Diagnostic performance of artificial intelligence in detecting bladder carcinoma.

backgroundAccurate diagnosis and staging of bladder carcinoma (BC), particularly distinguishing muscle-invasive (MIBC) from non-muscle-invasive (NMIBC) forms, are critical for treatment planning. While imaging aids diagnosis, its accuracy depends heavily on clinician expertise. Artificial intelligence (AI) offers potential to enhance diagnostic precision, but its comparative performance against clinicians remains unclear.

methodsWe conducted a systematic review and meta-analysis of studies published up to February 2025 that compared AI-based diagnostic models with clinicians in detecting and staging BC using CT, MRI, or ultrasound. Pooled sensitivity, specificity, likelihood ratios, diagnostic odds ratio (DOR), and area under the ROC curve (AUC) were calculated. Quality was assessed using the QUADAS-2 tool.

resultsTwenty-two studies involving 3176 patients were included. AI models showed higher pooled sensitivity and specificity (both 83%) compared to clinicians (78%). The pooled DOR and AUC were 24 and 0.89 for AI, versus 13 and 0.81 for clinicians. Subgroup analyses confirmed consistent AI superiority across imaging modalities and validation settings.

conclusionAI-based models outperform clinicians in diagnosing and staging bladder cancer, demonstrating higher accuracy and consistency across imaging types. These findings support the integration of AI into clinical workflows, though further external validation and standardization are needed for routine application.