Detection of motor nervous disease using deep learning based Duple feature extraction network

Area of Science:

• Neurology
• Computer Science
• Medical Imaging

Background:

• Motor nervous disease (MND) is a progressive neurological disorder affecting motor neurons and voluntary movement.
• Early detection of MND is crucial but challenging due to manual identification's time-consuming nature.
• Automated methods, particularly deep learning, are needed for rapid and accurate MND classification.

Purpose of the Study:

• To propose a novel deep learning-based Duple feature extraction network for early motor nervous disease (MND) identification.
• To enhance the accuracy and speed of MND detection compared to traditional manual methods.
• To develop an automated system for improved early-stage diagnosis of MND.

Main Methods:

• Pre-processing of Diffusion Tensor Imaging (DTI) using a Gaussian adaptive bilateral filter (GAB).
• Dual feature extraction: Colour Information Feature (CIF) with Local and Global sampling (LOG) via LinkNet, and Local Binary Pattern (LBP) with Edge sampling via MobileNet.
• Classification of MND levels using a Deep Neural Network (DNN) fed with extracted colour and texture features.

Main Results:

• The proposed Duple feature extraction network achieved a 99.62% accuracy rate.
• The DNN demonstrated improved F1-scores compared to FNN, GNN, and GRU (by 1.32% to 3.18%).
• Overall accuracy was significantly enhanced compared to CNN, SVM-RFE, MLP, and Tri-planar CNN (by 5.56% to 6.68%).

Conclusions:

• The novel deep learning-based Duple feature extraction framework shows significant promise for the early detection of motor nervous disease.
• The proposed method substantially improves accuracy and F1-scores, outperforming existing models.
• This framework offers a more efficient and accurate approach to diagnosing MND in its early stages.