| Summary: | Amyotrophic Lateral Sclerosis (ALS) is a rare disease and also known as Lou Gehrig’s disease. In addition, this disease is characterized by a progression in the nerve cells of the brain and spinal cord. These neuron cells control most of the functions of humans. ALS causes these neurons, the Lower Motor Neurons (LMNs) and the Upper Motor Neurons (UMNs), to deteriorate over time. As of today, there is no cure yet. However, physicians face challenges in diagnosing and preparing suitable treatment plans. Medical industries have implemented numerous technologies and systems to predict ALS to aid physicians. However, predicting ALS in its initial phase is challenging despite the advancement of technologies. To support these advancements, this article presents a Pipeline Multi-Level Encoders-Decoders-Based Ensemble Architecture Technology (PMLEDBEAT) based on dedicated U-shaped and Generative Adversarial Network (GAN) architectures to predict ALS and estimate its development rate. The proposed solution depends on two Neural Networks: UNET and GAN, and four parallel blocks of different convolutional and soft average pool layers to improve the system’s capabilities to capture different spatial dependencies across different patches. This proposed pipeline architecture prioritizes significant data and determines dependencies of the captured data by appointing higher weights to each patch according to their importance to underline crucial local features in the GAN block. In addition, the UNET block extracts substantial global characteristics and catches relations between different data. The prediction in the proposed model is performed on the extracted feature maps using a squeeze and excitation method. This approach was evaluated on a dataset from GitHub for different performance quantities using extensive tests. The proposed method achieved higher accuracy, precision, recall, and f-score compared to some implemented works, which ranged from 84% to 89% since it reached 92% accuracy. PMLEDBEAT reveals its capability to operate locally and globally on the patch level and highlights essential data on the spatial level, which implies that PMLEDBEAT can be applied to healthcare facilities to predict ALS.
|
|---|