This thesis presents a comprehensive analysis of existing research and improvements in numerous aspects of anomaly detection via CPU thermal sensors, addressing important difficulties and novel solutions. The results presented in this thesis demonstrate that thermal monitoring constitutes a feasible and complementary mechanism to improve the security of the system. Monitoring processor temperature data for intrusion and anomaly detection introduces promising directions for future research. I consider these techniques to be the foundation for increasing the security of processors using their heat. In fact, more research is needed to incorporate such approaches into real-time monitoring frameworks and improve data protection.

This thesis provides a comprehensive investigation into image inpainting forgery detection, addressing the growing challenges posed by increasingly sophisticated image editing techniques. As inpainting methods have evolved from basic restoration tools to advanced AI-driven systems capable of generating highly realistic content, the need for reliable forensic detection mechanisms has become critical. Overall, the thesis delivers novel datasets, hybrid analytical–deep learning detection architectures, and extensive empirical validation. It demonstrates that combining wavelet-based statistical modeling with modern neural networks offers a powerful and scalable framework for detecting sophisticated inpainting forgeries in digital images.

This thesis provides a comprehensive review of existing research and advancements in various aspects of microprocessor design and optimization, covering key challenges and innovative solutions. It introduces the issue bottleneck, a fundamental problem in microprocessor architectures, which refers to the limitations in the number of instructions that can be issued per clock cycle. It reviews different strategies and mechanisms proposed in the literature to overcome these limitations and improve overall processor throughput. In addition, it addresses the critical issues of security and data consistency in microprocessor systems with speculative execution capabilities. 

This doctoral thesis addresses automated coronary centerline extraction from Cardiac Computed Tomography Angiography (CCTA), a key problem in medical image segmentation with direct relevance to cardiovascular diagnosis. The final chapter concentrates on the core medical application: automated coronary centerline extraction from CCTA data. A 3D U-Net–based architecture is proposed, enhanced with a novel loss function and augmented patch-based training strategy. The method achieves high accuracy (90–95%) and strong overlap metrics (90–94%) on the Rotterdam Coronary Artery Centerline Extraction benchmark. Importantly, the network demonstrates robustness to severe class imbalance and sparse annotations, highlighting its practical viability in real-world clinical datasets.