Study presents lightweight AI prototype for prostate cancer risk assessment
Published in Artificial Intelligence, Journals, Tools.
Publication date: October 7, 2026.
A peer-reviewed study published in Scientific Reports on October 7, 2026 presents a multimodal AI prototype for post-diagnostic prostate cancer staging and advanced-stage risk assessment. The accepted manuscript is available ahead of final editorial production.
The framework combines numerical, categorical and text inputs. A comprehensive model classifies stages, while a smaller model assesses advanced-stage risk with fewer features.
Retrospective analysis using the PLCO screening-trial dataset produced an ROC-AUC of 0.95. Evaluation of a feature-restricted model on SEER data yielded an ROC-AUC of 0.86 with limited overlap between available features. These values measure discrimination, not the percentage of patients correctly diagnosed.
The minimal model occupies 9.51 MB and runs inference in under 50 milliseconds on a CPU. Low-confidence predictions can be flagged for expert review.
The authors describe a patient-facing prototype for future evaluation. Prospective home-use validation and improved clinical outcomes have not been demonstrated.
Source: Scientific Reports, DOI: 10.1038/s41598-026-75102-1.