Akademik Veri Yönetim Sistemi
1st National Neuroimaging Congress & 2nd Joint German-Turkish Symposium on Human Neuroscience, Ankara, Türkiye, 7 - 09 Eylül 2023, ss.11-12, (Özet Bildiri)
Objective: Pain decoding using hemodynamic responses is anobjective but challenging approach due to the variable natureof hemodynamic response. Moreover, the effects of differentanalgesic conditions increase the complexity of this problem.Inthis study, we aimed to decode the intensity level of nocicep-tive stimuli under analgesic conditions by utilizing fNIRSderived hemodynamic responses and a deep transfer learningapproach.
Methods: A previously collected fNIRS dataset collected from14 healthy male volunteers was utilized. Each subject had twosite visits where they were orally administered with a morphineor a placebo pill. At each site visit, subjects had 4 fNIRS scanswhich were taken during a nociceptive stimuli protocol a)beforeand b) after 30,60,90 minutes of drug administration. 6 noxiousand 6 innocuous stimuli were given to left thumb. After data pre-processing, a deep learning model was trained on the pre-drugdataset to classify painful and non-painful stimuli. Then, theknowledge obtained in this model was then transferred to classi-fy post-drug dataset.
Results: Accuracy performance of the pre-drug model was 0.97.Accuracy of post morphine drug models were 0.91 after 30 min,0.90 after 60 min and 0.91, after 90 min. For placebo adminis-tration, they were found as 0.92 after 30 min, 0.92 after 60 min,0.91 after 90 min respectively. Statistical comparison of per-formance metrics showed that accuracy values were significantlyhigher in pre-drug models compared to post-morphine andpost-placebo models.
Conclusion: Our deep transfer learning approach showed thatknowledge obtained from a pre-drug model trained by usinghemodynamic responses can be used to decode pain level afterdrug administration. We demonstrate the potential of fNIRSderived signals for transferring information from a modeltrained with baseline data to models built for different clinical ordaily life conditions where collection of training data may not befeasible/practical to build novel ML or DL models.