Site Editor

Leo I. Gordon, MD, FACP


ESMO 2022: Can Machine Learning–Based Model Predict DLBCL Subtypes From Whole-Slide Images?

By: Julia Fiederlein
Posted: Tuesday, September 13, 2022

A study conducted by Charlotte Syrykh, MD, of Centre Hospitalier Universitaire de Toulouse, France, and colleagues demonstrated the predictive power of a machine learning–based model to subtype diffuse large B-cell lymphomas (DLBCL) based on whole-slide images. Their findings, which were presented during the European Society for Medical Oncology (ESMO) Congress 2022 (Abstract 623MO), suggested such tools may be used to augment the analytic capacities of pathologists.

“In daily routine, cell of origin classification is replaced by immunochemistry stains using the Hans algorithm based on the expression of CD10, BCL6, and MUM1 proteins,” the investigators commented. “In addition, co-expression of BCL2 and MYC proteins is of prognostic value and defines the double-protein expressor subtypes, [which are] associated with worse prognosis. Fluorescent in situ hybridization is mandatory in the workup of DLBCL to detect MYC and BCL2 and/or BCL6 rearrangements.” Deep learning models may be applied when immunohistochemical staining or fluorescent in situ hybridization are unavailable.

The investigators trained a deep learning model to predict the cell of origin and MYC/BCL2 double-protein expressor status, presence of MYC rearrangements, and expression of BCL6, CD10, and MUM1 proteins in 565 whole-slide images stained with hematoxylin/eosin from the LYSA trial “GHEDI” data set. Model performance was assessed using several repetitions of stratified five-fold cross-validation.

The deep learning model achieved an area under the receiver operating characteristic curve of 0.624 for the germinal-center molecular subtype, 0.687 for double-protein expressor status, and 0.675 for MYC rearrangement. Based on Cox proportional hazard model analyses, predictions of double-protein expressor status (hazard ratio [HR] = 0.38; P = .016), MYC rearrangement (HR = 5.23; P < .001), and MUM1 protein expression (HR = 2.80; P = .027) were associated with worse overall survival outcomes.

Disclosure: For full disclosures of the study authors, visit

By continuing to browse this site you permit us and our partners to place identification cookies on your browser and agree to our use of cookies to identify you for marketing. Read our Privacy Policy to learn more.