Bladder cancer remains one of the most challenging cancers to diagnose and monitor. Patients often undergo repeated invasive procedures, such as cystoscopy, which are uncomfortable, costly, and resource intensive. While urine cytology offers a non-invasive alternative, traditional methods have well-known limitations, especially in detecting low-grade disease and producing consistent, reliable results.
Rethinking the Workflow Before Adding AI
Rather than simply applying artificial intelligence to existing laboratory methods, the researchers took a “workflow-first” approach. They recognized that inconsistent specimen quality, particularly cell loss during processing, is one of the biggest barriers to accurate urine-based cancer detection.
The newly developed liquid ICC platform keeps cells in suspension throughout staining and preparation, preserving cellular integrity and morphology. This standardized process ensures that each specimen meets adequacy requirements and is suitable for both immunocytochemical staining and digital analysis.
Multi-Marker Testing for Greater Accuracy
The platform evaluates urine samples for the presence of urothelial carcinoma using three protein biomarkers commonly associated with bladder cancer:
- hTERT, a marker linked to cellular immortality
- CK17, associated with abnormal urothelial differentiation
- GATA-3, a nuclear transcription factor frequently expressed in urothelial tumors
Using these markers together proved more powerful than relying on any single marker alone. The study demonstrated that:
- When any one marker was positive, the test achieved 100% sensitivity, meaning no cancers were missed.
- When all three markers were positive, the test achieved 100% specificity, meaning no false positives occurred.
This flexibility allows clinicians to tailor interpretation depending on the clinical scenario, for example, prioritizing sensitivity during cancer screening or specificity during diagnostic confirmation.
Where Machine Learning Adds Value
Once slides are prepared and stained, they are digitally scanned and analyzed using a machine learning algorithm. The algorithm identifies suspicious cells based on both biomarker expression and cytomorphologic features, such as nuclear size and staining patterns. Importantly, the system does not replace pathologists. Instead, it highlights candidate cells for expert review, improving efficiency while maintaining diagnostic oversight.
This human–AI collaboration helps reduce variability, supports consistent interpretation, and has the potential to ease workload pressures in busy pathology laboratories.
Strong Performance Across Disease Stages
In a prospective study of 150 patients, the platform consistently produced high-quality, cell-rich specimens. It performed well across a broad range of bladder cancer stages, including low-grade, non–muscle-invasive disease—an area where traditional urine cytology often struggles.
While larger studies are still needed to refine real-world performance and confirm specificity in broader populations, the initial results are encouraging and suggest meaningful clinical impact.
Looking Ahead
This research highlights an important lesson in digital pathology: advanced analytics are only as effective as the specimens they analyze. By prioritizing specimen integrity and workflow standardization before introducing machine learning, the study offers a practical blueprint for developing clinically useful AI-driven diagnostics.
If validated in larger, multi-center trials, this approach could reduce reliance on invasive monitoring procedures, improve early cancer detection, and open the door to similar innovations in other fluid-based cytology applications. To access this article, check out the full Digital Pathology Special Issue of the Journal of Histotechnology. Members can sign into their dashboard and click Journal of Histotechnology to access for free.
This article was created with the assistance of Generative AI and has undergone editorial review before publishing.