Multi-Focus Image Fusion Based on Convolution Neural Network for Parkinson's Disease Image Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurodegenerative Diseases

Background:

• Parkinson's disease (PD) diagnosis relies on early detection for effective treatment to halt progression.
• Computer-aided diagnostic (CAD) techniques are increasingly vital for PD diagnosis.
• Medical image fusion offers enhanced diagnostic value in various fields.

Purpose of the Study:

• To evaluate a deep learning-based multi-focus image fusion method for combining MRI and PET scans in PD diagnosis.
• To compare the diagnostic accuracy of single-modal MRI versus multi-modal fused images using various deep learning networks.

Main Methods:

• A multi-focus image fusion technique using deep convolutional neural networks was employed to integrate MRI and PET data.
• Four neural networks (Alexnet, Densenet, ResNeSt, Efficientnet) were utilized to classify both single-modal MRI and multi-modal fused datasets.
• Classification accuracy was assessed for each network on both dataset types.

Main Results:

• Multi-modal fusion datasets consistently yielded higher test accuracy rates across all four networks compared to single-modal MRI.
• Specific accuracy improvements were observed, with Densenet achieving 97.19% on the multi-modal dataset versus 87.76% on MRI.
• The deep learning multi-focus image fusion method demonstrated a clear enhancement in PD image classification accuracy.

Conclusions:

• Multi-modal image fusion, particularly using deep learning techniques, significantly improves the accuracy of Parkinson's disease classification.
• The integration of MRI and PET data through advanced fusion methods offers superior diagnostic potential over single-modality imaging.
• This approach holds promise for more accurate and earlier diagnosis of Parkinson's disease.