No-reference perceptual CT image quality assessment based on a self-supervised learning framework

Abstract Accurate image quality assessment (IQA) is crucial to optimize computed tomography (CT) protocols while keeping the radiation dose as low reasonably achievable. In medical domain, IQA based on how well an provides a useful and efficient presentation necessary for physicians make diagnosis. Moreover, results should be consistent with radiologists’ opinions quality, which accepted gold standard IQA. As such, goals of are greatly different from those natural addition, lack pristine reference images or in real-time clinical environment makes challenging. Thus, no-reference (NR-IQA) more desirable settings than full-reference (FR-IQA). Leveraging innovative self-supervised training strategy object detection models by detecting virtually inserted objects geometrically simple forms, we propose novel NR-IQA method, named deep detector (D2IQA), that can automatically calculate quantitative CT images. Extensive experimental evaluations anthropomorphic phantom demonstrate our D2IQA capable robustly computing perceptual it varies according relative levels. when considering correlation between evaluation metrics scores, marginally superior other even shows performance competitive FR-IQA metrics.