Hybrid parcellation mapping approach for the extraction of connectivity measures in autism spectrum disorder fMRI data
Kaushal Oza,Shree Udavant,Rushikesh Jyoti,Mamoon Rashid
DOI: https://doi.org/10.1007/s11042-024-19374-7
IF: 2.577
2024-05-23
Multimedia Tools and Applications
Abstract:A complicated neurodevelopmental illness known as Autism Spectrum Disorder (ASD) impacts behavior, speech, and social interaction. For early diagnosis and intervention, it is essential to classify people with ASD accurately. The hybrid Parcellation Mapping approach for the extraction of connectivity measures is used in this study to analyze the ABIDE I pcp dataset. The two popular feature extraction techniques used on 3 distinct Parcellations to decrease the dimensionality of the feature space and the processing time for the machine learning models are Fast Fourier Transform (FFT) and Principal Component Analysis (PCA). While FFT made extracting frequency-domain characteristics from rs-FMRI data easier, PCA made it possible to identify the connectivity matrices' most informative components. Then, 13 distinct machine learning models were trained on the extracted features of 871 subjects of the rs-fMRI data of the dataset, using four different connectivity measures to capture different elements of the interregional interactions. The results of the experimental analysis show that the Bagging Classifier, when used with all features, provided the best accuracy while requiring the most CPU resources. Following feature extraction, the bagging classifier was among the most accurate, with an accuracy rate of almost 80%. We were able to maintain a consistent and high accuracy with the help of the FFT and PCA optimal features, which were also quite quick to compute. After using our approach, we consistently achieved accuracy values of 0.80, 0.82, and 0.82 for Harvard–Oxford, BASC, and AAL respectively. While choosing minimal ROIs on each of the three parcellations. Our strategy delivers better accuracy while significantly cutting computing expenses by 70%. It outperforms existing methods and holds promise for uncovering brain response connectivity metrics in multi-subject fMRI data. In this study, we present a novel hybrid approach for ASD classification using rs-fMRI data. Our methodology combines parcellation mapping, FFT, and PCA in a novel and efficient way, paving the way for an innovative and efficient model. The article thoroughly evaluates a variety of machine learning models across various brain atlases and connectivity measures, demonstrating superior classification accuracies. Our plan of action is unique since requires a comprehensive approach to multiple factors, maintaining a thorough exploration of the complex ASD classification environment.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering