Multi-Center 3D CNN for Parkinson's disease diagnosis and prognosis using clinical and T1-weighted MRI data

Abstract

Objective: Parkinson's disease (PD) presents challenges in early diagnosis and progression prediction. Recent advancements in machine learning, particularly convolutional-neural-networks (CNNs), show promise in enhancing diagnostic accuracy and prognostic capabilities using neuroimaging data. The aims of this study were: (i) develop a 3D-CNN based on MRI to distinguish controls and PD patients and (ii) employ CNN to predict the progression of PD.

Methods: Three cohorts were selected: 86 mild, 62 moderate-to-severe PD patients, and 60 controls; 14 mild-PD patients and 14 controls from Parkinson's Progression Markers Initiative database, and 38 de novo mild-PD patients and 38 controls. All participants underwent MRI scans and clinical evaluation at baseline and over 2-years. PD subjects were classified in two clusters of different progression using k-means clustering based on baseline and follow-up UDPRS-III scores. A 3D-CNN was built and tested on PD patients and controls, with binary classifications: controls vs moderate-to-severe PD, controls vs mild-PD, and two clusters of PD progression. The effect of transfer learning was also tested.

Results: CNN effectively differentiated moderate-to-severe PD from controls (74% accuracy) using MRI data alone. Transfer learning significantly improved performance in distinguishing mild-PD from controls (64% accuracy). For predicting disease progression, the model achieved over 70% accuracy by combining MRI and clinical data. Brain regions most influential in the CNN's decisions were visualized.

Conclusions: CNN, integrating multimodal data and transfer learning, provides encouraging results toward early-stage classification and progression monitoring in PD. Its explainability through activation maps offers potential for clinical application in early diagnosis and personalized monitoring.

Keywords: MRI; Machine learning; Parkinson’s disease.

Fig. 1

Model architecture. The features extraction process consists of convolutions and pooling processes to extract relevant features from input MRI volume, the fully connected performs the classification based on the previously extracted features. Blue boxes represent the convolutions. Per each convolution block, the feature map's shape is reported. Orange blocks represent the pooling operation, which consists of convolution stride set as 2. Alternating feature increase and pooling operations are implemented. In the fully connected architecture: yellow circles represent the flattened features, with possibly concatenation of metadata; green circles represent hidden neurons, 1000 elements; red circles represent the model output with softmax activation function.

Fig. 2

K-means clustering based on UPDRS III assessments at baseline and follow-up. Each dot represents a patient. Red dots denote stable PD patients, green dots denote worsening PD patients, and blue dots denote the two cluster centers. Values are standardized by removing the mean and scaling to unit variance. Abbreviations: PD= Parkinson’s disease; UPDRS = Unified Parkinson’s Disease Rating Scale.

Fig. 3

CAM computed by the CNNs for three PD subjects involved, one for each cohort. The rows correspond to the 1) fiftieth, 2) seventieth, and 3) ninetieth layers along the z-axis. The columns correspond to 1) the GradCAM retrieved from the CNN, 2) the input to the model, and 3) the overlap between the CAM and the MRI input. Red areas correspond to areas where critical features concerning outcome 1 (PD patients) are extracted, blue areas correspond to areas where critical features concerning outcome 0 (controls) are extracted. Abbreviations: PD= Parkinson’s disease.