Female autism categorization using CNN based NeuroNet57 and ant colony optimization

Abstract

Autism identification and classification using biomedical medical image analysis has advanced recently. Research shows autistic females have different phenotypic and age-related brain variations than males. Gender-specific hormones and genes affect autistic female brain circuitry, unfortunately, female phenotypic and genotypic data is quite deficient. Since physicians spend much time in assessing autistic females manually. Advanced large-scale deep learning algorithms are in dire need of accurate medical diagnosis. This research proposed a 57-layer CNN architecture called NeuroNet57 that can extract features from fMRI factually. After pre-training on the Brain Tumour dataset, the NeuroNet57 model extracts female phenotypic features from autism brain imagining data exchange (ABIDE)-I+II datasets using T1 modality fMRI scans, resulting in feature matrices of 14372 × 4096 for ABIDE_I and 16168 × 4096 for ABIDE_II. Our model uses ant colony optimization (ACO) to select feature subsets for dimensionality reduction. Further, nine machine learning classifiers are used to categorize females with autism spectrum disorder (ASD) from females with control behavior. The KNN-based fineKNN (FKNN) classifier had 92.21% accuracy on ABIDE-I and 93.49% on ABIDE-II. This proves the effectiveness of our proposed model.