Akademik Veri Yönetim Sistemi
Özgür Yayınları, Gaziantep, 2026
This monograph presents a rigorous, evidence-based investigation into the effects of lossy image compression on diagnostic reliability in clinical imaging systems. Departing from the conventional assumption that pixel-level fidelity metrics such as PSNR and SSIM adequately characterize compression quality, the authors demonstrate—through mathematical proof and systematic empirical analysis—that variance preservation in intensity space does not guarantee preservation of diagnostically relevant information in representation space.
The work introduces the concept of the Variance–Information Fallacy: the implicit but erroneous identification of signal variance with diagnostic information content. Through controlled experiments on the multi-center TCGA-LUAD computed tomography cohort, the authors show that conventional compression can maintain global fidelity metrics at clinically acceptable levels while simultaneously inducing measurable topological perturbations, radiomic feature instability, and degradation of downstream classification performance.
The monograph develops a multi-layer evaluation framework that extends classical rate–distortion analysis to incorporate structural fidelity (persistent homology, Betti numbers) and task-level performance (AUC, staging accuracy, information retention ratio). A formal Fidelity–Topology Decoupling Theorem is established, proving that pixel-domain fidelity and structural integrity can diverge under lossy compression—a phenomenon empirically confirmed through phase-transition dynamics in topological descriptors.
In its final part, the work extends this framework to the era of clinical artificial intelligence, modeling compression as a geometric perturbation operator within deep learning pipelines. The analysis demonstrates how compression-induced artifacts cause manifold drift in latent feature spaces, alter decision boundary geometry, and introduce hidden distribution shifts between training and deployment environments.
Written for researchers in medical image processing, biomedical engineering, radiological physics, and quantitative imaging, this book provides a mathematically rigorous foundation for understanding why compression must be treated not as a peripheral storage mechanism but as a central variable in the design of trustworthy medical AI systems.