiology report summarisation across three imaging modalities. We also investigate scenarios with and without LM pre-training on PubMed abstracts. Our findings demonstrate that medical and domain-specific vocabularies outperformed widely used natural language alternatives when models are trained from scratch. Pre-training partially mitigates performance differences between tokenizers, whilst the domain-specific tokenizers achieve the most favourable results. Domain-specific tokenizers also reduce memory requirements due to smaller vocabularies and shorter sequences. These results demonstrate that adapting the vocabulary of LMs to the clinical domain provides practical benefits, including improved performance and reduced computational demands, making such models more accessible and effective for both research and real-world healthcare settings.
format Preprint
id arxiv_https___arxiv_org_abs_2508_09952
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Specialised or Generic? Tokenization Choices for Radiology Language Models
Warr, Hermione
Xu, Wentian
Anthony, Harry
Ibrahim, Yasin
McGowan, Daniel
Kamnitsas, Konstantinos
Computation and Language
Artificial Intelligence
Machine Learning
The vocabulary used by language models (LM) - defined by the tokenizer - plays a key role in text generation quality. However, its impact remains under-explored in radiology. In this work, we address this gap by systematically comparing general, medical, and domain-specific tokenizers on the task of radiology report summarisation across three imaging modalities. We also investigate scenarios with and without LM pre-training on PubMed abstracts. Our findings demonstrate that medical and domain-specific vocabularies outperformed widely used natural language alternatives when models are trained from scratch. Pre-training partially mitigates performance differences between tokenizers, whilst the domain-specific tokenizers achieve the most favourable results. Domain-specific tokenizers also reduce memory requirements due to smaller vocabularies and shorter sequences. These results demonstrate that adapting the vocabulary of LMs to the clinical domain provides practical benefits, including improved performance and reduced computational demands, making such models more accessible and effective for both research and real-world healthcare settings.
title Specialised or Generic? Tokenization Choices for Radiology Language Models
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2508.09952