RSNA: CAD and rads must join forces to reduce diagnosis error

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon

CHICAGO--There are many factors that can lead to a radiologist missing a cancer diagnosis during mammography screening; however, there are tools and methods currently in place that may reduce the occurrence of misdiagnosis, said Robyn Birdwell, MD, during the Computer-Aided Detection (CAD) in Practice: Hope and Hype session at the Radiological Society of North America (RSNA) annual conference on Tuesday.

CAD has been shown to increase the discovery of cancer by 7 to 20 percent.  However, CAD should be utilized by a radiologist only after their initial reading of the study has been completed. “[Radiologists] must make their own decision before engaging CAD,” Birdwell said.

CAD is an important adjunct method that can be put into practice to work with radiologists, due to the common interpretational and perceptional errors made that can be attributed to human error, she said.

In terms of the lack of perception of cancer, Birdwell noted three situations that can lead to the failure of diagnosis:

  • The CAD mark was not visible on the mammogram;
  • The mark was visible, but overlooked, an “observational oversight,” and;
  • The mark was visible and detected, but incorrectly interpreted as being normal.

Birdwell further explained circumstances under which a mistaken cancer-free diagnosis can take place, including a calcification being obscured by a vessel in the image.

Similarly, the modality itself can partially be to blame, as images presenting dense lesions or lesions at the end of the film may also contribute to a misdiagnosis.

Efforts can be made to minimize the aforementioned circumstances, said Birdwell, including the careful review of prior imaging studies, both screening and diagnostic, and the continuing medical education of the radiologists as being key elements.

Additionally, the double-read of a study had proven to increase cancer detection rates by 3 to 15 percent. “Two eyes are better than one-this hasn’t changed,” she said.

Despite the benefit of the double-read however, Birdwell noted problems associated with this method, including time, cost and which individual should ultimately be held responsible for the final read.

CAD, she explained, offers a significant opportunity to improve incidence of misdiagnosis. While there will always be issues with humans versus computers, CAD offers a “second opinion” in terms of discovering suspicious or abnormal marks on an image and supports the reduction of diagnosis error.

It is important to note the limitations of CAD, however, said Birdwell. When adjusting to CAD, call back rates have been shown to increase and potentially double within the first 21 months of its implementation, before returning to nearly typical levels. “We can never eliminate errors, regardless of screen modality,” she said.

“We need to understand human-computer issues. Do not change your mind about suspicious marks if CAD doesn’t mark it. The radiologist is the final decision maker,” concluded Birdwell.