1. A new approach for automatic classification of non-Hodgkin lymphoma using deep learning and classical learning methods on histopathological images.
- Author
-
Özgür, Emine and Saygılı, Ahmet
- Subjects
- *
MACHINE learning , *MANTLE cell lymphoma , *CHRONIC lymphocytic leukemia , *NON-Hodgkin's lymphoma , *FEATURE selection , *DEEP learning - Abstract
Lymph cancer, also known as lymphoma, refers to the uncontrolled proliferation of the body's defensive cells, resulting in their transformation into cancerous cells. Lymphoma belongs to the group of blood cancers and exhibits a higher incidence compared to other cancers within this category. Early and accurate diagnosis plays a crucial role in managing this disease. In this particular investigation, an expert support system was developed employing histopathological images of lymph cancer. The data set comprised images of various lymphomas, including chronic lymphocytic leukemia (CLL), follicular lymphoma (FL), and mantle cell lymphoma (MCL). The initial approach involved utilizing the GLCM method to extract features from these images, while subsequent approaches adopted transfer learning architectures. Additionally, principal component analysis was employed for feature selection and dimension reduction. For the classification stage, a combination of machine learning algorithms such as random forests, k-nearest neighbors (KNN), naive Bayes, and decision trees, as well as deep learning methods including VGG16, ResNet50, and DenseNet201 architectures were employed. The models were trained separately for double and triple classes, and their performance was evaluated. The highest accuracy values in binary classification are: 94% for CLL and FL, 92% for FL and MCL, and 82% for MCL and CLL. In triple classification, the highest accuracy rate is 82%. The lowest accuracy values in binary classification are: 52% for CLL and FL, 57% for FL and MCL, and 49% for MCL and CLL. The methods by which these accuracy values were obtained are stated in "Results" section. However, due to the difficulty in distinguishing between MCL- and CLL-type lymphomas, the classification accuracy for these lymphomas was lower compared to the other classifications. On the other hand, the classification accuracy for FL was higher, as it exhibits more distinctive features than the other two lymphomas. The lowest success in the study was achieved at 36% as a result of the use of the KNN algorithm in triple classification. Notably, the DenseNet201 method achieved the highest success in the study, accurately classifying FL and CLL with a 94% accuracy rate. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF