Machine learning-based evaluation of cancer pain from functional brain images
Journal
Transactions of Japanese Society for Medical and Biological Engineering
Journal Volume
Annual59
Journal Issue
Proc
Date Issued
2021-01-01
Author(s)
Abstract
Pain management is one of the important treatments of cancer patients. This study aims to develop a classifier of chronic cancer pain patients from their brain metabolic activity measured by FDG-PET. We compared FDG-PET brain images of 74 painful and 29 painless cancer patients to clarify the brain activity specific to chronic pain of cancer. Using the detected brain activity pattern, we further developed a classifier that determines the presence or absence of chronic pain from PET brain images by machine learning methods. The painful cancer patients showed significantly increased activity in the amygdala, hippocampus, and decreased activity in the cingulate gyrus and precuneus (p < 0.001, uncorrected). The proposed classifier was able to identify patients with chronic pain with a sensitivity and specificity of 85.1% and 62.1%, respectively. Further research is required to improve the specificity by selecting better regions of interest and classification algorithms.
Subjects
cancer pain | evaluation | functional brain image | machine learning | PET
SDGs
Type
journal article