"Radiomics dataset PCA Framework"

"Ensuring the quality of medical images before their use in automated diagnostic pipelines as radiomic is a critical yet often overlooked step. Even subtle visual degradations, such as blur, noise, or artifacts, can significantly compromise the reliability of automated analyses, such degradations may also affect the stability of extracted quantitative features, potentially reducing the reliability of downstream clinical decision-support systems. In current clinical practice, there is often no upstream quality control mechanism capable of flagging or filtering out poor-quality images before they are automatically processed. To address this issue, this paper proposes an unsupervised framework for automatic medical image quality assessment based on the extraction of general-purpose visual quality metrics from each image. The goal is to provide a quality-aware characterization of medical images before their inclusion in automated analysis pipelines such as radiomics and imaging-based artificial intelligence. These metrics are analyzed using Principal Component Analysis (PCA) to map the behavior of different types of visual alterations within a reduced-dimensional feature space. The core idea is that each form of degradation induces a coherent displacement along a specific direction in the PCA space, producing structured trajectories associated with specific artifact types. This representation enables the identification of potentially degraded images without relying on anatomical or pathological content, providing a generalizable and scalable approach for image quality evaluation. From a clinical perspective, the proposed framework can act as an upstream quality pre-screening and annotation tool for radiomics and imaging-based AI pipelines, helping prevent low-quality acquisitions from propagating errors into automated diagnostic analyses."