Development of a Long Short-Term Memory-Based Model for Early Detection of Prostatitis Using Synthetic Time-Series Clinical Data

Abstract

Prostatitis is a common urological condition affecting men and often remains undetected during its early stages because of subtle or overlapping clinical symptoms. This study proposes a deep learning-based framework for the early detection of prostatitis using synthetically generated longitudinal clinical data. A synthetic dataset comprising 120 patient records, including prostate-specific antigen (PSA) levels and other clinically relevant hematological parameters, was preprocessed and used to train a Long Short-Term Memory (LSTM) neural network. The proposed model achieved high predictive performance, with an accuracy of 98.5%, precision of 97.6%, recall of 98.2%, an F1-score of 97.9%, and an area under the receiver operating characteristic curve (AUC) of 0.992, indicating excellent discriminative capability. Performance evaluation using the confusion matrix and ROC curve further demonstrated the model's effectiveness in distinguishing prostatitis from non-prostatitis cases. The findings highlight the potential of LSTM-based models to support early clinical decision-making, particularly where access to large real-world datasets is limited. Furthermore, the study presents a scalable and privacy-preserving diagnostic framework that can be adapted to similar medical prediction tasks. Future research should validate the proposed model using real-world clinical datasets, including electronic health records (EHRs), to assess its generalizability and clinical applicability.