Classification and Morphological Analysis of DLBCL Subtypes in H\&E-Stained Slides

TL;DR

The proposed deep learning model demonstrates robust performance, achieving an average area under the curve (AUC) of $(87.4 \pm 5.7) during cross-validation and shows a high positive predictive value (PPV), highlighting its potential for clinical application, such as triaging for molecular testing.

Abstract

We address the challenge of automated classification of diffuse large B-cell lymphoma (DLBCL) into its two primary subtypes: activated B-cell-like (ABC) and germinal center B-cell-like (GCB). Accurate classification between these subtypes is essential for determining the appropriate therapeutic strategy, given their distinct molecular profiles and treatment responses. Our proposed deep learning model demonstrates robust performance, achieving an average area under the curve (AUC) of (87.4 pm 5.7)\% during cross-validation. It shows a high positive predictive value (PPV), highlighting its potential for clinical application, such as triaging for molecular testing. To gain biological insights, we performed an analysis of morphological features of ABC and GCB subtypes. We segmented cell nuclei using a pre-trained deep neural network and compared the statistics of geometric and color features for ABC and GCB. We found that the distributions of these features were not very different for the two subtypes, which suggests that the visual differences between them are more subtle. These results underscore the potential of our method to assist in more precise subtype classification and can contribute to improved treatment management and outcomes for patients of DLBCL.

Figures (7)

• Figure 1: Snapshot of H&E stained images from TMC Dataset: Top row are images of GCB subtype and bottom row, ABC subtypes of DLBCL respectively
• Figure 2: Patches are generated from H&E stained WSI with optimal preprocessing and fed to encoder (pretrained) to get the embedding.
• Figure 3: Schematic of the attention-based classification model. The features are processed through an attention block (NN), where importance weights are applied to each feature. The weighted features are then aggregated and passed to a classifier.
• Figure 4: The first row shows thumbnails of three WSIs from the ABC subtype, with corresponding heatmaps in the second row (red indicates high attention, blue indicates low attention). The third and fourth rows display the same for the GCB subtype.
• Figure 5: The top row displays the highly attended patches of the ABC class from multiple patients, followed by the second row, which shows their corresponding segmentation maps obtained from StarDist. The third row presents the highly attended patches of the GCB class from multiple patients, with the last row depicting the respective segmentation maps for these GCB patches