Please use this identifier to cite or link to this item:
https://ptsldigital.ukm.my/jspui/handle/123456789/513272Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | Shahnorbanun Sharan, Assoc. Prof. Dr. | |
| dc.contributor.author | Dheeb Abdelkarim Albashish (P72753 ) | |
| dc.date.accessioned | 2023-10-16T04:35:10Z | - |
| dc.date.available | 2023-10-16T04:35:10Z | - |
| dc.date.issued | 2017-03-31 | |
| dc.identifier.other | ukmvital:96923 | |
| dc.identifier.uri | https://ptsldigital.ukm.my/jspui/handle/123456789/513272 | - |
| dc.description | In the prostate histopathology image analysis (PHIA) domain, prostate cancer diagnosis are obtained through tissue image grading among other steps. Although numerous classification approaches have been proposed in recent years, the problem remains open due to the difficulty of discriminating between grades 3 and 4 and the heterogeneity among similar grades associated with the variation among tissue components. Computer Aided Diagnosis (CAD) has attracted considerable attention from researchers. CAD systems require quantitative histological analysis, which is currently achieved through a single classifier model. Generally, this model is not suitable for multi-component features of the tissue images as it leads to overfitting in the feature vectors. Another fundamental process that plays a fundamental role in PHIA is feature selection (FS), especially when features have a high dimensionality or multiple categories, which poses a relevance issue and redundancies. These limitations hinder the efficiency of PHIA classification, thus the need for highly effective FS methods. Most of the available literature deals with these limitations by means of the filter, wrapper, or embedded approaches. It has been shown that embedded methods (EM) are more efficient and lead to the best feature subsets. Perhaps the most notorious of these methods is the Support Vector Machine-Recursive Feature Elimination (SVM-RFE) due to its robustness against data overfitting compared to other schemes. This being said, SVM-RFE has its own drawbacks including its disregard of redundancy and complementarity. In addition to feature selection, PHIA involves the identification of multiples classes including Benign and Grades 3 and 4, which in many instances are very similar and may be extremely difficult to distinguish through CAD systems. Traditionally, multi-class classifications are solved through multiple binary classifications with a one-vs-one (OVO) or one-vs-all (Ovall) approach. The main issue with these schemes is that they fail to consider the interaction between classes. Fortunately, a group of multiclass approaches is used to construct an ensemble learning, which provides a close to optimal classifying system for PCa grading whereby results are combined by means of a majority voting (MV) strategy. Although MV has been shown to perform well in most cases, its performance becomes poor when votes are tied. This study intends primarily to suggest a complete PHIA framework based on ensemble learning to tackle the heterogeneity of tissue images. A linear FS method based on SVM-RFE and absolute cosine filter (SVM-RFE(AC)) is proposed to handle the redundancy issue. A third contribution of this study is the SVM-RFE(RCR), which not only considers the relevance and redundancy of features in SVM-RFE, but also their complementarity with the class label. The study culminates in a multi-level learning architecture (MLA) proposed to address the three-class classification problem in PHIA along with a novel adaptive majority voting (AMV) combination strategy that breaks the ties in the ensemble learning. The proposed framework was tested on prostate histopathology image datasets and achieved an outstanding AUC of 93.59% for the Grade 3 vs. 4 task, outperforming previous methods. The proposed SVM-RFE(AC) and SVM-RFE(RCR) achieved superior performance compared to the existing methods with less feature dimensions with an AUC of 94.45%, and 95.3% for Grade 3 vs. 4, respectively. The proposed MLA based on AMV also achieved a performance superior to OVO and Ovall, with an accuracy of 91.25%. Experimental results reveal that the five proposed methods are strong in terms of preserving important and useful knowledge and contribute to simplifying and improving the PHIA grading process.,Certification of Master's/Doctoral Thesis" is not available | |
| dc.language.iso | may | |
| dc.publisher | UKM, Bangi | |
| dc.relation | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
| dc.rights | UKM | |
| dc.subject | Computer Aided Diagnosis | |
| dc.subject | Histopathology | |
| dc.subject | Support vector machines | |
| dc.title | Supervised embedded feature selection methods based on support vector machine for histopathology grading | |
| dc.type | Theses | |
| dc.format.pages | 273 | |
| dc.identifier.callno | Q325.5.A399 2017 3 tesis | |
| dc.identifier.barcode | 002819(2017) | |
| Appears in Collections: | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| ukmvital_96923+SOURCE1+SOURCE1.0.PDF Restricted Access | 420.21 kB | Adobe PDF |  View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.