We examined in the previous chapter several common computable measures of image quality, as well as subjective quality ratings. Although these quality measures are useful in many ways, for medical images one wishes a quality measure to take proper account of the diagnostic purpose of the image. The ROC methodology discussed in the previous chapter is one approach to this. In this chapter, we present several studies that attempt to evaluate diagnostic utility of the images more directly. The radiologists were not specially trained or calibrated in any way for these judging tasks, as the goal of these studies was specifically to evaluate compression performance in the context of radiologists carrying out tasks that resembled their everyday work. No constraints were placed on the viewing time, the viewing distance, or the lighting conditions. The judges were encouraged to simulate the conditions they would use in everyday work. The tasks were detection of lung nodules and mediastinal adenopathy in CT images, measurement of blood vessels in MR chest scans, and detection and management tasks in mammography. As we shall see, the results indicate that when these images are used in situations resembling everyday work, substantial compression can be applied without affecting the interpretation of the radiologist.

Portions reprinted, with permission, from IEEE Trans. Medical Imaging, 12(4): 727-739, Dec. 1993 and Proceedings IEEE, 82(6): 919-932, June, 1994.

0 0

Post a comment