Artificial intelligence in urology: combining imaging and clinical data for improved prostate diagnostics

Recent advancements in computer-aided imaging have significantly improved the detection of prostate morphological alterations and cancer diagnosis. However, the focus on imaging has led to the neglect of other valuable patient-related data. This study explores the potential benefits of combining image data with diverse tabular data, including patient demographics and clinical findings, to maximize diagnostic accuracy. Our research utilized a dataset of 3800 biparametric MR studies from biopsy-naive patients and their respective demographic and clinical data including all available PSA values. We developed and evaluated fusion methodologies that integrate magnetic resonance imaging data with tabular data to assess both binary and multiclass prediction capabilities. Our best-performing fusion model achieved a ROC-AUC of 0.794 and a QWK of 0.464 in multiclass prediction of prostatitis and Gleason scores, significantly outperforming both image-only and tabular data-only approaches. These results demonstrate that the integration of MR imaging with clinical and demographic data significantly enhances predictive precision in prostate cancer diagnosis.