Neural Networks Modeling for Prediction of Required Resources for Personalized Endourologic Treatment of Urolithiasis

When scheduling surgeries for urolithiasis, the lack of information about the complexity of procedures and required instruments can lead to mismanagement, cancellations of elective surgeries and financial risk for the hospital. The aim of this study was to develop, train, and test prediction models for ureterorenoscopy. Routinely acquired Computer Tomography (CT) imaging data and patient data were used as data sources. Machine learning models were trained and tested to predict the need for laser lithotripsy and to forecast the expected duration of ureterorenoscopy on the bases of 474 patients over a period from May 2016 to December 2019. Negative predictive value for use of laser lithotripsy was 92%, and positive predictive value 91% before application of the reject option, increasing to 97% and 94% after application of the reject option. Similar results were found for duration of surgery at <30 min. This combined prediction is possible for 54% of patients. Factors influencing prediction of laser application and duration <30 min are age, sex, height, weight, Body Mass Index (BMI), stone size, stone volume, stone density, and presence of a ureteral stent. Neuronal networks for prediction help to identify patients with an operative time <30 min who did not require laser lithotripsy. Thus, surgical planning and resource allocation can be optimised to increase efficiency in the Operating Room (OR).

Automated summary

This study aimed to develop prediction models for ureterorenoscopy using machine learning techniques. The models were trained and tested using CT imaging data and patient information to predict the need for laser lithotripsy and the expected duration of the surgery. The predictions can help optimize surgical planning and resource allocation, improving efficiency in the operating room.