کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Improving the Performance of ST-GCN on Multi-Site rs-fMRI Data Through Time Repetition Alignment

نویسندگان :

Mehrana Calagari¹ Hamidreza Hakimdavoodi²

1- Isfahan University of Technology 2- Isfahan University of Technology

کلمات کلیدی :

time-alignment،time repetition،rs-fMRI،ST-GCN،autism spectrum disorder،longitudinal data،Sinc interpolation

چکیده :

In recent years, the integration of rs-fMRI datasets from various centers has expanded dataset size, addressing its limitations and supporting the broader application of deep learning in brain analysis. However, this integration had a drawback: it introduced increased data complexity due to variations in imaging protocols at different sites. Overcoming the challenges arising from these complexities drove numerous research studies. In this work, we tackled the issue of varying time repetition (TR) values in rs-fMRI data by using Sinc interpolation to time-align the data in the ABIDE I dataset. We proposed and implemented a simplified variant of the Spatio-temporal graph convolutional neural network (ST-GCN) referred to as lightweight ST-GCN, a deep model that leverages the inherently graph-like structure of the brain. Our results show that time-aligning the samples and using lightweight ST-GCN improve accuracy by 4%, boosting performance compared to prior studies using ST-GCN models for autism spectrum disorder classification.