Artificial intelligence for diagnosis and prognosis of thymic epithelial tumors: a systematic review.

Abstract

Background: Thymic epithelial tumors (TETs) are rare, but they are the most common tumors in the anterior mediastinum. The use of artificial intelligence (AI) in the medical field is rapidly advancing. Especially in a rare disease such as TET, AI can help to improve existing care and foster new innovations. However, there are many different AI models with various implications. This systematic review aims to give an overview of recent studies on AI applications for the diagnosis and prognosis of TETs.

Methods: In this systematic review, six electronic databases were searched for research papers with the keywords "thymoma", "thymic epithelial tumors", "thymic carcinoma", "artificial intelligence", "machine learning" and "deep learning". The screening was performed by two reviewers and conflicts were resolved by a third reviewer.

Results: After removal of duplicates, 582 articles were included. Of these, 65 articles were eligible for inclusion after screening. AI models are used mainly for distinguishing between different mediastinal tumor types (n=21), risk stratification (n=26), surgical planning (n=7) and for assisting in subtyping (n=4). For prognostic purposes, AI is used for predicting clinical outcome as well as the chance of metastasis or prognosis (n=7). Models using combined clinical and radiomics data performed better than models with a single type of data. Some of the AI models outperformed physicians or successfully supported physicians in their workflow.

Conclusions: Many AI models have been studied in the context of the diagnosis and prognosis of TETs. Even though results are promising, external validation is often lacking, and sample sizes are small. Therefore, most models are not yet ready for clinical implementation. Further research on AI models with larger datasets and external validation is necessary, with careful consideration of the risk of data leakage.