A Multi-Site Anti-Interference Neural Network for ASD Classification

Autism spectrum disorder (ASD) is a complex neurodevelopmental that can reduce quality of life and burden families. However, there lack objectivity in clinical diagnosis, so it very important to develop method for early accurate diagnosis. Multi-site data increases sample size statistical power, which convenient training deep learning models. heterogeneity between sites will affect ASD recognition. To solve this problem, we propose multi-site anti-interference neural network classification. The resting state brain functional image provided by the used train classification model. model consists three modules. First, site feature extraction module quantify inter-site heterogeneity, autoencoder dimension. Secondly, presentation extract features. Finally, uses output first two modules as labels inputs multi-task adversarial complete representation not affected confounding sites, realize adaptive results show average accuracy ten-fold cross validation 75.56%, better than existing studies. innovation our proposed lies problem traditional single-task be interfere with Our eliminates influence factors on through training, adapt data. Meanwhile, large-scale 1DconV introduced features network, provides support interpretability This expected take advantage multiple provide reference diagnosis treatment ASD.