Observer errors in assessing tumour size from liver scans

Examination of liver scans is a component in the study of hepatic cancer, with changes in the size of perfusion defects attributed progression or regression of the disease. The experiment reported below involved a group of experienced oncologists examining a set of actual liver scans. Estimated probabilities of misclassification of the different types were calculated and were found to be of rather disturbing extent. For example, erroneous attribution of progression (or regression) at the 50 percent level was found to occur at estimated probabilities of 0.23 and 0.37 in the two cases when a single investigator examined a scan on two occasions, or different investigators investigated the scan, respectively. Results from the study will be discussed in some detail and the influence of unjustified round-off of measured values examined as a major factor in producing misclassification.