轻量化几何流形深度网络的自闭症诊断方法

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2026）03-0070-06

Lightweight Geometric Manifold Deep Network for Autism Spectrum Disorder Diagnosis

WUJinying，MAHuibin （School of Information and Electronic Technology， Jiamusi University，Jiamusi 154oo7， China）

Abstract： AutismSpectrum Disorder （ASD）is acommoncentral neurodevelopmental disorder Its clinical diagnosis faces problemsof strong subjectivityandinsuficient accuracy.Toimprovethe performance offunctional Magnetic Resonance Imaging （fMRI） in computer-aided diagnosis，this paper constructs aLightweight Geometric Manifold Dep Network （LGMD-Net）.This network adoptsamulti-channeltwo-dimensionalresidualstructurecombined withmedical geometricprors toobtainthreedimensionalinformation，andachieves efcient featureextractionthrough lightweightresiualblocks.Inthefeaturedimension reductionstage，this paper introduces Spatial PyramidPooling tocapture multi-scale spatial features，andcombines Manifold Mixuptechnologytorealizefeatureenhancementandregularization.This paperconducts experimentsontheABIDEdatasetThe results showthatthis methodoutperformsConvolutionalNeuralNetworks，Transformer，andGraphNeuralNetworkmodels in termsofclassifcationaccuracyF1soreadAUC.Meanhle，tismetodpoessessigcantadvatagesinarateale inferencetime，and memoryusage.TheresearchresultsverifythefeasibilityofLGMD-Net inimprovingdiagnostic accuracyand engineering application.

Keywords： ASD; fMRI; lightweight network; Spatial Pyramid Pooling;Manifold Mixup

0 引言

自闭症谱系障碍（AutismSpectrumDisorder，ASD）是一种广泛性中枢神经系统发育障碍。（剩余8776字）