الفهرس 
Chapter 1 : IntroductionOnly 14 pages are availabe for public view
 |  | from | 113 |  |  |
| | | from | 113 | | |
Abstract
Thesis objectives
The main idea of this thesis concentrates on using medical imaging for disease diagnosis and manipulating images to generate effective framework that help in early detection and treatment and reduce the risk of diseases. For segmentation, we proposed an accurate method with a multilevel thresholding technique. Therefore, to find the optimal thresholding values for segmenting images, we improved the original version of GJO with the OBL called IGJO and the Otsu method. Then, for the classification process, we build a DCNN from scratch to classify dermatoscopic datasets.
The objectives of this thesis encompass the following:
• Studying the impact of soft computing techniques on medical imaging segmentation and classification.
• Highlighting most promising solutions to be improved for skin cancer segmentation and classification.
• Presenting a novel contribution in this theme for more accurate results in skin cancer diagnosis.
• Proving the performance of the proposed model using real datasets.
Methodology
Creating a framework for segmenting and classifying medical images that has the ability to detect diseases early and identify magic. The image segmentation problem must be solved using MT because the results of the unwanted edges are affected by dependency on the classification results. Therefore, using improved domain-specific algorithms can solve the creative problem. Whatever medical data the research contains should be analyzed to provide promising results for classification and diagnosis. CNN is considered one of the best techniques for image classification that can be used to provide the best solutions for diagnosis.
Thesis contributions
The primary contributions of this thesis can be summarized briefly as follows:
1- A comprehensive review is introduced that reflects on the soft computing techniques used for the
segmentation and classification of medical imaging as well as an overview of the development of
this field. It gave an overview of the various image segmentation and classification methods and
shed light on related challenges in using different medical imaging to diagnose diseases. Moreover, it concentrates this review on some of the methods, such as multilevel thresholding image segmentation and classification using CNNs.
2- An enhanced version of the meta-heuristics GJO algorithm, named IGJO, is proposed. The improvement has been made to the original GJO algorithm by the OBL mechanism.
3- The IGJO is applied to solve the image segmentation problem using multilevel thresholding using Otsu as an objective function.
4-The proposed segmentation algorithm was applied to dermatoscopic image datasets related to skin cancer. Its versatility extends to various medical image types and standard images. The algorithm’s performance was systematically evaluated across different segmentation levels to gauge its effectiveness.
5- An efficient classification model using a DCNN to diagnose skin cancer is proposed. The proposed model was evaluated on two dermatoscopic datasets: HAM10000 and ISIC-2019.
Thesis recommendations
1-The proposed techniques could be additionally tested on different engineering and real-world problems with unknown search spaces.
2-The framework undergoes comparison with dermatologists to assess real-world effectiveness, offering insights for healthcare decisions and potential integration into clinical practices, serving as a reliable second opinion, and substituting human-involved processes.
3-Improved meta-heuristic algorithms will be used for hyperparameter optimization with DCNN architecture.
4-Segmentation methods will be used with DCNN architecture to further improve the classification accuracy.
5-The proposed model could be further evaluated to address other challenges present in medical imaging, such as variations in image quality, lighting, and artifacts.