Spine Surgery Uses of Artificial Learning and Machine Learning: A LDH Treatment

The study evaluates the efficacy of various conventional techniques and ML (machine learning) models in predicting patients' 1-year follow-up outcomes based on preoperative factors. The study used the DaneSpine to identify sufferers who met the inclusion criteria and underwent (LDH) Lumbar Disc Herniation surgery. The model's initial training consisted of 16 distinct features, such as presurgical and demographic measures based on patient self-reports. The criteria for inclusion in this study encompassed sufferers who underwent LDH (Lumbar Disc Herniation) surgical treatment, recognized through the DaneSpine (Danish national registry for spine surgery). The patients were divided into groups based on whether they achieved the least clinically significant variation for EuroQol, VAS Back, Oswestry Disability Index (ODI), VAS (Visual Analog Scale) Leg, and their capacity to resume work duties after a one-year follow-up period. A random splitting method was used to create three subsets from the data, comprising testing, validation, and training sets with a ratio of 15%, 35%, and 50%, respectively. To compare the performance of various models like decision trees, deep learning, random forest, support vector machines, and boosted tree models were trained, while LR and MARS models were employed. Model fitness was evaluated by examining the performance and AUC for the duration of validation. The study generated seven models, with classification errors ranging from a minimum of 1% to a maximum of 4% standard deviation over the validation folds. Both deep learning and MARS (Multivariate Adaptive Regression Splines) models consistently performed well. The study developed two conventional and five ML (Machine Learning) predictive models to predict improvement in patients with LDH (Lumbar Disc Herniation) at the 1-year follow-up. The results indicate that building an ensemble of models requires minimal effort and is an initial basis for additional model selection and optimization.