Basic Research - Neurourology & LUTS/BPH & Others(구연) (E-027) Feasibility of a machine learning-based diagnostic platform to evaluate male lower urinary tract disorder using simple uroflowmetry. Urology, Samsung Medical Center, Urology, Samsung Medical Center, Urology, Samsung Medical Center, Urology, Samsung Medical Center, Medical AI Research Center, Samsung Medical Center, Urology, Samsung Medical Center Seokhwan Bang, Minki Baek, Deok Hyun Han, Hwang Gyun Jeon, Baek Hwan Cho, Kyu-sung Lee Background: We aimed to utilize well-known neural network architectures to classify Detrusor underactivity (DUA) and Bladder outlet obstruction (BOO) symptoms, each as a separate binary classification task, using uroflowmetry graphs. Our main contribution is a proposal to exploit deep learning networks in the domain of urology, since we have not encountered any deep learning based research dedicated to DUA and/or BOO classification to date. Hyper-parameters have been tuned excessively to adjust to uroflowmetry graph data. Furthermore, we applied often-used Grad-CAM++ for visual explanations of studied networks. Proposed research outcome is intended to help urologists successfully diagnose patients with lower urinary tract symptoms (LUTS). / Objectives: To develop prediction model for bladder outlet obstruction (BOO) and detrusor underactivity (DUA) using simple uroflowmetry by artificial intelligence. / Materials and Methods: We performed a retrospective review of 4,835 male patients ≥ 40 years who underwent urodynamic study (UDS) at a single center. We excluded patients with disease or history of prior surgery that could affect lower urinary symptoms. A total of 1792 patients were included. We extracted a simple uroflowmetry graph automatically using ABBYY®. We applied Convolutional Neural Network (CNN) deep learning system to predict detrusor underactivity and bladder outlet obstruction. As an evaluation metric, a 5-fold cross validation mean value of Area Under the Receiver Operating Characteristics (AUROC) curve was chosen since this metric provides richer measure of classification performance than accuracy when it comes to binary classification. / Results and Conclusion:  Among the 1792 patients, 482 (27%) had BOO, and 893 (49.83%) had DUA. There were significant differences between BOO and non-BOO patients in UDS parameters except time to peak flow. In DUA and non-DUA patients, there were significant differences in all of the pressure-flow study parameters, except age and voiding volume. The AUROC scores of BOO and DUA, which were measured using 5-fold cross validation, were 71.73% and 73.35%, respectively. Our study suggests that one can differentiate BOO and DUA using simple uroflowmetry analysis and CNN.   keywords : male LUTS, AI, urodynamic study