Medical Image Classification Based on Transformer Model and Ordinal Loss

Yan Liu

2024

Abstract

This study centers on the application of transformer models for general medical image classification, a crucial step towards automating medical diagnostics. By comparing transformer models with classical methods across diverse medical image datasets, this research aims to enhance performance on specific tasks within these datasets. The core model, Medical Vision Transformer (MedViT), effectively learns multi-scale features by integrating convolutional layers with specialized transformer modules, thereby catering to various medical image classification tasks across different categories. Moreover, this study introduces Ordinal Loss to augment the model's performance on ordinal regression subtasks. Unlike conventional cross-entropy loss, Ordinal Loss facilitates improved learning of sequential relationships between categories. Experiments conducted on MedMNIST validate that MedViT surpasses classical methods on most datasets, with Ordinal Loss further enhancing performance on ordinal regression subtasks. Visual analysis also confirms that the new loss function aids the model in effectively discerning key differences between adjacent categories. This research demonstrates the feasibility of employing a general-purpose transformer model to address medical image classification challenges across multiple domains. Additionally, plug-and-play modules can be leveraged to optimize the model for specific tasks, underscoring its versatility and potential for broader application in medical diagnostics.

Paper Citation

in Harvard Style

Liu Y. (2024). Medical Image Classification Based on Transformer Model and Ordinal Loss. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 708-713. DOI: 10.5220/0012969200004508

in Bibtex Style

@conference{emiti24,
author={Yan Liu},
title={Medical Image Classification Based on Transformer Model and Ordinal Loss},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={708-713},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012969200004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - Medical Image Classification Based on Transformer Model and Ordinal Loss
SN - 978-989-758-713-9
AU - Liu Y.
PY - 2024
SP - 708
EP - 713
DO - 10.5220/0012969200004508
PB - SciTePress