A Multi-modality-based Multiple Sclerosis Detection Model

Abstract

This study proposes a novel method for detecting multiple sclerosis (MS) by integrating multi-modality data fusion techniques. Leveraging the complementary information from both health records and magnetic resonance imaging (MRI), our approach aims to enhance the accuracy and reliability of MS detection. We utilized DenseNet 201 to extract features from MRI scans, exploiting its capability to capture intricate patterns in brain images associated with MS pathology. Additionally, we employed bidirectional long short-term memory networks to extract temporal patterns from health records, capturing longitudinal patient data crucial for understanding disease progression. A feature fusion technique was then applied to integrate the extracted features from MRI and health records, combining the spatial information from imaging data with the temporal dynamics captured in health records. Finally, a multi-layer perceptron was employed to perform the final prediction task based on the fused features. The proposed model was experimented with in the Kaggle datasets, covering 271 individuals. Remarkably, our proposed model achieved an impressive accuracy of 99.2% in MS detection, highlighting its effectiveness in leveraging multi-modality data for diagnostic purposes. By combining information from both MRI scans and health records, our approach offers a comprehensive and holistic understanding of the disease, enabling more accurate and timely diagnosis. Additionally, further validation studies in clinical settings are warranted to assess our approach’s real-world utility and clinical impact in improving patient outcomes and facilitating better management of MS.