A Hybrid CNN and ML Framework for Multi-modal Classification of Movement Disorders Using MRI and Brain Structural Features

TL;DR

This paper tackles the challenge of early differential diagnosis among APD subtypes and PD by presenting a two-stage hybrid framework that fuses multi-modal data—T1-weighted MRI, segmentation masks of 12 deep brain structures, and their corresponding volumes—through a tri-branch 3D CNN and a logistic regression classifier trained on CNN features plus volumetrics. The model achieves high discriminative performance across PSP vs. PD (AUC $0.95$), MSA vs. PD (AUC $0.86$), and PSP vs. MSA (AUC $0.92$), with ablation analyses showing the benefit of multi-modal fusion over single modalities. Interpretability is addressed with 3D Grad-CAM, revealing population-average attention maps that concentrate on disease-relevant regions such as the midbrain, ventricles, and striatum, aligning with known neuropathology. The framework offers a clinically relevant, explainable tool for early APD differentiation and lays groundwork for future multi-class and longitudinal investigations.

Abstract

Atypical Parkinsonian Disorders (APD), also known as Parkinson-plus syndrome, are a group of neurodegenerative diseases that include progressive supranuclear palsy (PSP) and multiple system atrophy (MSA). In the early stages, overlapping clinical features often lead to misdiagnosis as Parkinson's disease (PD). Identifying reliable imaging biomarkers for early differential diagnosis remains a critical challenge. In this study, we propose a hybrid framework combining convolutional neural networks (CNNs) with machine learning (ML) techniques to classify APD subtypes versus PD and distinguish between the subtypes themselves: PSP vs. PD, MSA vs. PD, and PSP vs. MSA. The model leverages multi-modal input data, including T1-weighted magnetic resonance imaging (MRI), segmentation masks of 12 deep brain structures associated with APD, and their corresponding volumetric measurements. By integrating these complementary modalities, including image data, structural segmentation masks, and quantitative volume features, the hybrid approach achieved promising classification performance with area under the curve (AUC) scores of 0.95 for PSP vs. PD, 0.86 for MSA vs. PD, and 0.92 for PSP vs. MSA. These results highlight the potential of combining spatial and structural information for robust subtype differentiation. In conclusion, this study demonstrates that fusing CNN-based image features with volume-based ML inputs improves classification accuracy for APD subtypes. The proposed approach may contribute to more reliable early-stage diagnosis, facilitating timely and targeted interventions in clinical practice.

Figures (3)

• Figure 1: The workflow of the proposed hybrid CNN and ML classification framework
• Figure 2: Confusion matrices for: (left) PSP vs. PD; (middle) PD vs. MSA; (right) PSP vs. MSA.
• Figure 3: 3D Grad-CAM attention maps for PD vs. PSP classification. The figure shows population-averaged class-discriminative attention maps computed on the held-out test set for the ventricular system (left), striatum (middle), and brainstem (right) branches. Attention maps are overlaid on the MNI (Montreal Neurological Institute) template. Warmer colors (red/yellow) indicate regions with a higher contribution to the model’s classification decision. The model consistently attends to the midbrain and ventricular regions, which are known to be associated with PSP-related neurodegeneration.