Title: UVSC-MT: an uncertain-guided mean teacher model with virtual adversarial and sample relation consistency for semi-supervised medical image classification

Authors: Kai Wu; Yubin Wei; Jinting Guan

Addresses: Department of Automation, Xiamen University, Xiamen, China; Key Laboratory of System Control and Information Processing, Ministry of Education, Shanghai, China; Department of Automation, Xiamen University Tan Kah Kee College, Zhangzhou, China ' Department of Automation, Xiamen University, Xiamen, China ' Department of Automation, Xiamen University, Xiamen, China; Key Laboratory of System Control and Information Processing, Ministry of Education, Shanghai, China; National Institute for Data Science in Health and Medicine, Xiamen University, Xiamen, China

Abstract: Semi-supervised deep learning models have been developed for medical image classification, enabling the utilisation of both labelled and unlabelled data. Among these, the Teacher-Student model has gained significant recognition. Despite its success, this model faces two critical limitations, the unreliability of target knowledge generated by the teacher model and the model's limited robustness and generalisation capabilities. We propose a semi-supervised medical image classification framework, termed uncertain-guided mean teacher model with virtual adversarial and sample relation consistency (UVSC-MT). Drawing inspiration from Monte Carlo uncertainty, we leverage uncertainty information to guide the student model toward more reliable and meaningful target learning. Furthermore, we incorporate sample relation consistency to enable the extraction of additional semantic insights from unlabelled data and thereby enhance utilisation. Additionally, we integrate virtual adversarial training as a regularisation strategy to bolster the model's classification performance and robustness. Experiments conducted on diverse datasets demonstrate that UVSC-MT surpasses existing methods, offering advancements and insights for semi-supervised learning in medical image classification.

Keywords: medical image classification; semi-supervised learning; uncertainty; sample relation consistency; virtual adversarial.

DOI: 10.1504/IJSCIP.2026.154733

International Journal of System Control and Information Processing, 2026 Vol.5 No.1, pp.30 - 48

Received: 02 Sep 2025
Accepted: 30 Dec 2025
Published online: 12 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article