Neural decoding of Aristotle tactile illusion using deep learning-based fMRI classification.

Area of Science:

• Neuroscience
• Cognitive Science
• Neuroimaging

Background:

• The Aristotle illusion, where one object is perceived as two, is a known tactile phenomenon.
• Previous electroencephalography (EEG) studies were limited by low spatial resolution in investigating its neural basis.
• Functional magnetic resonance imaging (fMRI) offers higher spatial resolution for exploring brain activity.

Purpose of the Study:

• To identify specific brain regions associated with the Aristotle illusion using fMRI.
• To apply deep learning techniques to analyze fMRI data for classifying tactile perception.
• To investigate the neural correlates of tactile illusions with enhanced spatial resolution.

Main Methods:

• Collected fMRI data from 30 participants experiencing Aristotle, Reverse, and Asynchronous tactile stimuli.
• Trained four convolutional neural network (CNN) models for perception-based and stimulus-based classification tasks.
• Utilized Gradient-weighted Class Activation Mapping (Grad-CAM) for region-of-interest (ROI) analysis.

Main Results:

• A Simple Fully Convolution Network (SFCN) achieved 68.4% accuracy in classifying Aristotle vs. Reverse illusion perception.
• CNN models achieved 80.1% accuracy in classifying the occurrence versus absence of Reverse illusion.
• Stimulus-based classification accuracies were around 50%, indicating difficulty in distinguishing between stimuli types.
• Grad-CAM analysis identified the somatosensory cortex and parietal regions as salient for perception-based classification.

Conclusions:

• Perception-driven neural responses related to tactile illusions are classifiable using fMRI data and CNN models.
• Saliency mapping confirms the involvement of the somatosensory cortex and parietal regions in tactile illusion perception.
• Additional brain regions, including the orbitofrontal cortex and supplementary motor area, were also identified as salient.