الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 166 |  |  |
| | | from | 166 | | |
Abstract
This thesis presents Cancer as a serious public health problem worldwide, presenting high mortality rates when not accurately diagnosed and treated and overloading both public and private health systems. Despite the considerable progress reached through diagnostic imaging technologies, the death rate is still high. using a microscope to examine histology pictures visually. The histological analysis is a time-consuming, highly specialized activity that is greatly reliant on the pathologists’ experience and directly impacted by elements like weariness and a decline in attention.
It begins with an introduction to deep learning as an efficient technology and why the need for deep learning in life, then it introduces the outline and the objectives of the research.
Next, the research discusses, the effectiveness of most traditional classification systems depends on the proper data representation, and a large portion of the efforts are put into feature engineering, a challenging and time-consuming technique that employs prior expert domain knowledge of the data to start creating useful features. by contributing to closing this gap.
Variations in the size and texture of melanoma make the classification procedure more complex in a computer-aided diagnostic (CAD) system for melanomas. The research’s first experiment proposes an innovative hybrid deep learning-based layer-fusion and neutrosophic-set technique for identifying skin lesions. The off-the-shelf networks are examined to categorize eight types of skin lesions using transfer learning on International Skin Imaging Collaboration (ISIC) -2019 skin lesion datasets. The top two networks which are GoogleNet and DarkNet achieved an accuracy of 77.41% and 82.42% respectively. The proposed method works in two successive stages: first, boosting the classification accuracy of the trained networks individually. A suggested feature fusion methodology is applied to enrich the extracted features’ descriptive power, which promotes the accuracy to 79.2% and 84.5%, respectively. The second stage explores how to combine these networks for further improvement. The Error Correcting Output Codes (ECOC) paradigm is utilized for constructing a set of well-trained true and false Support Vector Machine (SVM) classifiers via fused DarkNet and GoogleNet feature maps, respectively. The ECOC’s coding matrices are designed to train each true classifier and its opponent in a one-versus-other fashion. Consequently, contradictions between true and false classifiers in terms of their classification scores create an ambiguity zone quantified by the indeterminacy set. Recent neutrosophic techniques resolve this ambiguity to tilt the balance towards the correct skin cancer class. As a result, the classification score is raised to 85.74%, outperforming the recent proposals by an obvious step. The trained models alongside the proposed Single Valued Neutrosophic Sets (SVNSs) implementation will be publicly available for aiding relevant research fields.
Breast cancer is among the most prevalent cancers, and early detection is crucial to successful treatment. One of the most crucial phases of breast cancer treatment is a correct diagnosis. Numerous studies exist about breast cancer classification in the literature. However, analyzing the cancer dataset in the context of clusterability for unsupervised modeling is rare. The second experiment analyses pointedly the breast cancer dataset clusterability by applying the widely used c-means clustering algorithm and its evolved versions fuzzy and neutrosophic ones. An in-depth comparative study is conducted utilizing a set of quantitative and qualitative clustering efficiency metrics. The study’s outcomes divulge the presented neutrosophic c-means clustering superiority in segregating similar breast cancer instances into clusters.
Recent laws governing image processing and machine learning techniques enable the development of CAD (Computer-Aided Diagnosis) systems, which can help pathologists be more efficient, objective, and reliable when making diagnoses. Sadly, there are relatively few comprehensive, accessible, and documented histology image resources designed for CAD study. Machine learning systems must be developed and validated using annotated databases.
The research presents how the convolutional neural network is structured giving a detailed explanation of the system of the DarkNet and GoogleNet. It explains Fusion as an efficient technology used for enhancing the classification rate for cancer and how it can contribute to saving lives. Furthermore, using the clustering techniques to get a comparative result for the most superior one in detecting cancer for unsupervised learning.
Finally, the study presents a comparison of results between the introduced policies and the existing policies to assess the proposed model enhancement.
The thesis is divided into six chapters as listed below:
Chapter 1
This chapter introduces the thesis by presenting the research motivations, objectives, challenges, methodology, and thesis organization.
Chapter 2
The chapter gives a general overview of the terminology and techniques used in pathology to diagnose cancer. It also introduces techniques based on Neutrosophic technology and deep learning. The most recent studies and the results of deep learning in the classification of human cancer are also reviewed and presented.
Chapter 3
This chapter focuses on the proposed main hypothesis for detecting skin cancer applied to the ISIC 2019 dataset. It also includes a brief explanation of the data and measurements and details the integration of the characteristics in the images through processing with more than one type of known networks to select the best one, as well as the integration work within the deep characteristics of each network. Then the two best networks are combined, and the correct results are extracted using ECOC, and then the Neutrosophic system is applied to increase the accuracy of detection of the disease.
Chapter 4
This chapter explains the proposed second hypothesis for the detection and identification of breast cancer using comparative advantage of several classification methods. With a brief explanation of the methods used and mention of their measurements. Then the comparison was applied to confirm the preference of the Neutrosophic C means method in disease identification.
Chapter 5
This chapter explains the three experiments that were performed. First, the conclusions drawn on the ISIC 2019 dataset are detailed. Also, on the WDBC data, and it ends with a comparison between previous work and thesis hypotheses. These findings are also discussed to establish a baseline for moving forward with the research.
Chapter 6
This chapter ends the thesis by conclusions and the expected future work.
Key words:
Cancer, imaging methods, Deep learning, Convolutional neural network, Transfer learning, Fused Deep Features, Skin Cancer Classification, Multi-Support Vector Machine, Error Correcting Output Codes, Single Valued Neutrosophic Sets, Breast cancer dataset clusterability, Fuzzy c-means clustering, Neutrosophic c-means clustering, t-SNE, Silhouette coefficient.