<p>The early identification of Autism Spectrum Disorder (ASD) remains a critical challenge in neurodevelopmental research, with current diagnostic processes often delayed by subjective assessments and limited clinical resources. This paper presents a memory-efficient Neural Architecture Search framework that autonomously identifies optimum neural network structures for ASD classification. Unlike existing genetic algorithm-based NAS approaches requiring over 16GB GPU memory, our framework achieves 76% memory reduction while maintaining superior performance. Our approach presents three key innovations: (1) a novel search space integrating simple, residual, and bottleneck operations with <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({\mathcal {O}}(3^L)\)</EquationSource> </InlineEquation> complexity for <i>L</i> layers; (2) a memory-efficient genetic algorithm that decreases GPU memory consumption by <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(76\%\)</EquationSource> </InlineEquation> relative to current methodologies while preserving search efficacy; and (3) an adaptive fitness function that equilibrates model performance with computational complexity. Through comprehensive experiments utilizing a substantial dataset (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(N = 1,262,856\)</EquationSource> </InlineEquation>; <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(n_{ASD} = 501,733\)</EquationSource> </InlineEquation>, <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(n_{TD} = 502,541\)</EquationSource> </InlineEquation>), our methodology attained a classification accuracy of <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(95.23\%\)</EquationSource> </InlineEquation> (<InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(95\%\)</EquationSource> </InlineEquation> CI: 94.89-<InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(95.57\%\)</EquationSource> </InlineEquation>) and area under the Receiver Operating Characteristic (ROC) curve of 0.986, which markedly surpassed existing state-of-the-art techniques (traditional CNN: 92.3%, ResNet-based: 94.1%, LSTM: 93.7%). The framework achieves this performance with 2.8M parameters and 15ms processing time per sample, demonstrating practical viability for clinical deployment in resource-constrained settings where current diagnostic procedures extend 4-5 years after symptom onset.</p>

GeneticNAS: a novel self-evolving neural architecture for advanced ASD screening

期刊论文
出版模式：

开放获取
发表日期：

2026年1月27日

查看全文

Abdullah R. Alzahrani,
Dabiah Alboaneen,
Ibrahim R. Alzahrani

摘要

The early identification of Autism Spectrum Disorder (ASD) remains a critical challenge in neurodevelopmental research, with current diagnostic processes often delayed by subjective assessments and limited clinical resources. This paper presents a memory-efficient Neural Architecture Search framework that autonomously identifies optimum neural network structures for ASD classification. Unlike existing genetic algorithm-based NAS approaches requiring over 16GB GPU memory, our framework achieves 76% memory reduction while maintaining superior performance. Our approach presents three key innovations: (1) a novel search space integrating simple, residual, and bottleneck operations with \({\mathcal {O}}(3^L)\) complexity for L layers; (2) a memory-efficient genetic algorithm that decreases GPU memory consumption by \(76\%\) relative to current methodologies while preserving search efficacy; and (3) an adaptive fitness function that equilibrates model performance with computational complexity. Through comprehensive experiments utilizing a substantial dataset ( \(N = 1,262,856\) ; \(n_{ASD} = 501,733\) , \(n_{TD} = 502,541\) ), our methodology attained a classification accuracy of \(95.23\%\) ( \(95\%\) CI: 94.89- \(95.57\%\) ) and area under the Receiver Operating Characteristic (ROC) curve of 0.986, which markedly surpassed existing state-of-the-art techniques (traditional CNN: 92.3%, ResNet-based: 94.1%, LSTM: 93.7%). The framework achieves this performance with 2.8M parameters and 15ms processing time per sample, demonstrating practical viability for clinical deployment in resource-constrained settings where current diagnostic procedures extend 4-5 years after symptom onset.