Labial-gland artificial intelligence model screening for autoimmune thyroiditis among patients with connective tissue...

The aim of this study is to construct a deep learning-based prediction model to accurately predict the risk of autoimmune thyroiditis (AIT) in patients with connective tissue disease (CTD) using whole section images (WSI) of labial gland pathological tissue.

This was a retrospective study. The labial gland pathological sections of total 121 CTD patients were collected. According to the results of thyroid autoantibodies, including thyroglobulin antibody (TgAb) and thyroid peroxidase antibody (TPOAb), the patients were divided into positive group (Ab+ Group) and negative group (Ab- Group). The pre-trained model EfficientNet-B5 was used to extract image features, and combined with multi-instance learning and ensemble learning techniques, the high-risk prediction model for CTD patients with AIT was constructed.

The integrated model showed excellent prediction performance in both the internal validation set and the external validation set, with the area under the receiver operating characteristic curve (AUC) of 0.829. At the same time, the model can effectively identify the key pathological features of labial gland tissues related to the high risk of AIT in CTD patients.

This study confirmed that the prediction model of labial gland WSI based on deep learning had good efficacy in evaluating the risk of AIT in CTD patients, which provided a new technical support and theoretical basis for early clinical identification of high-risk groups and optimization of diagnosis and treatment decisions.