Radiologists top computers in breast cancer screening

Article

When it comes to spotting suspicious lesions on a screening mammogram, computer-aided detection is no match for a dedicated breast imaging specialist, according to a large study by researchers at Yale University.

When it comes to spotting suspicious lesions on a screening mammogram, computer-aided detection is no match for a dedicated breast imaging specialist, according to a large study by researchers at Yale University.

The prospective study examined 5875 consecutive screening mammograms performed at Yale's Breast Imaging Center between February and September 2005. Two-thirds of the studies were performed on conventional film-based systems and one-third on digital equipment.

The study compared CAD markings with interpretation by the facility's radiologists, whose experience in breast imaging ranged from four to 20 years. The researchers looked at overall correlation and compared characterization of masses, calcifications, density, and architectural distortion. They also reviewed follow-up diagnostic mammograms, but these were not available in all cases.

Radiologists' interpretations indicated 735 lesions requiring action in 637 patients. CAD markings correlated with these interpretations in just 48% of lesions.

"There is poor correlation between CAD and findings that require action. These results erode confidence in CAD," said Dr. Lara Bryan-Rest, who presented the study at the 2005 RSNA meeting. "Complete reliance on CAD by nonexperts is strongly discouraged."

The level of correlation varied depending on the type of lesion. CAD markings correlated with radiologist interpretation in 65% of calcification cases but in only 43% of masses. There was no difference based on digital or analog technique.

Another study, however, found that CAD on a full-field digital mammography system best helps the less experienced reader. Dr. Roberta Jong and colleagues from Sunnybrook and Women's College in Toronto reviewed 220 full-field digital mammograms, 70 of them cancerous, prior to and then with CAD. The readers consisted of one resident, four fellows in breast or women's imaging, and five experienced breast imagers.

The sensitivity of experienced and trainee radiologists without CAD was 79% and 71%, respectively. Using CAD, sensitivities improved to 80% and 74%. The specificity with CAD decreased only slightly more in the trainees (from 65% to 62%) than the experienced readers (from 73% to 72%). The sensitivity of CAD alone was better than any of the radiologists (94%), but specificity was considerably lower (39%).

Recent Videos
New Mammography Studies Assess Image-Based AI Risk Models and Breast Arterial Calcification Detection
Can Deep Learning Provide a CT-Less Alternative for Attenuation Compensation with SPECT MPI?
Employing AI in Detecting Subdural Hematomas on Head CTs: An Interview with Jeremy Heit, MD, PhD
Pertinent Insights into the Imaging of Patients with Marfan Syndrome
What New Brain MRI Research Reveals About Cannabis Use and Working Memory Tasks
Current and Emerging Legislative Priorities for Radiology in 2025
How Will the New FDA Guidance Affect AI Software in Radiology?: An Interview with Nina Kottler, MD, Part 2
A Closer Look at the New Appropriate Use Criteria for Brain PET: An Interview with Phillip Kuo, MD, Part 2
How Will the New FDA Guidance Affect AI Software in Radiology?: An Interview with Nina Kottler, MD, Part 1
A Closer Look at the New Appropriate Use Criteria for Brain PET: An Interview with Phillip Kuo, MD, Part 1
Related Content
© 2025 MJH Life Sciences

All rights reserved.