NEW YORK—Computer-aided detection (CAD) of breast cancer using digitized mammograms could have detected malignancies at least 1 year earlier than film assessment by radiologists alone, according to findings from the Elizabeth Wende Breast Clinic, Rochester, New York.
In a retrospective study, the investigators concluded that "CAD has the potential to decrease the false-negative rate by more than one-third," and in a prospective study they determined that if radiologists at their clinic had acted on mammogram areas marked by CAD as suspicious, an additional 16% of breast cancers would have been detected.
Stamatia V. Destounis, MD, a staff radiologist at the Elizabeth Wende Breast Clinic and associate professor of radiology at the University of Rochester, reported the results at a media briefing on advances in breast cancer diagnosis and treatment, sponsored by the Radiological Society of North America (RSNA) and held at Mount Sinai Hospital in New York City. Both studies were recently published in the journal Radiology.
Dr. Destounis and her colleagues conducted a retrospective study of 64,442 patients who had been screened for breast cancer at the Elizabeth Wende Breast Clinic in the year 2000. Of 519 women with a breast cancer diagnosed, 318 had had at least one prior mammogram 1 or more years previously.
The investigators reviewed both prior and current films of these 318 patients and identified 52 patients who had actionable false-negative screenings that they believed were cancer.
The researchers used a commercially available CAD system to assess whether the technique could enhance detection of breast cancer by picking up cases that had been missed in prior film assessment by radiologists.
CAD, which has been referred to as a second pair of eyes, utilizes computer-learning software to highlight suspect areas on digitized versions of mammography films, with distinct marks to indicate masses, microcalcifications, and other areas for further investigation. The marks made by CAD still must be interpreted by a radiologist.
Since 1995 at the Wende Clinic, all mammogram films have been double-read independently by two radiologists, with the second radiologist nonblinded to the diagnosis made by the first one. An internal audit for 1996-2000 at the clinic revealed that double-reading alone had led to a 9% increase in the number of detected breast cancers.
CAD Marked Missed Cancers
CAD marked 37 (71%) of the 52 cancers that had been missed by double-reading at least 1 year earlier. Of these 52 cancers, CAD detected cancer in 3 of 9 women who later returned to the clinic with breast cancer symptoms (such as a lump, breast pain, or nipple discharge) and in 34 of 43 women who the investigators determined had received a false-negative diagnosis.
The investigators then assessed a variety of patient-related and other factors to ascertain how those 52 cancers had been missed. "We looked at the composition of the breast tissue, we looked at tumor size, and we looked at lymph node metastasis," Dr. Destounis said. The majority of false negatives, nearly 65%, were in dense breasts, with an average tumor size of 8 mm on the earliest films vs 12 mm at final detection and diagnosis.
CAD is not a perfect detector, she explained, in that only 34% (75 of 218) of marks actually turned out to indicate cancer: "For every one mark that indicated cancer, there were two marks that were false positive," she said. In its favor, however, CAD did correctly mark all three of the patients who had lymph node metastases 1 year before these were identified conventionally.
"We concluded that CAD had the potential to decrease the false-negative rate" in the setting of a double-read of mammography films, Dr. Destounis told ONI. With the 27 occult cancers included, conventional assessment at the clinic yielded a 31% false-negative rate (98 of 318), yet "if we had used CAD and actually been prompted by all the marks, it would have reduced the false-negative rate to 19% [61 of 318]," she added.
Prospective Study
Based on these retrospective findings, the investigators initiated a prospective study to assess the positive predictive value of CAD and the associated recall rate, by comparing breast screening outcomes as performed by two radiologists vs one radiologist backed up by CAD.
The investigators enrolled 19,586 routine breast cancer screening patients and analyzed them with CAD from November 2000 to July 2002. In contrast to their standard mammogram-assessment procedure, the double-reads for this study were double-blinded, so neither of a pair of radiologists interpreting a film was aware of the other’s assessment. CAD was then applied to film assessment by each reader. A total of 116 visible cancers were identified.
Six More Cancers
"We found six more cancers by listening to the computer-aided marks . . . which prompted us to work up an additional 7% of cancers [with small-needle biopsy]," Dr. Destounis said. The mean size of the cancers was 9.7 mm, all were invasive, and only 57% were minimal, meaning less than 1 cm of invasion or any-size ductal carcinoma in situ.
"The conclusion we drew from this prospective trial is that two doctors are still better than one doctor plus CAD, because two doctors found 93% of the cancers and were prompted only 7% of the time; however, even when we double-read, CAD did reduce our overall false-negative rate by 5%," she added. Two radiologists found 85 of 116 visible cancers, including interval cancers (73%), and CAD marked 87 of 116 (75%).
Noting that radiologists do find many breast cancers with high accuracy and reproducibility, she emphasized that film interpretation by a radiologist supplemented with a second interpretation using CAD still yielded "a 5% to 7% increase in the detection rate" achieved by double-reading.
The problem, she said, is that radiologists don’t pay enough attention to marks indicated by CAD. Although CAD marks potentially malignant areas on a digitized film, radiologists "don’t always act on it. If we had listened to CAD, we would have found another 16% of the cancers earlier on. In order for us to get better, we have to find some way to have more confidence in the computer and listen to the correct marks."
Dr. Destounis noted that CAD is becoming the standard of breast cancer diagnosis in many hospitals, but she added that, while costs will decrease over time, "CAD is an expensive up-front cost at $15 to $22 per mammogram, the upkeep is expensive, and it does slow down the physician because you have to decipher the marks and determine what’s important. CAD is not meant to replace the doctor; it’s a tool to help us decide in threshold cases whether to recall the patient or not."
Dr. Destounis said that she recommends CAD for all mammograms. "There are multiple versions of CAD systems available, and, in general, all are very good at finding the microcalcifications that indicate early disease," she said. "And I think reimbursement is sufficient to cover the extra cost associated with use of CAD."
