RES-MND: Motor neuron disease detection using Res4Net-convolutional block attention module

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Motor neuron diseases (MNDs) are progressive neurodegenerative disorders causing muscle weakness.
• Accurate diagnosis of MNDs necessitates integrating clinical data, lab findings, and multimodal imaging.
• Current diagnostic methods require enhancement for improved accuracy and efficiency.

Purpose of the Study:

• To propose a novel Residual Network for Motor Neuron Disease detection (RES-MND) framework.
• To leverage multimodal imaging data (MRI, CT, PET, DTI) for enhanced MND detection.
• To improve the diagnostic accuracy and performance of MND classification.

Main Methods:

• Utilized adaptive dynamic histogram equalization and total variation bilateral filter for image preprocessing.
• Employed Res4Net-CBAM for feature extraction and a dove swarm optimization algorithm for feature selection.
• Developed a deep belief network (DBN) for classifying normal controls and four types of MNDs (ALS, PLS, PBP, PMA).

Main Results:

• The RES-MND framework achieved a high accuracy rate of 99.65%.
• The proposed DBN demonstrated superior accuracy compared to SNN, DNN, and CNN.
• RES-MND outperformed existing methods like miRNA, vGRF, and SVM-RFE in overall accuracy.

Conclusions:

• The RES-MND framework offers a highly accurate and effective approach for detecting motor neuron diseases.
• Multimodal imaging combined with deep learning techniques shows significant promise in advancing MND diagnostics.
• This novel framework represents a substantial improvement over current methods for identifying various types of MNDs.