Using Deep Learning and Resting-State fMRI to Classify Chronic Pain Conditions

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Chronic pain is a complex, often under-diagnosed condition with limited effective treatments.
• Advancements in brain imaging and machine learning offer new avenues for diagnostic support.

Purpose of the Study:

• To evaluate machine learning models, including deep learning, for classifying chronic pain patients using resting-state fMRI data.
• To compare the efficacy of different connectivity measures and machine learning algorithms.

Main Methods:

• Resting-state functional magnetic resonance imaging (fMRI) data were acquired from 150 participants.
• Functional brain connectivity was computed, and machine learning models (Deep Learning, SVM, RFC) were trained and tested.
• Data preprocessing utilized the MSDL probabilistic atlas, and dynamic time warping (DTW) was employed as a connectivity measure.

Main Results:

• Deep learning models, particularly a convolutional neural network with MSDL preprocessing and DTW, achieved the highest classification performance.
• Balanced accuracy ranged from 69% to 86%, with an area under the ROC curve from 0.84 to 0.93.
• Dynamic time warping outperformed correlation as a connectivity measure.

Conclusions:

• Resting-state fMRI data, analyzed with advanced machine learning techniques, show potential as a biomarker for chronic pain.
• Deep learning models demonstrate superior performance compared to traditional machine learning algorithms for this diagnostic task.