Prognostic value of graph theory-based tissue architecture analysis in carcinomas of the tongue.

Several studies on oral squamous cell carcinomas (OSCC) suggest that the clinical value of traditional histologic grading is limited both by poor reproducibility and by low prognostic impact. However, the prognostic potential of a strictly quantitative and highly reproducible assessment of the tissue architecture in OSCC has not been evaluated. Using image analysis, in 193 cases of T1-2 (Stage I-II) OSCC we retrospectively investigated the prognostic impact of two graph theory-derived structural features: the average Delaunay Edge Length (DEL_av) and the average homogeneity of the Ulam Tree (ELH_av). Both structural features were derived from subgraphs of the Voronoi Diagram. The geometric centers of the cell nuclei were computed, generating a two-dimensional swarm of point-like seeds from which graphs could be constructed. The impact on survival of the computed values of ELH_av and DEL_av was estimated by the method of Kaplan and Meier, with relapse-free survival and overall survival as end-points. The prognostic values of DEL_av and ELH_av as computed for the invasive front, the superficial part of the carcinoma, the total carcinoma, and the normal-appearing oral mucosa were compared. For DEL_av, significant prognostic information was found in the invasive front (p < 0.001). No significant prognostic information was found in superficial part of the carcinoma (p = 0.34), in the carcinoma as a whole (p = 0.35), or in the normal-appearing mucosa (p = 0.27). For ELH_av, significant prognostic information was found in the invasive front (p = 0.01) and, surprisingly, in putatively normal mucosa (p = 0.03). No significant prognostic information was found in superficial parts of the carcinoma (p = 0.34) or in the total carcinoma (p = 0.11). In conclusion, strictly quantitative assessment of tissue architecture in the invasive front of OSCC yields highly prognostic information.