How Does Artificial Intelligence Redefine N:C Ratio Thresholds for Diagnosing High-Grade Urothelial Carcinoma?

Summary: Introduction: The Paris System (TPS) requires a nuclear to cytoplasmic (N:C) ratio of > 0.7 as a primary criterion for the cytologic diagnosis of high-grade urothelial carcinoma (HGUC). We evaluated the validity of this cut-off value using an artificial intelligence (AI)-assisted tool (AIxURO) for quantitative assessment of the N:C ratio and nuclear area of known carcinoma cases. Materials and Methods: Our retrospective study assessed 109 patients with biopsy-proven urothelial carcinoma within 6 months of concurrent urine cytology. Cytology slides were digitized as whole slide images (WSIs) for analysis with AIxURO, which quantifies the N:C ratio and nuclear area for each selected cell with some or all cytomorphological features of HGUC (Figure 1). Results: Eighty-four percent (92/109) of carcinoma cases were cytologically reported as suspicious for high-grade urothelial carcinoma (SHGUC) or HGUC (Table 1). AIxURO quantitative analysis for both suspicious (having all TPS criteria for HGUC, Table 2) and atypical cells (having some TPS criteria for HGUC, Table 3) showed fewer suspicious cells than atypical cells (mean value 160.0 vs. 929.6). However, suspicious cells had higher mean N:C ratios (0.66 vs. 0.58) and larger mean nuclear areas (107.2 μm^2 vs. 86.9 μm^2). The mean N:C ratio for suspicious cells across all biopsy-proven categories of carcinoma was 0.65-0.68. The nuclear area differed, with a 34.7% larger mean nuclear area in suspicious versus atypical cells in atypical urothelial cells (AUC), 13.9% in SHGUC, and 24.4% in HGUC categories. The largest mean nuclear areas for both suspicious cells (117.3 μm^2) and atypical cells (101.1 μm^2) were observed in biopsy cases of carcinoma in situ (CIS). Conclusions: A lower N:C ratio cut-off between 0.65-0.68 in addition to considering nuclear area as a criterion, may be more predictive of HGUC than the current >0.7 TPS algorithm.