Understanding the multiclass classification of lymphomas from simple descriptors

The Lymphoma treatment can be more effective when its type is correctly diagnosed. Several researches have focused on developing accurate classification of the disease from histological images. However, they often use complex features and black box models, making them difficult for specialists to understand. In this work, we use morphological and non-morphological descriptors extracted from cell nuclei for multi-class classification of non-Hodgkin lymphomas. Three ensemble methods were evaluated and the best one achieved an average accuracy 0.956 using linear regression. This performance is close to the state of the art, even our model using simpler descriptors. We also employ explainable artificial intelligence techniques to provide understandable explanations for model’s decisions.