Non-Hodgkin Lymphoma Classification Using Multi-Scale Attention Mechanism with Convolutional Neural Networks

Non-Hodgkin Lymphoma (NHL) is a group of blood cancers that affect the lymphatic system, a critical component of the body's immune defense. It involves the classification of lymphatic cancers based on blood cells, primarily B-cells or T-cells. NHL classification specifically excludes other cancer types such as leukemias and solid tumors. However, accurate classification of NHL using histopathological images remains challenging due to high intra- and inter-class similarities in tissue structures and subtle variations in cell distribution and morphology, which reduce model accuracy. To overcome these challenges, a Multi-Scale Attention Mechanism-Convolutional Neural Network (MSAM-CNN) is proposed for precise NHL classification. The MSAM is embedded within the generator to effectively learn correlations between corrupted and uncorrupted regions in histopathological images. During preprocessing, image resizing, min-max normalization, and data augmentation are applied to enhance classifier performance. Additionally, the Grey Level Co-Occurrence Matrix (GLCM) is employed for feature extraction, enabling efficient differentiation of NHL images from the preprocessed data. The proposed model achieves an accuracy of 99.21%, significantly outperforming existing approaches such as DCGAN and SampEn.