Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza
Cancer disease is drastically increasing worldwide over the past few years. Among all types of cancers in women, breast cancer (BrC) is the main cause of abnormal deaths. For a confident diagnosis of BrC, histopathology (Hp) images are usually suggested by the doctors. BrC detection is a diagnostic...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/14492/ http://studentsrepo.um.edu.my/14492/1/Ghulam_Murtaza.pdf http://studentsrepo.um.edu.my/14492/2/Ghulam_Murtaza.pdf |
| _version_ | 1848774981897420800 |
|---|---|
| author | Ghulam , Murtaza |
| author_facet | Ghulam , Murtaza |
| author_sort | Ghulam , Murtaza |
| building | UM Research Repository |
| collection | Online Access |
| description | Cancer disease is drastically increasing worldwide over the past few years. Among all types of cancers in women, breast cancer (BrC) is the main cause of abnormal deaths. For a confident diagnosis of BrC, histopathology (Hp) images are usually suggested by the doctors. BrC detection is a diagnostic test for benign (non-cancerous) and malignant (cancerous) breast tumors (BrT). Once the BrT is diagnosed, then it needs to be classified for subtypes of benign and malignant to start specific treatment. Several studies developed BrC detection and classification models using Hp images. However, the existing models required high computational resources, long training time, and their performance is compromised due to a higher misclassification rate. Thus, this research is aimed to develop two models. First, the BrC detection model is developed to diagnose BrT basic types like benign and malignant. Second, the BrT classification model is developed to diagnose subtypes of benign and malignant tumors. To perform overall experiments, Hp images of the BreakHis dataset are utilized. BreakHis is a large and complex dataset (i.e., four subtypes of each benign and malignant BrTs) that publicly available. For BrC detection, an efficient and reliable model namely Ensemble BrC Detection Network (EBrC-Net) and three misclassification reduction (McR) algorithms are developed. The proposed EBrC-Net model is based on deep learning (DL) based approach. EBrC-Net architecture is designed to require less training time and computational resources like a normal desktop computer. The trained EBrC-Net is used to extract discriminative features. The extracted features are evaluated through six machine learning (ML) classifiers namely softmax, k-nearest neighbor (kNN), support vector machine, linear discriminant analysis, decision tree, and naive Bayes. Experimentally, it has been observed that kNN outperformed the rest of the five ML classifiers. Furthermore, three McR algorithms are developed and implemented in a cascaded manner to reduce the false predictions (i.e., misclassification) of the aforementioned six ML classifiers. The proposed BrC detection model for five folds of features achieved mean accuracy, sensitivity, and patient recognition rate by 97.78%, 97.28%, and 97.92% respectively. On the other hand, BrT classification is aimed to develop an efficient and reliable model namely Biopsy Microscopic Image Cancer Network (BMIC-Net) to classify Hp images into eight subtypes of BrT through a DL-based hierarchical classification approach. BMIC-Net model can be trained using less computational resources in less time. The trained BMIC-Net is used to extract discriminative features from Hp images. To reduce the misclassification, a feature selection algorithm (using information gain and principal component analysis schemes) is developed to elicit the most discriminative feature subset. Finally, the aforementioned six ML classifiers are analyzed to acquire the best performing classifier. The experimental results revealed that BMIC-Net outperformed for five folds of features by obtaining a mean accuracy of 95.33% for first-level hierarchical classifier and 94.70%, 92.53% for second-level hierarchical classifiers. Moreover, the performances of both BrC detection and BrT classification are compared with existing state-of-art baseline studies. Findings discovered that the proposed models are efficient (i.e., consume less computational resources and training time) and reliable (i.e., reduce misclassification to show better and unbiased results even using a complex dataset) in comparison with the existing SoA baseline studies. Thus, the proposed BrC detection and classification models can assist doctors to serve on the basis of the second opinion for early diagnosis of BrC.
|
| first_indexed | 2025-11-14T14:06:57Z |
| format | Thesis |
| id | um-14492 |
| institution | University Malaya |
| institution_category | Local University |
| last_indexed | 2025-11-14T14:06:57Z |
| publishDate | 2021 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | um-144922023-06-14T23:51:42Z Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza Ghulam , Murtaza QA75 Electronic computers. Computer science Cancer disease is drastically increasing worldwide over the past few years. Among all types of cancers in women, breast cancer (BrC) is the main cause of abnormal deaths. For a confident diagnosis of BrC, histopathology (Hp) images are usually suggested by the doctors. BrC detection is a diagnostic test for benign (non-cancerous) and malignant (cancerous) breast tumors (BrT). Once the BrT is diagnosed, then it needs to be classified for subtypes of benign and malignant to start specific treatment. Several studies developed BrC detection and classification models using Hp images. However, the existing models required high computational resources, long training time, and their performance is compromised due to a higher misclassification rate. Thus, this research is aimed to develop two models. First, the BrC detection model is developed to diagnose BrT basic types like benign and malignant. Second, the BrT classification model is developed to diagnose subtypes of benign and malignant tumors. To perform overall experiments, Hp images of the BreakHis dataset are utilized. BreakHis is a large and complex dataset (i.e., four subtypes of each benign and malignant BrTs) that publicly available. For BrC detection, an efficient and reliable model namely Ensemble BrC Detection Network (EBrC-Net) and three misclassification reduction (McR) algorithms are developed. The proposed EBrC-Net model is based on deep learning (DL) based approach. EBrC-Net architecture is designed to require less training time and computational resources like a normal desktop computer. The trained EBrC-Net is used to extract discriminative features. The extracted features are evaluated through six machine learning (ML) classifiers namely softmax, k-nearest neighbor (kNN), support vector machine, linear discriminant analysis, decision tree, and naive Bayes. Experimentally, it has been observed that kNN outperformed the rest of the five ML classifiers. Furthermore, three McR algorithms are developed and implemented in a cascaded manner to reduce the false predictions (i.e., misclassification) of the aforementioned six ML classifiers. The proposed BrC detection model for five folds of features achieved mean accuracy, sensitivity, and patient recognition rate by 97.78%, 97.28%, and 97.92% respectively. On the other hand, BrT classification is aimed to develop an efficient and reliable model namely Biopsy Microscopic Image Cancer Network (BMIC-Net) to classify Hp images into eight subtypes of BrT through a DL-based hierarchical classification approach. BMIC-Net model can be trained using less computational resources in less time. The trained BMIC-Net is used to extract discriminative features from Hp images. To reduce the misclassification, a feature selection algorithm (using information gain and principal component analysis schemes) is developed to elicit the most discriminative feature subset. Finally, the aforementioned six ML classifiers are analyzed to acquire the best performing classifier. The experimental results revealed that BMIC-Net outperformed for five folds of features by obtaining a mean accuracy of 95.33% for first-level hierarchical classifier and 94.70%, 92.53% for second-level hierarchical classifiers. Moreover, the performances of both BrC detection and BrT classification are compared with existing state-of-art baseline studies. Findings discovered that the proposed models are efficient (i.e., consume less computational resources and training time) and reliable (i.e., reduce misclassification to show better and unbiased results even using a complex dataset) in comparison with the existing SoA baseline studies. Thus, the proposed BrC detection and classification models can assist doctors to serve on the basis of the second opinion for early diagnosis of BrC. 2021-02 Thesis NonPeerReviewed application/pdf http://studentsrepo.um.edu.my/14492/1/Ghulam_Murtaza.pdf application/pdf http://studentsrepo.um.edu.my/14492/2/Ghulam_Murtaza.pdf Ghulam , Murtaza (2021) Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza. PhD thesis, Universiti Malaya. http://studentsrepo.um.edu.my/14492/ |
| spellingShingle | QA75 Electronic computers. Computer science Ghulam , Murtaza Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza |
| title | Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza |
| title_full | Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza |
| title_fullStr | Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza |
| title_full_unstemmed | Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza |
| title_short | Deep learning-based breast cancer detection and classification using histopathology images / Ghulam Murtaza |
| title_sort | deep learning-based breast cancer detection and classification using histopathology images / ghulam murtaza |
| topic | QA75 Electronic computers. Computer science |
| url | http://studentsrepo.um.edu.my/14492/ http://studentsrepo.um.edu.my/14492/1/Ghulam_Murtaza.pdf http://studentsrepo.um.edu.my/14492/2/Ghulam_Murtaza.pdf |