Murtaza, Ghulam and Shuib, Liyana and Abdul Wahab, Ainuddin Wahid and Mujtaba, Ghulam and Raza, Ghulam (2020) Ensembled deep convolution neural network-based breast cancer classification with misclassification reduction algorithms. Multimedia Tools and Applications, 79 (25-26). pp. 18447-18479. ISSN 1380-7501, DOI https://doi.org/10.1007/s11042-020-08692-1.
Full text not available from this repository.Abstract
Breast cancer (BrC) is the leading cause of abnormal death in women. Mammograms and histopathology (Hp) biopsy images are generally recommended for early diagnosis of BrC because Hp image-based diagnosis enables doctors to make cancer diagnostic decisions more confidently than with mammograms. Several studies have used Hp images to classify BrC. However, the performance of classification models is compromised due to the higher misclassification rate. Therefore, this study aimed to develop a reliable, accurate, and computationally cost-effective ensembled BrC classification network (EBrC-Net) model with three misclassification algorithms to diagnose breast malignancy in early stages using Hp images. The proposed EBrC-Net model is based on the deep convolutional neural network approach. For experiments, the publicly available BreakHis dataset was used and split into training, validation, and testing sets. In addition, image augmentation was adopted for the training set only, and features were extracted through the well-trained EBrC-Net. Thereafter, the extracted features were further evaluated by six machine learning classifiers, of which two best performing classifiers (i.e., softmax and k-nearest neighbour kNN]) were selected on the basis of five performance metric evaluation results. Furthermore, three misclassification reduction (McR) algorithms were developed and implemented in cascaded manner to reduce the false predictions of the softmax and kNN classifiers. After the implementation of the McR algorithms, experiments showed that the kNN results were much better and reliable than the softmax. The proposed BrC classification model achieved accuracy, specificity, and sensitivity rates of 97.74%, 100%, and 97.01%, respectively. Moreover, the performance of proposed BrC classification model was compared with that of state-of-the-art baseline models. Findings showed that the proposed EBrC-Net classification model, coupled with the proposed McR algorithms, achieved the best results in comparison with the baseline classification models. The proposed EBrC-Net model and the McR algorithms are a reliable source for doctors aiming for second opinion in making early diagnostic decisions for BrC using Hp images.
| Item Type: | Article |
|---|---|
| Funders: | None |
| Uncontrolled Keywords: | Breast cancer; Image classification; Convolutional neural network; Deep learning; Transfer learning; Histopathology biopsy image |
| Subjects: | R Medicine > RB Pathology R Medicine > RC Internal medicine |
| Divisions: | Faculty of Computer Science & Information Technology > Department of Computer System & Technology |
| Depositing User: | Ms Zaharah Ramly |
| Date Deposited: | 14 Nov 2023 02:50 |
| Last Modified: | 03 Dec 2023 02:11 |
| URI: | http://eprints.um.edu.my/id/eprint/36590 |
Actions (login required)
 |
View Item |
