THE computer systems widely used to help spot breast cancers in the US do not provide the kind of improvements the manufacturers claim, say researchers who have done independent tests.
In the US it is becoming standard practice to use computer-aided detection systems to help spot tumours in mammogram images. After radiologists have looked at a mammogram, the CAD system highlights any abnormal features they might have missed. But Paul Taylor of the Centre for Health Informatics at University College London questions their value. He told a conference on computer-assisted radiology in London this week that CAD systems mistake normal features for possible cancers far too often.
His team was assessing the technique to see if it could help relieve the UK鈥檚 acute shortage of radiologists. With a CAD system as back-up, only one radiologist, instead of two, would have to look at each mammogram. The researchers asked 45 radiologists to look at 180 mammograms, and then to reassess each one with the help of the ImageChecker system made by R2 Technology of Sunnyvale, California. On the basis of these results, Taylor calculates that a radiologist checking 5000 patients a year would receive about 10,000 prompts from the system, of which just five 鈥� 1 in 2000 鈥� would be a cancer that had been missed on the first examination.
Advertisement
In routine use, the high rate of false alarms means radiologists might start to ignore the prompts. Alternatively, they might needlessly recall patients for a second scan, Taylor says, putting them through unnecessary anxiety and exposing them to additional X-rays.
鈥淭he company has carried out extensive tests of its technology to get approval from the US Food and Drug Administration,鈥� says team member Rosalind Given-Wilson, a radiologist at St Georges Hospital in London. 鈥淭he people involved in those trials were very enthusiastic but in our study we have not been able to demonstrate the same sort of benefits.鈥�
Jimmy Roehrig, chief science officer for R2 Technology, rejects the criticism. 鈥淭here is growing evidence in the US, where there are 1000 systems installed, that it does help,鈥� he says. Indeed, one trial in the US suggested CAD systems can improve cancer detection by up to 20 per cent.
But Taylor thinks the dramatic improvement occurred because the radiologists did not look as closely at the images the first time round. 鈥淲hen you know you have a second look, you don鈥檛 look as hard as you should.鈥� He wants to carry out a large-scale study that would eliminate these flaws.
Other experts point out that CAD systems are constantly being improved. 鈥淭here is no doubt that it will get better,鈥� Given-Wilson says. 鈥淏ut the rate at which it is being taken up in the US is not justified by what it can do at the moment.鈥� Medical insurers in the US now pay an extra $17 per patient to departments that adopt CAD technology, and this is encouraging premature take-up, she says.