SIAMESE VERIFICATION FRAMEWORK FOR AUTISM IDENTIFICATION DURING INFANCY USING CORTICAL PATH SIGNATURE FEATURES
Xin Zhang1,2, Xinyao Ding1, Zhengwang Wu2
1School of Electronic and Information Engineering, South China University of Technology, China.
Proceedings. IEEE International Symposium on Biomedical Imaging
|August 23, 2021
Summary
This study introduces a new method for identifying Autism Spectrum Disorder (ASD) using brain imaging. The Siamese verification model with path signature features achieved high accuracy in early ASD detection.
Area of Science:
- Neuroscience
- Medical Imaging
- Machine Learning
Background:
- Autism Spectrum Disorder (ASD) lacks biological diagnostic markers, necessitating advanced identification methods.
- Brain imaging data offers a promising avenue for objective ASD identification.
Purpose of the Study:
- To develop and evaluate a novel model for identifying ASD using cortical features from brain imaging at 6 and 12 months.
- To leverage a verification framework and path signature features for enhanced ASD detection accuracy and reliability.
Main Methods:
- A Siamese verification model was employed to compare cortical features of subjects.
- Path signature (PS) features were introduced to capture longitudinal brain development dynamics.
- The model was trained and tested using brain imaging data from infants at 6 and 12 months.
Main Results:
- The proposed Siamese verification model achieved 87% accuracy, 83% sensitivity, and 90% specificity.
- This performance surpasses existing state-of-the-art methods for ASD identification.
- The integration of path signature features significantly improved classification performance.
Conclusions:
- The Siamese verification model with path signature features demonstrates a highly effective approach for early ASD identification.
- This method offers a reliable and accurate tool for diagnosing ASD using neuroimaging data.
- The verification framework enhances model training by increasing data size and improving robustness.
