Modified Exigent Features Block in JAN Net for Analysing SPECT Scan Images to Diagnose Early-Stage Parkinson's...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Dopamine transporter (DaT) quantification in the midbrain serves as a biomarker for Parkinson's disease (PD).
• Single-photon emission computed tomography (SPECT) imaging is used to accurately measure dopamine levels.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN), named JAN Net, specifically designed for Parkinson's disease identification.
• To utilize Volume Rendering Image Slices (VRIS) from SPECT scans for enhanced PD detection.

Main Methods:

• The JAN Net employs a modified exigent feature (M-ExFeat) block with convolutional and additive layers to preserve striatal features.
• Different-sized convolutional layers (1x1, 3x3, 5x5) extract multi-level features, which are combined in an additive layer to improve neuron learnability.
• Network performance was evaluated using stride 1 and stride 2 configurations.

Main Results:

• The JAN Net achieved 100% training and validation accuracy with minimal loss for stride 2, using data from the Parkinson's Progression Markers Initiative (PPMI) database.
• The proposed architecture demonstrated superior performance compared to various deep learning models and machine learning techniques, including Extreme Learning Machines (ELM) and Artificial Neural Networks (ANN).

Conclusions:

• The developed JAN Net shows significant efficacy in identifying Parkinson's disease.
• This research offers a valuable tool for neurologists to potentially prevent neuronal impairment in PD patients.