A Novel Dual Interactive Network for Parkinson's Disease Diagnosis Based on Multi-modality Magnetic Resonance Imaging

Parkinson's disease is a neurodegenerative disorder that often leads to abnormal atrophy in certain brain regions. Due to the ability to display brain structures non-invasively, magnetic resonance imaging (MRI) is widely used in the diagnosis of Parkinson's disease. However, existing methods have limited exploration of the rich information present in multi-modality MRI, such as T1 and T2 weighted MRI, and rarely utilized the correlation between them. Therefore, we propose a dual interaction network (DINet) for Parkinson's disease diagnosis using T1 and T2 MRI. Specifically, considering modality specificity, two separate convolutional neural networks are employed for feature extraction from T1 and T2 MRI, respectively. Taking into account the correlation between multi-modality MRI, we introduce a novel modality interaction module DINet to extract cross-modality complementary information at multiple scales. To capture disease-related local and global information, we also propose a scale interaction module to explore the consistency information across adjacent scales. Finally, full-scale features are fused for Parkinson's disease diagnosis. We validate our proposed DINet on the Parkinson's Progression Markers Initiative dataset. Experimental results show our proposed DINet achieves accuracy of 93.11% and 89.85% in Parkinson's disease diagnosis and early diagnosis, respectively. Our proposed DINet not only enables accurate diagnosis of Parkinson's disease but also effectively explores feature interaction mechanisms in modality and scale views.