Diffuse Large B-Cell Lymphoma is a subtype of non-Hodgkin lymphoma that occurs worldwide; around 80 thousand new cases were recorded in 2022. The accurate diagnosis and subtyping of non-Hodgkin lymphoma pose significant challenges that necessitate expertise, extensive experience, and meticulous morphological analysis. To address these challenges, a study was developed to segment and classify the slides of diffuse large B-cell lymphoma. This study utilizes a dataset consisting of 108 images of H&E-stained slides, including MYC-positive, MYC-negative, and normal slides. The images are initially pre-processed using colour deconvolution to highlight haematoxylin-stained nuclei before they are segmented to detect a nuclear area using the watershed algorithm and extract a morphological feature of nuclei. Results indicate that area showed a significant difference (P < 0.05), highlighting variations in nuclear size among groups. In contrast, perimeter, diameter, and circularity showed no significant differences. Colour analysis revealed significant differences in the standard deviation of the L component in LAB space and all RGB standard deviations, while the mean LAB and RGB values showed no significant differences.
Predicting MYC gene rearrangements in diffuse large B-cell lymphoma (DLBCL) from Hematoxylin and Eosin (H&E) images can optimize diagnostic triaging. This is crucial because Fluorescence in Situ Hybridization (FISH) testing remains costly and time-consuming. We evaluated five optimized Vision Transformer (ViT) paradigms to benchmark digital prescreening: Basic ViT, T2T-ViT, CLIP-ViT, Swin Transformer, and Hierarchical ViT. Models were tested on 463,200 normalized tissue patches from 120 whole-slide images (WSIs). Architectures were assessed on predictive accuracy, Area Under the Receiver Operating Characteristic (AUROC), and inference throughput. Hierarchical ViT achieved the highest absolute classification accuracy through multi-scale feature modeling. Among end-to-end models, Swin Transformer offered a favorable AUROC-throughput trade-off. It operated nearly three times faster than Hierarchical ViT at 5.16 vs. 1.78 batches/s. Meanwhile, T2T-ViT achieved the fastest overall inference at 11.17 batches/s. Grad-CAM visualizations showed top models focused on morphologically relevant tumor regions. However, visual heatmaps do not confirm biological correctness. Overall, multi-scale ViT architectures effectively predict patch-level MYC status in DLBCL, with Swin Transformer balancing speed and performance. Because this study uses a single-center, patch-level dataset, multicenter and patient-level WSI validations remain necessary before establishing clinical utility.
Background/Objectives: Diffuse Large B-Cell Lymphoma (DLBCL) is the most common subtype of non-Hodgkin lymphoma and demands precise segmentation and classification of nuclei for effective diagnosis and disease severity assessment. This study aims to evaluate the performance of HoVerNet, a deep learning model, for nuclei segmentation and classification in CMYC-stained whole slide images and to assess its integration into a user-friendly diagnostic tool. Methods: A dataset of 122 CMYC-stained whole slide images (WSIs) was used. Pre-processing steps, including stain normalization and patch extraction, were applied to improve input consistency. HoVerNet, a multi-branch neural network, was used for both nuclei segmentation and classification, particularly focusing on its ability to manage overlapping nuclei and complex morphological variations. Model performance was validated using metrics such as accuracy, precision, recall, and F1 score. Additionally, a graphic user interface (GUI) was developed to incorporate automated segmentation, cell counting, and severity assessment functionalities. Results: HoVerNet achieved a validation accuracy of 82.5%, with a precision of 85.3%, recall of 82.6%, and an F1 score of 83.9%. The model showed powerful performance in differentiating overlapping and morphologically complex nuclei. The developed GUI enabled real-time visualization and diagnostic support, enhancing the efficiency and usability of DLBCL histopathological analysis. Conclusions: HoVerNet, combined with an integrated GUI, presents a promising approach for streamlining DLBCL diagnostics through accurate segmentation and real-time visualization. Future work will focus on incorporating Vision Transformers and additional staining protocols to improve generalizability and clinical utility.
Molecular stratification of diffuse large B-cell lymphoma (DLBCL) based on MYC rearrangement status (confirmed via fluorescence in situ hybridization) guides frontline therapeutic selection. Although Vision Transformers (ViTs) effectively model global tissue context, lacking localized inductive biases extends training convergence. We developed PyConv, an asymmetrical hybrid architecture engineered for the three-class histopathological classification of MYC-rearranged (\(MYC+\)) DLBCL, non-rearranged (\(MYC-\)) DLBCL, and reactive lymph node (RLN). PyConv combines local ConvFormer blocks with global Pyramid ViT blocks via Multi-Head Cross-Attention (MHCA) fusion. We evaluated the network on a patient-stratified cohort of 180 whole-slide images (69,480 testing patches) against 12 baseline backbones at both patch and patient levels. PyConv achieved a patch classification accuracy and macro F1-score of 0.9962 [95% CI: 0.9884–0.9991 via patient-level clustered bootstrapping] and a macro AUROC of 0.9999. Patient-level probability pooling yielded 100% classification concordance on the held-out test cohort (18/18 patients), showing no statistically significant discriminative loss compared to top-performing baselines (e.g., ConvFormer, \(p = 0.9866\)) while integrating global stromal context. PyConv converged by epoch 20, reducing training duration by 68.7% compared to pure Pyramid ViT. PyConv provides an accurate, computationally efficient benchmarking framework for automated lymphoma classification. Although high patient concordance demonstrates internal feasibility, the model serves as an exploratory proof of concept requiring multi-center and multi-scanner validation before clinical translation.