This master’s thesis investigates the application of texture analysis techniques to prostate ultrasound images, with a specific focus on how different pre-processing schemes affect the analysis outcomes. Conducted at the Cyprus University of Technology, the research addresses a critical need in medical imaging: improving the diagnostic accuracy and reliability of prostate cancer detection through advanced image processing methods. The study is supervised by Christos P. Loizou and is situated within the Department of Electrical Engineering, Computer Engineering, and Informatics.
Comprehensive Evaluation of Pre-processing Schemes: The thesis systematically compares multiple pre-processing approaches applied to prostate ultrasound images. These schemes may include noise reduction, contrast enhancement, normalization, and filtering, each of which can significantly impact the quality and interpretability of texture features extracted from the images.
Texture Feature Extraction and Analysis: The research explores various texture descriptors—such as statistical, structural, and model-based features—to quantify tissue characteristics within the prostate. By analyzing how pre-processing influences these features, the thesis provides insights into optimal workflows for robust texture analysis.
Experimental Validation: Using a dataset of prostate ultrasound images, the thesis evaluates the performance of different pre-processing and texture analysis combinations. Metrics such as feature robustness, discriminative power, and potential for clinical application are assessed, offering a data-driven foundation for selecting pre-processing strategies in future studies.
The findings of this thesis have significant implications for both research and clinical practice in medical imaging. By clarifying the effects of pre-processing on texture analysis, the work guides practitioners and researchers toward more reliable and reproducible image analysis pipelines. This is particularly relevant for prostate cancer diagnostics, where subtle textural differences in ultrasound images can indicate pathological changes. The methodology and results can be extended to other organs and imaging modalities, contributing to the broader field of computer-aided diagnosis. Ultimately, the thesis supports the development of automated, objective tools for early detection and characterization of prostate cancer, with the potential to improve patient outcomes and optimize healthcare resources.