Experimental Evaluation of Amyotrophic Lateral Sclerosis (ALS) Disease Prediction based on Improved Deep Learning Mechanism

Abstract

Amyotrophic Lateral Sclerosis (ALS) also referred to as Lou Gehrig's Disease represents a degenerative neurological disease which devastates motor neurons thus causing weakening muscles before permanently impairing voluntary body movements. The study provides experimental results for predicting Amyotrophic Lateral Sclerosis (ALS) disease through an advanced deep learning mechanism system. The progressive neurodegenerative disorder ALS substantially degrades speech patterns while enabling acoustic analysis as a non-contaminating approach for an early diagnosis. The research used the Amyotrophic Lateral Sclerosis (ALS) dataset acquired from Kaggle which contained 131 acoustic features extracted from vowel phonations of both ALS patients and healthy subjects. A complete preprocessing sequence applied to the dataset consisted of noise reduction and feature scaling and data augmentation by applying SMOTE for handling class imbalance. The model employs Hybrid CNN-GRU-Attention to combine Convolutional Neural Networks (CNNs) as features extractors with Gated Recurrent Units (GRUs) that analyze sequential patterns and an Attention component for identifying key voice elements that indicate ALS. The proposed model reached a test accuracy rate of 95.8% surpassing Elevated Hybrid Deep Learning Strategy (EHDLS) by achieving 94.5% accuracy as its benchmark. The robust performance of the model is verified by Precision at 94.5% and Recall at 96.0% alongside the AUC-ROC Score of 97.6%. The SHAP analysis confirmed that the model identified important acoustic features which match medical significance parameters. The study demonstrates how deep learning technologies can detect ALS outside traditional testing methods while creating new automated diagnostic equipment.

Topics & Concepts