539: Response predictors to non-invasive motor cortex stimulation and mirror therapy in phantom limb pain

Topic: Clinical Science
Subtopic: Pain category (ICD-11), Chronic neuropathic pain, Developing methodology/technology, Neuromodulation (e.g. TMS, TDCS, FUS, SCS, MCS, DBS, vagus nerve stimulation), Pain in Special Populations

Background and Aims: Phantom limb pain is a chronic pain condition with difficult management and a heterogeneous response. Non-invasive motor cortex stimulation and mirror therapy are potential treatments; however, the predictors of patient response remain unclear. This study aims to use machine learning (ML) algorithms to predict responses to non-invasive brain stimulation and mirror therapy in patients with phantom limb pain (PLP). This was a secondary analysis of a randomized, blinded, sham-controlled, 2 × 2 factorial trial. 112 participants with traumatic lower limb amputation were randomized into treatment groups (active or sham M1 tDCS combined with Mirror therapy).

Methods: The predicted outcome was a 50% improvement in PLP levels after treatment. Clinical and motor cortex excitability variables were included as predictors. An ML approach of feature engineering, data pre-processing, and model optimization was used to create the most accurate predictive model. We performed cross-validation (80% of the data for training and 20% for testing). We trained multiple ML classifiers. We used the AUC-ROC for model comparison. A feature-importance analysis was performed to identify the most discriminative features.

Results: The MLP was the most accurate classifier across all treatments (active and sham). In the test data, AUC was higher for predicting M1 tDCS response (0.71) than for mirror therapy response (0.59). The important features to predict response to M1 tDCS were anxiety levels and intracortical inhibition. On the other hand, mirror therapy response was predicted by phantom limb sensation intensity and age.

Conclusions: Artificial Intelligence is a potential tool to optimize non-invasive brain stimulation (NIBS) treatments. M1 cortical excitability biomarkers seem to predict tDCS responses, and phantom limb sensations seem to predict MT responses. There is a need for validation with new cohorts and larger sample sizes.

References: 1. Pacheco-Barrios K, Meng X, Fregni F. Neuromodulation techniques in phantom limb pain: a systematic review and meta-analysis. Pain Medicine. 2020 Oct;21(10):2310-22.
2. Pacheco-Barrios K, Gunduz ME, Bonin Pinto C, Duarte D, Vélez FG, Gianlorenco AC, Teixeira PE, Giannoni-Luza S, Crandell D, Battistella LR, Simis M. Effects of combined and alone transcranial motor cortex stimulation and mirror therapy in phantom limb pain: a randomized factorial trial. Neurorehabilitation and neural repair. 2021 Aug;35(8):704-16.

Presenting Author: Kevin Pacheco-Barrios
Submitting Author: Kevin Pacheco-Barrios

Senior Author: Felipe Fregni

IASP 2026 Global Year theme of Neuropathic Pain: Yes

• Carla Pastora Sesin: MD - Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA, USA.
• Jisu Park: BS - Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA, USA.
• Joao Victor Ribeiro: MD - Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA, USA.
• Felipe Fregni: MD, PhD - Spaulding Rehabilitation Hospital, Harvard Medical School, Boston, MA, USA.