In coronary cineangiography, both X-ray absorption and light scatter in the image intensifier tend to degrade image quality, and so affect the accuracy of densitometric measurement of vessel diameter. To investigate this problem, we compared the accuracy and precision of the densitometric method and the edge detection method in the automated detection of stenosis in both vessel phantom and clinical studies. In the phantom study, the X-ray penetration was varied by altering the thickness of the model, and the change in the measured diameter obtained by each of the two methods was evaluated simultaneously. A difference of 5 mm in the thickness of the model was found to alter significantly (P &lt
0.01) measurement of the diameter obtained using the densitometric method, but not that obtained by the edge detection method. In the clinical part of the study, the accuracy of each method in the automated detection of coronary stenosis was evaluated. With respect to the detection of stenosis, the level of disagreement between the assessment of the 3 observers and what was detected by densitometry (22.8%) was 2.9 times higher than the disagreement between the observers' assessment and what was detected using the edge detection method (7.9%). When the background density of the coronary cineangiogram along the axis of the vessel was uneven, many vessel segments which had been evaluated as normal when edge-detection was used were evaluated as stenosed when densitometry was used. This study, then, demonstrates that the Lambert-Beer law does not apply in cases where the thickness of the subject varies in different locations along the axis of the same vessel. We therefore conclude that densitometry is not a reliable means of assessing coronary stenosis in such cases, due to veiling-glare and scatter, and recommend that it not be routinely used in the automated detection of coronary arterial stenosis. © 1994 Kluwer Academic Publishers.