harishdeepak / parallel-feature-fusion-network-for-autism-detection

Parallel Feature Fusion Network for Autism Detection

Overview

This repository contains the implementation of a deep learning image classifier designed to differentiate between Autism Spectrum Disorder (ASD) and Typical Development (TD). The classifier employs a dual-track feature fusion network architecture that combines a Swin Transformer with a customized Convolutional Neural Network (CNN), achieving a remarkable classification accuracy of 98.7%. Additionally, a Quantum Support Vector Machine (QSVM) is utilized for feature classification, operating on a quantum simulator with 16 qubits, and attaining a classification accuracy of 96%.

Architecture

1. Dual-Track Feature Fusion Network

• Swin Transformer: Excels in capturing long-range dependencies within images, facilitating a deeper understanding of the interrelations among different image components.
• Customized CNN: Specializes in extracting local features, contributing to improved classification performance by considering both local and global features.

2. Quantum Support Vector Machine (QSVM)

• Utilizes features extracted from the feature fusion network.
• Operates on a quantum simulator with 16 qubits.
• Achieves a classification accuracy of 96%.
• Leverages quantum computing to potentially accelerate predictions, reduce computational time, and enhance overall efficiency.

Results

• Feature Fusion Network: 98.7% classification accuracy.
• QSVM: 96% classification accuracy.

Prerequisites

Ensure you have the following software and libraries installed:

• Python 3.x
• PyTorch
• TensorFlow
• Qiskit (for quantum simulations)
• NumPy
• SciPy
• scikit-learn
• Matplotlib