3D Compressed Convolutional Neural Network Differentiates Neuromyelitis Optical Spectrum Disorders From Multiple...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Magnetic resonance imaging (MRI) is vital in medical diagnostics.
• Deep learning (DL) shows promise for analyzing complex medical imaging data.
• Differentiating rare diseases like neuromyelitis optical spectrum disorder (NMOSD) from multiple sclerosis (MS) using MRI is challenging due to overlapping lesions.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for the automatic differentiation of NMOSD from MS using MRI.
• To address the limitations of traditional models in analyzing rare diseases with subtle, overlapping pathologies.

Main Methods:

• Proposed a novel 3D convolutional neural network (CNN) model with a two-view compression block to process T2-FLAIR MRI images.
• Compared the novel model's performance against traditional 3D CNNs.
• Utilized transfer learning by pre-training the model on the ImageNet dataset.

Main Results:

• The novel 3D compressed CNN models demonstrated superior performance compared to traditional 3D CNNs.
• Pre-trained models achieved average accuracies of 0.75 (34-layer) and 0.725 (18-layer), with high sensitivities and specificities.
• Traditional 3D CNNs lacked the capacity to effectively distinguish between NMOSD and MS.

Conclusions:

• The proposed 3D compressed CNN model can automatically differentiate NMOSD from MS.
• This approach is effective for diagnosing rare diseases with small datasets and scattered, overlapping lesions.
• The novel model offers a significant advancement over traditional methods for challenging neurological condition diagnosis.