MULTI-MODAL MULTI-CLASS PARKINSON DISEASE CLASSIFICATION USING CNN and DECISION LEVEL FUSION
Sushanta Kumar Sahu, Ananda S. Chowdhury
Department of Electronics and Telecommunication Engineering, Jadavpur University, Kolkata, India.
International Conference on Pattern Recognition and Machine Intelligence (PReMI 2023): Pattern Recognition and Machine Intelligence pp 737–745
Abstract.
Parkinson’s disease (PD) is the second most common neurodegenerative disorder, as reported by theWorld Health Organization (WHO). In this paper, we propose a direct three-Class PD classification using two different modalities, namely, MRI and DTI. The three classes used for classification are PD, Scans Without Evidence of Dopamine Deficit (SWEDD) and Healthy Control (HC). We use white matter (WM) and gray matter (GM) from the MRI and fractional anisotropy (FA) and mean diffusivity (MD) from the DTI to achieve our goal.We train four separate CNNs on the above four types of data. At the decision level, the outputs of the four CNN models are fused with an optimal weighted average fusion technique. We achieve an accuracy of 95.53% for the direct three-class classification of PD, HC and SWEDD on the publicly available PPMI database. Extensive comparisons including a series of ablation studies clearly demonstrate the effectiveness of our proposed solution.
Keywords: Parkinson’s disease (PD) · Direct three-Class classification · Multimodal Data· Decision level fusion.
We now summarize our contributions as below:
1. We address a direct three-class classification task (PD, HC and SWEDD) for Parkinson’s disease, which is certainly more challenging than the current trend of a single binary classification (for 2-class problem) or multiple binary classifications (for 3-class problem).
2. We make effective use of the underlying potential of multi-modal neuroimaging, namely T1-weighted MRI and DTI. In particular, we train four different CNNs on WM, GM data from MRI and FA, MD data from DTI. Such in-depth analysis of multi-modal neuroimaging data is largely missing in the analysis of PD.
3. Finally, at the decision level, the outputs of each CNN model are fused using an Optimal Weighted Average Fusion (OWAF) strategy to achieve state-of-the-art classification accuracy.