An Isolated CNN Architecture for Classification of Finger-Tapping Tasks Using Initial Dip Images: A Functional...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique.
• Analyzing hemodynamic responses (HR) is crucial for understanding brain activity.
• Classifying subtle motor tasks requires advanced analytical methods.

Purpose of the Study:

• To investigate the classification of finger-tapping tasks using fNIRS data.
• To evaluate the effectiveness of different convolutional neural network (CNN) architectures.
• To determine the optimal duration for capturing hemodynamic responses for accurate classification.

Main Methods:

• Utilized functional near-infrared spectroscopy (fNIRS) to capture brain activity during finger-tapping tasks.
• Developed and tested isolated convolutional neural network (CNN) models with varying layers (16, 19, 22, 25).
• Constructed functional t-maps based on the initial dip duration of hemodynamic responses (0.5 to 4s).

Main Results:

• The 22-layered isolated CNN model achieved the highest classification accuracy of 89.2%.
• Classification using initial dip hemodynamic responses proved more effective than delayed responses.
• Distinct brain activity patterns were identified for thumb and little finger tapping tasks.

Conclusions:

• The 22-layered CNN model offers an efficient approach for classifying fNIRS data from motor tasks.
• Initial hemodynamic dip features are highly discriminative for differentiating subtle brain activities.
• fNIRS-based analysis of initial hemodynamic responses holds potential for diagnosing cerebral oxygen exchange abnormalities.