Construction of a Multimodal Machine Learning Model for Papillary Thyroid Carcinoma Based on Pathomics and Ultrasound Radiomics Dataset.

The use of machine learning to integrate and analyse multimodal information has broad prospects for enhancing the precision of tumour diagnosis. Our study constructed a multimodal diagnostic model for papillary thyroid carcinoma (PTC) by integrating pathomics and ultrasound radiomics characteristics using artificial intelligence machine learning methods, aiming to improve the efficiency of pathologists. A retrospective analysis was conducted on 222 cases with postoperative diagnoses of PTC and 163 cases with postoperative diagnoses of benign thyroid nodules. Scanning was used to obtain cytopathology digitised images, cut image blocks and outline the corresponding ultrasound imaging lesions to extract pathomics and ultrasound radiomics features. Three methods, as eXtreme gradient boosting (XGBoost), support vector machine (SVM), and random forest (RF) algorithms, were applied to construct PTC cytopathological diagnostic models. The efficacy of the models was evaluated and validated with the area under receiver operating characteristic curve (AUC), and the performance of single-modal, multimodal, and artificial diagnostic models was compared.