What question did this study set out to answer?

This research aims to enhance the clinical diagnosis and prediction of amyotrophic lateral sclerosis (ALS) using deep learning techniques.

March 15, 2026

Enhancing Clinical Diagnosis of Amyotrophic Lateral Sclerosis through Advanced Prediction and Analysis Using Deep Learning

Key Points

This research aims to enhance the clinical diagnosis and prediction of amyotrophic lateral sclerosis (ALS) using deep learning techniques.
Proposed a Bidirectional Three-Dimensional Quasi-Recurrent Neural Network (PALS-Bi-3DQRNN) for ALS prediction.
Utilized images from the Canadian ALS Neuroimaging Consortium (CALSNIC) dataset for analysis.
Implemented the Unsharp Mask Guided Filtering (UMGF) for motion correction during image pre-processing.
Employed Spatial-Frequency Transformer (SFT) for feature extraction of images.
Optimized Bi-3DQRNN using the Waterwheel Plant Optimization Algorithm (WPOA) for improved performance.
Achieved higher sensitivity by 22.36%, accuracy by 19.36%, and F1-score by 28.27% compared to existing techniques.
PALS-Bi-3DQRNN outperformed existing models in ALS classification.
Results included metrics like sensitivity, precision, accuracy, and ROC for assessing effectiveness.

Abstract

Amyotrophic Lateral Sclerosis (ALS) is a neurological illness that progresses unabated and currently lacks an effective disease-modifying treatment. In this research, the prediction of Amyotrophic Lateral Sclerosis using Bidirectional Three-Dimensional Quasi-Recurrent Neu-ral Network (PALS-Bi-3DQRNN) is proposed. Initially, the input image is gathered from the Canadian ALS Neuroimaging Consortium (CALSNIC) dataset. Then, the image is provided for the pre-processing phase. During the pre-processing phase, the Unsharp Mask Guided Filtering (UMGF) is used for motion correction. The pre-processed images are provided to the feature extraction process; the Spatial-Frequency Transformer (SFT) is used to extract local and global discriminative features, such as colour distribution, spatial layout, edges, and textures. Although Bi-3DQRNN can learn discriminative representations via gradient-based training, its performance depends strongly on the selection of hyperparameters. Therefore, the Waterwheel Plant Optimization Algorithm (WPOA) is employed to optimally tune the Bi-3DQRNN parameters, leading to improved classification of ALS and healthy subjects. The proposed method is carried out using Python. Several performance metrics are used to assess the effectiveness of the proposed PALS-Bi-3DQRNN method, including Sen-sitivity, Precision, Accuracy, F1-score, Specificity, and Receiver Operating Characteristic (ROC). The proposed PALS-Bi-3DQRNN method attains 22.36%, 15.42% and 18.27% higher sensitivity, 19.36%, 26.42% and 23.27% higher accuracy, 28.27%, 19.36% and 23.42% higher F1-score contrasted with existing techniques, Examples include the identifi-cation of manifold learning for amyotrophic lateral sclerosis functional loss assessment (ML-ALS-FLA), amyotrophic lateral sclerosis from multi-centre magnetic resonance imag-ing data using a spatial and frequency fusion transformer (ALS-MCMD-SFFT), and the comparison of King’s clinical staging in multinational amyotrophic lateral sclerosis cohorts (CKCS-MALSC).

Bookmark

Enhancing Clinical Diagnosis of Amyotrophic Lateral Sclerosis through Advanced Prediction and Analysis Using Deep Learning

Key Points

Abstract

Cite This Study