DLBCL-Morph: Morphological features computed using deep learning for an annotated digital DLBCL image set

Diffuse Large B-Cell Lymphoma (DLBCL) is the most common non-Hodgkin\nlymphoma. Though histologically DLBCL shows varying morphologies, no\nmorphologic features have been consistently demonstrated to correlate with\nprognosis. We present a morphologic analysis of histology sections from 209\nDLBCL cases with associated clinical and cytogenetic data. Duplicate tissue\ncore sections were arranged in tissue microarrays (TMAs), and replicate\nsections were stained with H&E and immunohistochemical stains for CD10, BCL6,\nMUM1, BCL2, and MYC. The TMAs are accompanied by pathologist-annotated\nregions-of-interest (ROIs) that identify areas of tissue representative of\nDLBCL. We used a deep learning model to segment all tumor nuclei in the ROIs,\nand computed several geometric features for each segmented nucleus. We fit a\nCox proportional hazards model to demonstrate the utility of these geometric\nfeatures in predicting survival outcome, and found that it achieved a C-index\n(95% CI) of 0.635 (0.574,0.691). Our finding suggests that geometric features\ncomputed from tumor nuclei are of prognostic importance, and should be\nvalidated in prospective studies.\n