Photo Credit: iStock.com/Marco Marca

AI enhances urothelial carcinoma diagnosis and management, but only as an assistive tool, not a replacement, for cytopathologists, according to a recent review.

Artificial intelligence (AI) holds promise to enhance the precision and speed of urothelial carcinoma (UC) diagnosis and management. Yet, it is poised to function as an assistive tool, not a replacement, for cytopathologists, according to a scoping review article published in the World Journal of Urology.

“AI algorithms excel in tasks such as image analysis and pattern recognition with speed and precision that can greatly enhance the efficiency of diagnostic workflows,” observed co-corresponding author Ee Jean Lim, MBBS, MMed, MPH, of Singapore General Hospital, and colleagues. “However, cytopathologists provide critical expertise in integrating clinical context, interpreting complex cases, and making nuanced decisions beyond the capabilities of automated algorithms.”

Enhancing Diagnostic Workflows

The review synthesized findings from 16 studies evaluating AI applications in urine cytology. Automated identification and characterization of atypical cells emerged as the primary use case, with deep learning models consistently achieving areas under the receiver operating characteristic curve (AUC) up to 0.989.

Another main application of AI in urine cytology is risk stratification of abnormal cells in urine samples. One AI-based model cited in the review classified cytology images as negative (benign) or positive (atypical or malignant) with over 90% accuracy across diverse cell types. “Such a computerized system can be used to identify cells that are likely atypical or malignant, potentially reducing the workload of cytopathologists and ensuring more consistent classifications,” the authors emphasized.

Advances in Risk Stratification & Recurrence Prediction

Beyond detection, AI has been leveraged to stratify risk and predict histologic outcomes. One AI system discussed in the review demonstrated an AUC of 0.78 and 63% sensitivity for high-grade UC histology prediction, surpassing pathologist cytology sensitivity of 46%, albeit with comparable specificity and overall accuracy. The AI assessments required an average of 139 seconds per sample, suggesting a feasible integration into clinical throughput. In another study, predictors of bladder cancer recurrence identified by machine learning provided more comprehensive insights than manual cytological and histological examinations alone.

Challenges & Future Direction

Despite these advances, the authors emphasize the need for larger, multicenter datasets and rigorous external validation to ensure algorithmic robustness across diverse populations and laboratory practices. They also stressed that prospective trials must evaluate the optimal integration of AI into existing surveillance protocols to maximize patient outcomes and workflow efficiency. “Nonetheless,” the authors concluded, “the integration of AI holds immense potential to revolutionize the diagnosis and management of UC,” provided that collaborative efforts address current limitations and ethical considerations in algorithm deployment.