Computer detection for mammograms

28 Oct 2008

Study shows that computer-aided detection has a role to play in reading mammograms

For breast cancer screening, mammograms utilising single reading with computer-aided detection (CAD) offer an alternative to double reading, concludes a study in the New England Journal of Medicine (October 16, 2008). The sensitivity of screening mammography for detecting small breast cancer has been shown to be higher when a mammogram is read by two readers as opposed to a single reader, with studies suggesting that double reading increases the rate of cancer detection by 4 to 14 %, and by 10 % according to a meta-analysis. Double reading is now standard practice in at least 12 European countries.

CAD systems use computer algorithms to analyse digital mammographic images, and identify and mark potentially “suspicious” regions that attract the radiologist's attention to features that might otherwise be overlooked. In the US , where single reading is standard practice, CAD systems are used in 25 to 30 % of all mammogram readings. This study, which was led by Fiona Gilbert from the Aberdeen Biomedical Imaging Centre ( University of Aberdeen , Scotland ), set out to determine whether the performance of a single reader using a CAD system would match the performance achieved by two separate readers.

Between September 2006 and August 2007, 31,057 women undergoing routine screening by film mammography at three breast screening programme centres in England  ( Manchester , Nottingham  and Warwickshire) were randomly assigned to one of three reading regimens. In the study, group A (with a planned number of 1000 subjects) was assigned to receive double reading only, group B (with a planned number of 1000 subjects) was assigned to receive single reading with CAD) and group C (with a planned number of 28,000 subjects) was assigned to receive both a single reading with CAD and a double reading.

Overall results show that for group C, who received both single reading with CAD and double readings, 227 cancers were detected among 28,204 subjects. When double reading was considered separately the proportion of cancers detected was 199 out of 227 (87.7%), and when single reading with CAD was considered separately the proportion of cancers detected was 198 out of 227 (87.2%), providing a P value of 0.89. The results produced cancer-detection rates of 7.02 per 1000 for the single reader using CAD, and 7.06 per 1000 for two readers. Additionally, the overall recall rates were 3.9% for the single reading group with CAD and 3.4 % for the double reading group (P<0.001). The estimated sensitivity, specificity, and positive predictive value for single reading with CAD were 87.2%, 96.9% and 18.0 %, with the corresponding figures for double reading being 87.7%, 97.4% and 21.1%.

The authors concluded that single reading with CAD offers an alternative to double reading. They went on to say that: “The additional costs of computer-aided detection equipment and the costs associated with an increase in recall must be balanced against the potential savings in reader time”, and noted that: “The small increase in the number of women recalled for assessment in the group given a single reading with computer-aided detection as compared with the group given a double reading is consistent with the results of previously published studies.”  They also recommended that the performance of CAD in full-field digital mammography and its performance in film mammography, used in the current study, needs to be investigated.

Dr Marco Rosselli Del Turco, President of the European Society of Breast Cancer Specialists (EUSOMA), congratulated the authors on having the foresight to initiate the study. “Up to now double reading has been the gold standard of mammography. We have been waiting for a well designed, prospective, randomised study to establish the role of CAD. This study provides a definitive answer about the value of adding CAD to single reading, and is likely to lead to a change in European guidelines,” he said, cautioning that the results only held for experienced radiologists, who read more than 5,000 mammograms per year. “Less experienced radiologists would be likely to experience higher recall rates.”

It would also be beneficial, Dr Rosselli Del Turco felt, to perform a trial looking to see whether there were advantages in adding CAD to the double reading procedure. On the subject of full field digital mammography, he felt that only a small scale study was required to establish the role of CAD. “It should be enough just to check several thousand patients to show that it performs in the same way,” he said.