Ensemble Transfer Learning for Lymphoma Classification
Marija Habijan, Irena Galić · 2024
The classification of lymphoma types using deep learning models presents a promising avenue for enhancing diagnostic accuracy in medical imaging. This study evaluates the effectiveness of multiple pre-trained convolutional neural networks (CNNs), namely VGG-19, DenseNet201, MobileNetV3, and ResNet50V2, in classifying three common types of lymphoma: chronic lymphocytic leukemia (CLL), follicular lymphoma (FL), and mantle cell lymphoma (MCL). We tailored each model via transfer learning to adapt to the specific task of lymphoma classification. Our results indicate that DenseNet201 achieved the highest accuracy with 98.04%, followed by ResNet50V2, MobileNetV3, and VGG-19 with accuracies of 90.13%, 89.07%, and 87.11% respectively. Additionally, an ensemble approach combining all four models demonstrated a significant performance improvement, achieving an accuracy of 98.89%. These findings underscore the potential of advanced CNN architectures and ensemble methods in improving the diagnostic processes for lymphoma through medical imaging, offering a robust tool for clinical support and a pathway toward automated diagnostic systems.