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

Background/aim: Motor neuron diseases (MNDs) are progressive neurological disorders that cause muscle weakness and wasting as a result of ongoing neurodegeneration. MNDs require comprehensive diagnostic approaches that integrate clinical symptoms, laboratory findings, and multimodal imaging data. Materials and methods: In this study, a novel residual network for motor neuron disease detection (RES-MND) framework is proposed for detecting MNDs using multimodal imaging data. Initially, the input multimodal images, including MRI, CT, PET, and DTI, are preprocessed using adaptive dynamic histogram equalization and the total variation bilateral filter. The preprocessed multimodal images are processed through Res4Net-CBAM for feature extraction to enhance image recognition performance. A dove swarm optimization algorithm is employed to select the most relevant features from the multimodal images. Finally, the deep belief network (DBN) classifies five categories, including one control group (normal) and four MND types: ALS, PLS, PBP, and PMA. Results: The performance of the proposed RES-MND method is evaluated using standard metrics such as accuracy, precision, recall, and F1-score. According to the results, the proposed RES-MND method achieved the highest accuracy rate of 99.65%, outperforming existing methods. Conclusion: The proposed DBN achieved 0.68%, 0.41%, and 0.9% higher accuracy than SNN, DNN, and CNN, respectively. The proposed RES-MND method achieved 1.08%, 2.18%, and 1.7% higher overall accuracy compared to existing methods such as miRNA, vGRF, and SVM-RFE, respectively.