The value of predictive models assisted diagnosis of benign and malignant lacrimal gland tumor based on dynamic contrast-enhanced combined with conventional MRI.

Abstract

Background or purpose: To explore the value of predictive models assisted diagnosis of benign and malignant lacrimal gland tumor based on dynamic contrast enhanced (DCE) combined with conventional MRI.

Methods: This study retrospectively analyzed the patients with primary lacrimal gland tumors using conventional and DCE-MRI. 15 base predictive models were conducted to compare the diagnostic value.

Results: A total number of 76 patients were enrolled, including 41 were malignant, 35 were benign. Compared with the benign and malignant tumors, MRI parameters including tumor diameter, tumor shape, adjacent bone damage, K^trans, K_ep, V_e and TIC were selected into predictive models. Six-performance metrics of models with DCE quantitative parameters was significantly better than the models without. According to the test results, Extra Trees Classifier (ET) showed the highest MCC and F1 score, the accuracy was 91 %, precision was 100 %, and the recall was 83 %, based on DCE quantitative parameters plus conventional MRI features.

Discussion or conclusions: Our findings indicate that the predictive models based on DCE quantitative parameters plus conventional MRI features have shown better performance than the models based on conventional MRI features alone. ET could be selected as the best model in the diagnosis of benign and malignant lacrimal tumors.