CAD emphasis shifts to diagnosis

Computer-aided detection and diagnosis tools were showcased at Saturday's "ESR meets Germany" session. Speakers highlighted four key clinical areas where advances could make a real difference to diagnostic decision making.

The potential of CAD in breast imaging was first noted back in 1967, said Prof. Ulrich Bick, a professor of radiology at the Charité Medical University of Berlin. CAD algorithms have evolved considerably since that initial report, leading to a steady improvement in performance. He showed data from a 2008 study revealing 100% sensitivity for breast CAD in detecting microcalcifications and 91% sensitivity for malignant masses.

"Masses are easy to see but difficult to interpret," Bick said. "Basically, the entire breast parenchyma is made up of masses. So if you look at a mammogram, it is very difficult to tell which of these are cancers and which are not. Detection of associated architectural distortion is difficult for the computer."

Bick predicts that future breast CAD systems will be equipped with intelligent workstations that can help determine if lesions are malignant. This will be done by comparing key features seen on the mammogram to those in a database.

Radiologists must also learn how to use CAD appropriately. Given the large number of false-positive readings generated by breast CAD, a measured approach is necessary. Bick recommends neither blind faith in the computer nor complete skepticism.

He suggested that radiologists will gradually come to regard CAD as a trusted second reader for screening mammography, rather than as a "spellchecker." No one is yet obliged to use CAD when reading screening mammograms, but if future data demonstrate that CAD can reduce the mortality of breast cancer, then this issue may need to be revisited.

"At some stage, CAD will be superior to radiologists," he said.

Chest imaging is another area where CAD is used at present, but it has the potential to do much more, according to Prof. Hans-Ulrich Kauczor, chair of the radiology department at University Hospital Heidelberg.

CAD is currently helping radiologists find pulmonary nodules and early signs of lung cancer on CT. Detection in itself is not enough, however, he said. The ideal CAD system should also be able to perform accurate volume measurements. Suspicious lesions can then be monitored to see if they change in size and, if so, by how much.

"You want to measure the tumor doubling time," he said. "But we also have to be aware that there is a significant error in making volumetry measurements. You have to demonstrate an increase in volume of more than 30% or 40% to be really sure that a nodule is actually growing."

Kauczor would like to see chest CAD used to assess a far greater spectrum of diseases related to smoking. He envisages computerized systems that could evaluate chronic obstructive pulmonary disease. It should be possible to create tools that can distinguish between emphysema-predominant COPD and airway-predominant COPD, a distinction that could influence treatment, he said.

Other advances forecast by Kauczor include a greater use of CAD for emphysema, 4D imaging of lung hyperinflation, and the inclusion of vascular imaging tools in chest CT workup.

Continuing the theme of computer-aided analysis, Prof. Heinz-Otto Peitgen of the Centre for Medical Diagnostics and Visualization (MeVis) in Bremen described a variety of state-of-the-art tools that may assist liver imaging and intervention. His presentation explored three different aspects of computer-aided diagnosis and therapy: response evaluation in chemotherapy, surgical planning, and radiofrequency ablation.

Detection rather than diagnosis will remain the primary role for CAD in the colon, said Prof. Andrik Aschoff, an associate professor of radiology at the University of Ulm. Early identification and removal of polyps promises to reduce the rate of colon cancer. But if this type of screening is to be performed with CT colonography, then CAD may be needed to reduce variations in reader sensitivity.

Aschoff outlined the three alternative paradigms for using CAD: as an initial reader to filter out "normal" cases, as an aid to the reporting radiologist, and as an independent second reader. This latter option tends to give the best sensitivity, though at the cost of slightly extended reading times, he said.

"CAD is very promising in CT colonography. It has the potential to support radiologists in providing the best possible detection rates of polyps. This is extremely important. If we do CT colonography on our patients, we want to give them the assurance that they have no large polyps and are not going to develop colorectal cancer," Aschoff said.