Abstract

Objectives: Epidural pulsed radiofrequency (ePRF) interventions are successfully used in the treatment of patients with cervical, thoracic, and lumbar pain but they can be associated with high radiation doses. The scope of this study was to evaluate the ability of an artificial

Materials and Methods For 46 patients treated with ePRF, the dose area product (DAP) and procedure times were retrospectively analyzed. Additional patient, symptom, and surgical characteristics were collected based on the surgery protocols. An ANN was constructed to predict the excessive radiation as compared to the mean DAP value.

Results: Twenty-one patients were male (45.7%) and 25 females (54.3%). Mean values and ranges: age (61.76±2.2; 29-86 years), duration (26.65±1.43; 12-53 minutes), and DAP (694.63±113.83; 130.71- 3,711.64 Gy/cm2). The resulted ANN contained 7 scaling neurons (inputs), 3 hidden neurons, and one probabilistic neuron (target). Important inputs for the acquired ANN were age, sex, diagnosis, side, and level of intervention. The experience of the surgeon and the duration of the surgery were not significant contributors in this ANN. The network exhibited a sensitivity of 1, a specificity of 0.43, an AUC (Area Under Curve, ROC chart) of 0.714.

Conclusion: It was possible to construct an ANN, which could predict if the radiation burden during an ePRF procedure would be higher than the average one. This could prove useful in optimizing the planning of ePRF procedures. Further studies with a larger series of patients are warranted.

Keywords: Artificial neural networks; Interventions; Prediction; Pulsed radiofrequency; Radiation risk.