A Self-Adaptive Network for Multiple Sclerosis Lesion Segmentation From Multi-Contrast MRI With Various Imaging Sequences

Abstract

Deep neural networks have shown promises in the lesion segmentation of multiple sclerosis (MS) from multi-contrast MRI including T1, T2, PD and FLAIR sequences. However, one challenge in deploying such networks into clinical practice is missing MRI sequences due to the variability of image acquisition protocols. Therefore, trained networks need to adapt to practical situations where specific MRI sequences are unavailable. In this paper, we propose a DNN-based MS lesion segmentation framework with a novel technique called sequence dropout. Without altering network architecture, our method ensured the robustness of the network to missing sequences and could achieve its maximal possible performance from a given set of input sequences.Experiments were performed on the IEEE ISBI 2015 Longitudinal MS Lesion Challenge dataset and our method is currently ranked 2nd with a Dice similarity coefficient of 0.684. Experiments also showed our network achieved its maximal performance with one missing sequence during deployment by comparing with separate networks of the same architecture but trained using the corresponding set of input sequences. Our network achieved a non-inferior performance without re-training. Experiments with multiple missing sequences further showed the robustness of our network. Also, with this framework, we studied the quantitative impact of each MRI sequence on the MS lesion segmentation task without training separate networks.