Multimodal Retinal Imaging Classification for Parkinson's Disease Using a Convolutional Neural Network.

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Retinal structural and microvascular changes correlate with brain alterations.
• Previous research indicates distinct retinal differences in individuals with Parkinson's disease (PD).
• Machine learning models have shown promise in detecting neurodegenerative diseases like Alzheimer's from retinal data.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for classifying Parkinson's disease (PD) using multimodal retinal imaging.
• To assess the efficacy of different retinal imaging modalities in PD detection via AI.

Main Methods:

• A CNN was trained on optical coherence tomography (OCT) ganglion cell-inner plexiform layer (GC-IPL) thickness maps, OCT angiography macular images, and ultra-widefield (UWF) fundus photographs.
• The model utilized a pretrained VGG19 feature extractor with image-specific transformations.
• Performance was evaluated using receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) values.

Main Results:

• The study included 371 eyes from 249 controls and 75 eyes from 52 PD subjects.
• The best CNN variant achieved a high AUC of 0.918 in classifying PD.
• Ultra-widefield (UWF) color photographs proved to be the most effective imaging input, while GC-IPL thickness maps were least contributory.

Conclusions:

• A pilot CNN successfully identified individuals with Parkinson's disease from retinal images, demonstrating proof of concept.
• This study highlights the potential of retinal imaging and AI for PD detection.
• Larger imaging datasets are needed to develop clinical-grade algorithms for automated PD diagnosis.