2016 年 56 巻 1 号 p. 1-7
Image quality is a concept to represent information content in the image. Therefore, medical images must have sufficient information needed for the correct diagnosis. Basic knowledge about diagnostic image quality is essential to perform imaging diagnosis at high level.
Physical image quality consists of three components; (spatial) resolution, contrast (resolution), and noise. Resolution and contrast make signals needed for detection task, while noise is the factor that disturbs signal detection. Some measures, such as signal-to-noise ratio or contrast-to-noise ratio are used to represent integrated physical image quality.
Observer performance changes according to the physical image quality. High observer performance can be expected by using high physical quality images to some extent. However, there is a limit where observer performance reaches maximum even if the physical image quality further increases.
Receiver Operating Characteristic (ROC) curves are used to evaluate final diagnostic outcome from the images. The diagnostic image quality must be assessed using ROC analysis. Observer performance using some phantoms is highly correlated with the diagnostic image quality obtained from ROC curves. Therefore, physical image quality, observer performance, diagnostic image quality are closely associated with each other. The key factor in evaluating diagnostic image quality using ROC analysis is appropriate selection of the samples and the observers for the given task.