Classification of rim-enhancing brain abscess, glioblastoma, and brain metastasis using deep learning on multi-modality MRI

Motivation: Classification of rim-enhancing brain abscess, glioblastoma, and brain metastasis solely based on conventional MRI is challenging. Goal(s): To develop a fully automatic system for classification of rim-enhancing lesions using deep learning with multi-modality MRI Approach: An incremental strategy was used to train three models independently, including Model-B (CE-T1WI, T2WI, and T2-FLAIR), Model-BD (plus DWI and ADC), and Model-BDS (plus SWI). Furthermore, 3-T data were included to obtain Model-B+, BD+, and BDS+. Results: Diffusion MRI improves the overall accuracy as expected, and the inclusion of 3T data possibly extend the flexibility of our model, reaching the highest accuracy of 0.796. Impact: Our results demonstrated the value of multi-modality MRI in differentiation of three rim-enhancing lesions. We also highlighted the adaptability of our model on 1.5-T and 3-T data, possibly expanding its clinical use.