Classification of lymph cancer images with machine learning methods

Lymph cancer, or lymphoma, is the uncontrolled proliferation of the body's defense cells by becoming cancerous. Lymphoma is in the group of blood cancers and it is frequency of the disease is higher than other cancers in this group. Early and correct diagnosis is very important in this disease. Lymphoma, a dangerous and serious type of cancer, can result in death depending on the situation. Specialist support systems are being developed to help specialist early and correct diagnosis. Specialist support systems in the healthcare field provide classification of the data of a certain disease using machine learning and deep learning methods. When the studies in the literature are examined, wonderwork of specialist support systems are seen. In this study, a dataset containing histopathological images of lymph cancer was studied. The dataset includes images of lymphomas of the CLL, FL and MCL types. In order to increase the dataset, rotation at 90, 180, 270 and 360 degree angles was performed. In order to extract features from these images, GLCM was used in the first approach and transfer learning method was used in other approaches. PCA was used for feature selection and dimension reduction. At the classification stage; machine learning algorithms RF, KNN, NB, DT and deep learning architectures VGG16, ResNet50 and DenseNet201 were used. Model trainings were performed separately for double and triple classes and the results were examined. It is difficult to distinguish between MCL and CLL lymphomas. Therefore the classification results of this lymphoma are lower than the other classification results. The results of the classification with FL are high because FL can be distinguished from the other two lymphomas. DenseNet201 was over performed in classifying FL and CLL with 94% accuracy rate.