Classification of autistic subjects employing modified volume local binary pattern (MVLBP) and stacked Autoencoder (SAE) on functional magnetic resonance imaging (fMRI)

• Published in: Multimedia tools and applications Vol. 84; no. 20; pp. 22161 - 22186
• Main Authors: M., Kaviya Elakkiya, Dejey
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2025; Springer Nature B.V
• Subjects: Accuracy; Artificial intelligence; Autism; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Feature extraction; Independent component analysis; Machine learning; Magnetic resonance imaging; Medical imaging; Multimedia Information Systems; Principal components analysis; Special Purpose and Application-Based Systems; Tensors; Track 2: Medical Applications of Multimedia; Deep learning; Autism; fMRI; Feature extraction; Image processing; Classification
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-19881-7

Autism Spectrum Disorder (ASD) or Autism is a developmental disorder that impairs the ability to communicate and interact. Screening of autism is strenuous because there are no medical tests for the detection of autism instantly. Accurate diagnosis of autism is still arduous. Neuroradiologists map the brain with functional Magnetic Resonance Imaging (fMRI) to locate the specific regions of crucial functions such as sensing, speaking and moving. The existing autism screening tools are based on the questionnaire and only few neuroradiologists are available to know about fMRI. The object of the proposed research is to automatically classify autistic subjects utilizing fMRI with high degree of accuracy. The proposed work involves pre-processing, feature extraction, feature selection, and classification modules. The raw fMRI data from ABIDE II database is pre-processed adopting SPM12 to remove the artifacts. For handling high dimensional fMRI data, the pre-processed fMRI data undergoes Group Principal Component Analysis (Group PCA) to reduce the dimension and Group Independent Component Analysis (Group ICA) to analyze the Group Independent Components (ICs) respectively. To extract the features from Group ICs, Modified Volume Local Binary Pattern (MVLBP) is applied. The extracted features are converted into fMRI tensor data. Unsupervised Deep Learning Model is exercised for the classification of Typical Controls (TC) or autistic subjects. The performance of the proposed system generates superior experimental results while comparing with existing approaches in terms of metrices namely accuracy, precision and recall.