Development of a transformer-based deep learning algorithm for diabetic peripheral neuropathy classification using...

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Diabetic peripheral neuropathy (DPN) is a common complication of diabetes that often progresses unnoticed.
• Early detection of DPN is crucial for timely intervention and management to prevent severe complications.

Purpose of the Study:

• To develop and evaluate a transformer-based deep learning algorithm (DLA) for classifying corneal confocal microscopy (CCM) images to identify DPN in diabetic patients.
• To compare the performance of the transformer-based DLA against traditional convolutional neural network (CNN) models.

Main Methods:

• A dataset of 940 CCM images from 94 participants (57 with DPN, 37 without) was utilized.
• A Swin transformer network with a hierarchical architecture was employed for image classification.
• The dataset was randomly split into training, validation, and test sets at a 7:1:2 ratio, ensuring participant-level separation.

Main Results:

• The transformer-based DLA demonstrated high diagnostic accuracy, achieving an area under the curve (AUC) of 0.9405 for participant-level classification and 0.8996 for image-level classification.
• The model outperformed established CNN architectures like ResNet50, Inception_v3, and DenseNet121 in single-image predictions.
• Grad-CAM visualization techniques were used to interpret the model's decision-making process.

Conclusions:

• Transformer-based deep learning algorithms show superior performance compared to CNNs for rapid binary classification of DPN from CCM images.
• The developed DLA holds significant potential for clinical application in the early screening and diagnosis of DPN.