Performance of hybrid artificial intelligence in determining candidacy for lumbar stenosis surgery

• DOI: 10.1007/s00586-022-07307-7
• Publication Date: 2022-07-08T08:13:19Z
• Authors: Raphael Mourad; Serhii Kolisnyk; Yurii Baiun; Alessandra Falk; Titenkov Yuriy; Frolov Valerii; Aleksey Kopeev; Olga Suldina; Andrey Pospelov; Jack Kim; Andrej Rusakov; Darren R. Lebl
• Keywords: Candidacy; CHAID

Abstract Purpose Lumbar spinal stenosis (LSS) is a condition affecting several hundreds of thousands adults in the United States each year and associated with significant economic burden. The current decision-making practice to determine surgical candidacy for LSS often subjective clinician specific. In this study, we hypothesize that performance artificial intelligence (AI) methods could prove comparable terms prediction accuracy panel spine experts. Methods We propose novel hybrid AI model which computes probability recommendations LSS, based on patient demographic factors, clinical symptom manifestations, MRI findings. combines random forest trained from medical vignette data reviewed by surgeons, an expert Bayesian network built peer-reviewed literature opinions multidisciplinary team surgery, rehabilitation medicine, interventional diagnostic radiology. Sets 400 100 vignettes surgeons were used training testing. Results demonstrated high predictive accuracy, root mean square error (RMSE) between predictions ground truth 0.0964, while average RMSE individual doctor's was 0.1940. For dichotomous classification, AUROC Cohen's kappa 0.9266 0.6298, corresponding metrics 0.8412 0.5659, respectively. Conclusions Our results suggest can be automate evaluation physicians.