Intrareader variability in mammographic diagnostic and perceptual performance amongst experienced radiologists in...

The purpose of this study was, for a group of experienced radiologists, to identify the magnitude of and statistical significance of intrareader variability in mammographic diagnostic performance or cancer diagnosis and mammographic perceptual performance or microcalcification detection.

Eight radiologist readers (8-30 years experience in radiology, five current BreastScreen readers) read a set of 100 digital mammograms on two separate reads with random case orders. Twenty-three of the 100 had proven malignancies, and 52 of the 100 had confirmed microcalcifications. The same mammograms were presented for both reads. The radiologists were requested to clear or call back cases and to indicate if any benign and malignant microcalcifications were present on the mammograms. Reading conditions were standardised.

Intrareader variability in accuracy was demonstrated to be between 0% and 6% for the diagnostic task of breast cancer diagnosis and between 0% and 16% for the perceptual task of microcalcification detection. Intrareader agreement in the group of readers was high; between 75% and 93% (κ=0.36-0.72) for cancer diagnosis and between 77.5% and 93% (κ=0.17-0.77) for microcalcification detection. There was no correlation between reader's experience in radiology or being a BreastScreen reader and level of intrareader variability in cancer diagnosis and microcalcification detection.

There exists intrareader variability in diagnostic and perceptual performance. Despite this variability, intrareader agreement remains high.