High Histological Skeletonization As a Measure of Heterogeneity is Associated with Poor Overall Survival and Death Within 2 Years of Diffuse Large B-Cell Lymphoma

Background: Diffuse large B-cell lymphoma (DLBCL) is an aggressive lymphoma that is histologically and clinically heterogeneous. Skeletonization is a computer vision technique used in image description, segmentation, and pattern recognition. Objectives: This retrospective observational study analyzed the 2D-topological heterogeneity of DLBCL. Methods: A series of 153 patients, including 110 patients with DLBCL and 43 patients with reactive lymphoid tissue, were analyzed by skeleton analysis using hematoxylin and eosin (H&E) digitized images. The skeletonization variable was correlated with several clinicopathological characteristics of the patients, including cell of origin Hans’ classifier and Lymph2Cx assay, MYC and BCL2 rearrangement by FISH, immunohistochemistry of BCL2, CASP8, cCASP3, CD163, CDK6, cPARP, CSF1R, E2F1, ISY1, Ki69, LMO2, MDM2, MYC, MYO, TNFAIP8, PD-L1, and p53, and gene expression using a pancancer immune profiling panel. Results: In comparison with reactive tissue, DLBCL was characterized by lower skeletonization: 5665.20 ± 1012.82 vs. 6341.16 ± 548.23, respectively (p < 0.001). Within the DLBCL diagnostic category, high skeletonization correlated with poor overall survival (hazard risk = 2.5, p = 0.003). High skeletonization was also correlated with higher Epstein–Barr virus (EBV) EBER positivity, death within the first two years, high histological entropy, and lower CD5, E2F1, BCL2, ISY1, and TNFAIP8 protein levels (all p values < 0.05). Gene expression was available in a representative subset of 30 cases, including 18 cases with high skeletonization and 12 cases with low skeletonization. High skeletonization was characterized by upregulation of 166 immuno-oncology genes, such as CD274 (PD-L1), NFKB1A, CD68, CSF1R, CCR5, ITGAM, and TGFB1, and enrichment of IL6 JAK STAT3 signaling, inflammatory response, TNF, NFKB, KRAS, apoptosis, and macrophage function pathways by gene set enrichment analysis (GSEA). In multivariate COX regression analysis for overall survival, the prediction value of skeletonization was independent of Hans’ classifier, the International Prognostic Index (IPI), and EBER (p = 0.030, hazard risk = 2.2). Conclusions: DLBCL is characterized by lower skeletonization than reactive lymphoid tissue. In DLBCL, high skeletonization is associated with poor prognosis and enrichment of immuno-oncology markers.