ANN Model Predicts Post-MVD Pain Among Patients With Trigeminal Neuralgia

Tinnitus. Closeup up side profile sick female having ear pain touching her painful head. Concept photo with indicating location of the pain. Health care concept
Using a constructed ANN model, researchers sought to predict the long-term outcomes of microvascular decompression surgery and the factors that influence prognosis in patients with trigeminal neuralgia.

An artificial neural network (ANN) model could predict long-term pain prognosis among patients with trigeminal neuralgia (TN) who underwent microvascular decompression (MVD), according to study findings published in World Neurosurgery.

MVD is considered the main surgical method to treat patients with TN. However, clinical practice has noted patients with TN don’t always report favorable outcomes and can even experience a reoccurrence following MVD. The use of traditional empirical prognostic judgments or standard linear regression models have fallen short of predicting favorable outcomes in this patient population. The objective of the current study was to predict disease prognosis for surgical procedures using an ANN model in patients with TN after MVD.

Patients (N=1041) with TN who received MVD surgery at the Hangzhou First People’s Hospital in China between 2013 and 2018 were included in this analysis. Epidemiological features were used as input in an ANN model to predict long-term pain outcomes. Patients were randomly subdivided into training (n=833) and validation (n=208) cohorts.

Patients were 58.7% women, aged mean 53.6 (range, 38-83) years, duration of disease was 4.33±6.5 years, 52.7% had facial pain on the left side of the face, and 51.6% had pain in second and third division of the trigeminal nerve.

During the operation, 8.5% were found to have noncorresponding pain and neurovascular compression locations, 8.4% had both offending arteries and veins, 1.4% had no offending vessels, and 31.8% had grade 3 vessel compression.

The day after surgery, 90.9% were free from pain. The most common short-term complications included headache (23.5%), fever (9.4%), and facial numbness (8.6%).

At the final 2-year follow-up, Barrow Neurological Institute (BNI) pain score categories were 1 (86.0%), 2 (4.4%), 3 (4.1%), 4 (3.7%), and 5 (1.3%).

Among the 15 input factors, the most important for predicting postsurgical pain were whether the location of pain corresponded with offending vessel (area under the curve [AUC], 0.705; accuracy, 74.5%), immediate pain remission after surgery (AUC, 0.761; accuracy, 78.6%), degree of nerve compression (AUC, 0.793; accuracy, 87.2%), type of offending vessels (AUC, 0.810; accuracy, 90.1%), and disease duration (AUC, 0.831; accuracy, 93.3%).

The final model predicted 179 true positives, 19 true negatives, 6 false positives, and 4 false negatives, with an AUC of 0.862.

This study may have been limited by the small proportion of patients who experienced persistent pain, limiting power.

In this study, an ANN was found to be feasible for predicting long-term pain outcomes among patients with TN undergoing MVD. “The model was, furthermore, able to assess the importance of each factor in the prediction of pain outcome,” the researchers concluded. This tool should be confirmed using data from multiple centers.


Hao W, Cong C, Ding W, et al. Multi-data analysis based on an artificial neural network model for long term pain outcome and key predictors of microvascular decompression in trigeminal neuralgia. World Neurosurg. Published online April 28, 2022. doi:10.1016/j.wneu.2022.04.089