Author ORCID Identifier

https://orcid.org/0000-0002-6378-360X

Date of Award

2-10-2024

Document Type

Thesis

School

School of Computing

Programme

Ph.D.-Doctoral of Philosophy

First Advisor

Dr.R.Muthaiah

Keywords

Brain Tumor Detection, Anisotropic Diffusion Filtering, Medical Image Segmentation, YOLOv5, YOLOv8, Computer Vision for Medical Imaging

Abstract

As the body's central control system, the human brain is susceptible to a wide variety of disorders, including tumors characterized by abnormal cell growth. It is imperative to detect these tumors as early as possible to plan effective treatment and improve patient outcomes. By using contemporary medical imaging methods, this research seeks to improve the accuracy and efficiency of brain tumor detection through the careful preprocessing and analysis of images, particularly Magnetic Resonance Imaging (MRI) [1]. To provide context for the subsequent research efforts, the challenges inherent in brain tumor detection are discussed comprehensively, including segmentation accuracy, small lesion detection, and variability in imaging data [2]. A sophisticated image processing algorithm tailored for detecting brain tumors is being developed and implemented. Pre-processing techniques such as noise reduction, intensity normalization, contrast enhancement, and spatial registration are meticulously applied to enhance medical images quality and consistency. As a result of these pre-processing steps, automated detection algorithms are optimized, artifacts are mitigated, and interpretation of the imaging data is improved. The main aim of this research is to formulate a framework for processing MRI images in order to detect brain tumors early and accurately.

Consequently, the development of automated computer-aided diagnosis systems aims to enhance and segment brain scans to locate tumor regions with greater accuracy. With the introduction of a dual-condition diffusion coefficient, this chapter introduces a new anisotropic diffusion filter capable of adapting better to both healthy and pathological tumor areas. It is a dual-condition diffusion coefficient that permits dynamic adaptation of the diffusion process within tumor regions to the local image characteristic in order to introduce flexibility into the anisotropic diffusion process. A selection of contrast enhancement with edges conserved for the most part within areas of tumors that have large gradient intensity and very complex structures; hence, it achieves proper segmentation. This line drawing is trusted for the tumor outline because this process, being diffusion, is sensitive to these complex features at boundaries that the targeted methodology satisfied.

A full pipeline that surpasses state-of-the-art approaches is achieved through the combination of anisotropic diffusion filter enhancements and morphology-driven segmentation. By demonstrating enhanced results, the integrated framework contributes to developing effective systems that assist radiologists in detecting tumors and intervening in time. It is critical to pay attention to the minute details present in MR brain images during diagnosing tumor classification problems, which are most often overlooked in existing tumor detection methods. It is also important to note that a limited amount of annotated ground truth data is available. Hence, it is proposed that a new version of YOLOv5 be developed to improve brain tumor detection efficiency in MRI [3]. Furthermore, a modified cuckoo search algorithm is employed to achieve faster convergence and more accurate results, allowing the optimization algorithm to optimize the performance of the proposed network. From its conventional methods, the modified Cuckoo Search Algorithm inherits some of the important advantages. Firstly, the highly enhanced search capability leads to the better convergence rate and overall solution accuracy.

Furthermore, some new modifications greatly enhance the algorithm efficiency in running, such as variable step optimization, optimized Levy flight patterns, and the fact that it's less prone to straying off due to local minima in strength toward global optima. The greatly improved exploration and exploitation abilities of the algorithm ensure better completeness and effectiveness of the search in problems terrain that is typical of high-dimensional and complex problem domains, as happens in medical image analysis. Several key metrics are examined as part of the segmentation accuracy assessment, including Mean Squared Error (MSE), Structural Similarity Index (SSIM), Feature Similarity Index (FSIM), Peak Signal-to-Noise Ratio (PSNR), and CPU Time. YOLOv8-MM framework is next presented to highlight its innovative features, including separating the Deep Convolutional Neural Network (DCNN) into two parts that enhance information transfer.

As part of this study, Modified Fuzzy C-means Clustering (MFCC) is also introduced as a technique for MRI image segmentation. The latter is representative of Pareto optimization, balancing the accuracy and smoothness of tumor-segmented areas against other competing objectives for increased segmentation performance. cluster centre refinement discusses modifications to the fuzzy C-means clustering method aimed at improving its performance with regard to intrinsic ambiguity in medical images, through minimization of an objective function by an iterative process of cluster center refinement. Also, the improvement here in Pareto optimisation is the best possible compromise between these goals. Thus, Pareto optimisation provides the means for better accuracy in brain tumor identification for MRI images by allowing more efficient and correct segmentation, particularly while outlining complex borders of tumours.

MFCC can provide a robust foundation for accurate tumor geometry identification through iterative and Pareto optimization [4]. This chapter concludes with a discussion of the experimental setup, dataset characteristics, and performance evaluation methods, as well as rigorous experiments demonstrating the proposed methodology's robustness in detecting, segmenting, and classifying brain tumors utilizing publicly available datasets. Hence, developing a computer vision model with suitable feature extraction provides accurate brain tumor detection from MRI images.

Download

Share

COinS