Related Experiment Video
Updated: Jun 3, 2025
08:23
A Multimodal Imaging- and Stimulation-based Method of Evaluating Connectivity-related Brain Excitability in Patients with Epilepsy
Published on: November 13, 2016
11.1K
Hybrid optimization enabled DenseNet for autism spectrum disorders using MRI image
Sakthi Ulaganathan1, Pon Harshavardhanan2, N V Ganapathi Raju3
1Department of Computational Intelligence, School of Computing, SRM Institute of Science and Technology, Kattankulathur, Chennai 603203, India.
Computational Biology and Chemistry
|January 10, 2025
Summary
This study introduces JSTO-DenseNet, a novel model for autism spectrum disorder (ASD) detection using brain MRI images. The model achieves high accuracy, improving early diagnosis and reducing complications associated with traditional methods.
Area of Science:
- Neuroscience
- Medical Imaging
- Artificial Intelligence
Background:
- Autism spectrum disorder (ASD) is a neurodevelopmental disorder impacting social interaction and communication.
- Current early detection methods for ASD face challenges including time constraints and limited diagnostic accessibility.
- Advanced computational models are needed to improve the accuracy and efficiency of ASD diagnosis.
Purpose of the Study:
- To develop and evaluate the JSTO-DenseNet model for accurate and efficient detection of Autism Spectrum Disorder (ASD) using brain Magnetic Resonance Imaging (MRI).
- To enhance the early diagnosis of ASD by overcoming limitations of existing detection techniques.
Main Methods:
- Image pre-processing techniques including Gaussian filtering for noise reduction and Region of Interest (ROI) extraction.
- Feature extraction using Grey Level Co-occurrence Matrix (GLCM) and texture features (LTP, LOOP, HOG) from MRI scans.
- Development of a novel Jaya Sewing Training Optimization (JSTO) algorithm, combining Jaya algorithm and Sewing Training-Based Optimization (STBO), for feature selection and DenseNet model training.
Main Results:
- The JSTO-DenseNet model demonstrated superior performance with an accuracy of 94.8%, True Positive Rate (TPR) of 90%, and True Negative Rate (TNR) of 90.5%.
- Achieved an un-weighted average recall (UAR) of 89.8% and the lowest False Negative Rate (FNR) of 86.7%.
- Outperformed traditional methods including Explainable Artificial Intelligence (XAI) and various deep learning approaches on the Abide 1 dataset.
Conclusions:
- The JSTO-DenseNet model presents a highly accurate and effective approach for the early detection of Autism Spectrum Disorder (ASD) from MRI data.
- The novel JSTO optimization algorithm significantly enhances the performance of the DenseNet model in ASD classification.
- This research offers a promising tool for improving diagnostic outcomes and accessibility for individuals with ASD.
