Influence of Individual Differences in fMRI-Based Pain Prediction Models on Between-Individual Prediction Performance

Pilot trial testing the effects of exercise on chemotherapy-induced peripheral neurotoxicity (CIPN) and the interoceptive brain system

Purpose

Chemotherapy-induced peripheral neurotoxicity (CIPN) is a prevalent, dose-limiting, tough-to-treat toxicity involving numbness, tingling, and pain in the extremities with enigmatic pathophysiology. This randomized controlled pilot study explored the feasibility and preliminary efficacy of exercise during chemotherapy on CIPN and the role of the interoceptive brain system, which processes bodily sensations.

Methods

Nineteen patients (65±11 years old, 52% women; cancer type: breast, gastrointestinal, multiple myeloma) starting neurotoxic chemotherapy were randomized to 12 weeks of exercise (home-based, individually tailored, moderate intensity, progressive walking and resistance training) or active control (nutrition education). At pre-, mid-, and post-intervention, we assessed CIPN symptoms (primary clinical outcome: CIPN-20), CIPN signs (tactile sensitivity using monofilaments), and physical function (leg strength). At pre- and post-intervention, we used task-free ("resting") fMRI to assess functional connectivity in the interoceptive brain system, involving the salience and default mode networks.

Results

The study was feasible (74-89% complete data across measures) and acceptable (95% retention). We observed moderate/large beneficial effects of exercise on CIPN symptoms (CIPN-20, 0-100 scale: -7.9±5.7, effect size [ES]=-0.9 at mid-intervention; -4.8±7.3, -ES=0.5 at post-intervention), CIPN signs (ES=-1.0 and -0.1), and physical function (ES=0.4 and 0.3). Patients with worse CIPN after neurotoxic chemotherapy had lower functional connectivity within the default mode network (R2=40-60%) and higher functional connectivity within the salience network (R2=20-40%). Exercise tended to increase hypoconnectivity and decrease hyperconnectivity seen in CIPN (R2 = 12%).

Conclusion

Exercise during neurotoxic chemotherapy is feasible and may attenuate CIPN symptoms and signs, perhaps via changes in interoceptive brain circuitry. Future work should test for replication with larger samples. ClinicalTrials.gov identifier NCT03021174.

Recommendations for the Development of Socioeconomically-Situated and Clinically-Relevant Neuroimaging Models of Pain

Pain is a complex, multidimensional experience that emerges from interactions among sensory, affective, and cognitive processes in the brain. Neuroimaging allows us to identify these component processes and model how they combine to instantiate the pain experience. However, the clinical impact of pain neuroimaging models has been limited by inadequate population sampling - young healthy college students are not representative of chronic pain patients. The biopsychosocial approach to pain management situates a person's pain within the diverse socioeconomic environments they live in. To increase the clinical relevance of pain neuroimaging models, a three-fold biopsychosocial approach to neuroimaging biomarker development is recommended. The first level calls for the development of diagnostic biomarkers via the standard population-based (nomothetic) approach with an emphasis on diverse sampling. The second level calls for the development of treatment-relevant models via a constrained person-based (idiographic) approach tailored to unique individuals. The third level calls for the development of prevention-relevant models via a novel society-based (social epidemiologic) approach that combines survey and neuroimaging data to predict chronic pain risk based on one's socioeconomic conditions. The recommendations in this article address how we can leverage pain's complexity in service of the patient and society by modeling not just individuals and populations, but also the socioeconomic structures that shape any individual's expectations of threat, safety, and resource availability.

Characterization of Source-Localized EEG Activity During Sustained Deep-Tissue Pain

Musculoskeletal pain is a clinical condition that is characterized by ongoing pain and discomfort in the deep tissues such as muscle, bones, ligaments, nerves, and tendons. In the last decades, it was subject to extensive research due to its high prevalence. Still, a quantitative description of the electrical brain activity during musculoskeletal pain is lacking. This study aimed to characterize intracranial current source density (CSD) estimations during sustained deep-tissue experimental pain. Twenty-three healthy volunteers received three types of tonic stimuli for three minutes each: computer-controlled cuff pressure (1) below pain threshold (sustained deep-tissue no-pain, SDTnP), (2) above pain threshold (sustained deep-tissue pain, SDTP) and (3) vibrotactile stimulation (VT). The CSD in response to these stimuli was calculated in seven regions of interest (ROIs) likely involved in pain processing: contralateral anterior cingulate cortex, contralateral primary somatosensory cortex, bilateral anterior insula, contralateral dorsolateral prefrontal cortex, posterior parietal cortex and contralateral premotor cortex. Results showed that participants exhibited an overall increase in spectral power during SDTP in all seven ROIs compared to both SDTnP and VT, likely reflecting the differences in the salience of these stimuli. Moreover, we observed a difference is CSD due to the type of stimulus, likely reflecting somatosensory discrimination of stimulus intensity. These results describe the different contributions of neural oscillations within these brain regions in the processing of sustained deep-tissue pain.

Designing individual-specific and trial-specific models to accurately predict the intensity of nociceptive pain from single-trial fMRI responses

Using machine learning to predict the intensity of pain from fMRI has attracted rapidly increasing interests. However, due to remarkable inter- and intra-individual variabilities in pain responses, the performance of existing fMRI-based pain prediction models is far from satisfactory. The present study proposed a new approach which can design a prediction model specific to each individual or each experimental trial so that the specific model can achieve more accurate prediction of the intensity of nociceptive pain from single-trial fMRI responses. More precisely, the new approach uses a supervised k-means method on nociceptive-evoked fMRI responses to cluster individuals or trials into a set of subgroups, each of which has similar and consistent fMRI activation patterns. Then, for a new test individual/trial, the proposed approach chooses one subgroup of individuals/trials, which has the closest fMRI patterns to the test individual/trial, as training samples to train an individual-specific or a trial-specific pain prediction model. The new approach was tested on a nociceptive-evoked fMRI dataset and achieved significantly higher prediction accuracy than conventional non-specific models, which used all available training samples to train a model. The generalizability of the proposed approach is further validated by training specific models on one dataset and testing these models on an independent new dataset. This proposed individual-specific and trial-specific pain prediction approach has the potential to be used for the development of individualized and precise pain assessment tools in clinical practice.

An Autoencoder-based Approach to Predict Subjective Pain Perception from High-density Evoked EEG Potentials

Pain is a subjective experience and clinicians need to treat patients with accurate pain levels. EEG has emerged as a useful tool for objective pain assessment, but due to the low signal-to-noise ratio of pain-related EEG signals, the prediction accuracy of EEG-based pain prediction models is still unsatisfactory. In this paper, we proposed an autoencoder model based on convolutional neural networks for feature extraction of pain-related EEG signals. More precisely, we used EEGNet to build an autoencoder model to extract a small set of features from high-density pain-evoked EEG potentials and then establish a machine learning models to predict pain levels (high pain vs. low pain) from extracted features. Experimental results show that the new autoencoder-based approach can effectively identify pain-related features and can achieve better classification results than conventional methods.