<![CDATA[More than 85.720 new cases and 21.000 fatalities from lymphoma were reported globally in 2021. This type of cancer can spread through the body using the lymphatic system and then enter the blood. Since lymphoma affects the lymphatic system, it can be hard to diagnose correctly because there are many different subtypes, such as Mantle Cell Lymphoma (MCL), Follicular Lymphoma (FL), and Chronic Lymphocytic Leukemia (CLL). The diagnostic complexity of lymphoma highlights the need for more accurate and reliable automated diagnostic methods. This research proposes a transfer learning approach employing pre-trained Convolutional Neural Network (CNN) models using DenseNet-201, Xception, and ResNet-50, for lymphoma subtype classification. The dataset consists of microscopic histopathology images from three lymphoma classes (MCL, FL, and CLL). Each image was resized and segmented into 24 non-overlapping patches, followed by Macenko stain normalization and data augmentation. Model performance was evaluated using a random sampling with a fixed random seed train–validation–test split, and validated using cross-validation method. The proposed approach achieved classification accuracies of 96.7% for DenseNet-201, 97.15% for Xception, and 96.3% for ResNet-50. These results indicate that deeper architectures with efficient feature reuse and depthwise separable convolutions improve the detection of subtle morphological differences among lymphoma subtypes. Despite limitations related to dataset size and external validation, the findings demonstrate the potential of transfer learning-based CNN models as decision-support tools for lymphoma diagnosis.]]>
Copyrights © 2026
