Early Diagnosis of Multiple Sclerosis Using Swept-Source Optical Coherence Tomography and Convolutional Neural...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Multiple sclerosis (MS) diagnosis can be challenging in early stages.
• Swept-source optical coherence tomography (SS-OCT) provides detailed retinal imaging.
• Developing automated diagnostic tools is crucial for timely intervention.

Purpose of the Study:

• To implement an automated system for early multiple sclerosis (MS) diagnosis.
• To utilize convolutional neural networks (CNNs) for classifying SS-OCT images.
• To identify key retinal structures for MS detection.

Main Methods:

• SS-OCT images from 48 MS patients and 48 controls were analyzed.
• Retinal layer thicknesses (retina, RNFL, GCL+, GCL++, choroid) were measured.
• A deep convolutional generative adversarial network (GAN) augmented the image dataset for CNN training.

Main Results:

• The GCL++ layer showed the highest discriminant capacity (44.99%).
• Complete retina (26.71%) and GCL+ (22.93%) were also significant.
• A CNN model achieved 1.0 sensitivity and 1.0 specificity.

Conclusions:

• Feature pre-selection combined with CNNs shows promise for early MS diagnosis.
• The proposed method is non-harmful, low-cost, and easy to implement.
• SS-OCT thickness data can effectively aid in early MS detection.