Early Autism Spectrum Disorder Detection Using Adaptive Fused Spatial-Temporal Graph Convolutional Network Optimized With Artificial Lemming Algorithm.
L’essentiel
Autism spectrum disorder (ASD) is characterized by diversity of behavioural abnormalities, and successful intervention depends on an early diagnosis. Conventional diagnostic techniques depend on interviews and observational evaluations, which can occasionally result in errors. In order to enhance the accuracy of ASD recognition, we propose an early ASD detection using Adaptive Fused Spatial-Temporal Graph Convolutional Network Optimized with Artificial Lemming Algorithm (AFSTGCN-ALA). Initially, ASD-related data are collected from ASD Dataset and Adaptive Fast Desensitized Kalman Filter (AFDKF) preprocessing; it effectively eliminates impulsive noise to enhance the accuracy of the data. The Multi-Synchro Squeezing Transform (MSST) is then used to extract features from the preprocessed data, such as connections and frequency band coherence. To differentiate between instances with and without ASD, these extracted traits are then categorized using the AFSTGCN. However, adaptive optimization for parameter tweaking is absent from traditional AFSTGCN, which might affect the accuracy of detection. In order to solve this, the ALA was introduced, which optimizes the weight parameters of AFSTGCN to guarantee correct ASD categorization. The proposed AFSTGCN-ALA model is evaluated using performance criteria such as compute time, F1-score, accuracy and precision. According to experimental data, the suggested technique outperforms current approaches in terms of accuracy by 18.97%, 24.57% and 32.68% while cutting down on computing time by 19.84%, 24.93% and 31.62%. These results open the door to more accurate diagnosis and prompt therapies by demonstrating the effectiveness and dependability of AFSTGCN-ALA for early ASD identification.
Synthèse détaillée
Résumé original
Autism spectrum disorder (ASD) is characterized by diversity of behavioural abnormalities, and successful intervention depends on an early diagnosis. Conventional diagnostic techniques depend on interviews and observational evaluations, which can occasionally result in errors. In order to enhance the accuracy of ASD recognition, we propose an early ASD detection using Adaptive Fused Spatial-Temporal Graph Convolutional Network Optimized with Artificial Lemming Algorithm (AFSTGCN-ALA). Initially, ASD-related data are collected from ASD Dataset and Adaptive Fast Desensitized Kalman Filter (AFDKF) preprocessing; it effectively eliminates impulsive noise to enhance the accuracy of the data. The Multi-Synchro Squeezing Transform (MSST) is then used to extract features from the preprocessed data, such as connections and frequency band coherence. To differentiate between instances with and without ASD, these extracted traits are then categorized using the AFSTGCN. However, adaptive optimization for parameter tweaking is absent from traditional AFSTGCN, which might affect the accuracy of detection. In order to solve this, the ALA was introduced, which optimizes the weight parameters of AFSTGCN to guarantee correct ASD categorization. The proposed AFSTGCN-ALA model is evaluated using performance criteria such as compute time, F1-score, accuracy and precision. According to experimental data, the suggested technique outperforms current approaches in terms of accuracy by 18.97%, 24.57% and 32.68% while cutting down on computing time by 19.84%, 24.93% and 31.62%. These results open the door to more accurate diagnosis and prompt therapies by demonstrating the effectiveness and dependability of AFSTGCN-ALA for early ASD identification.