There's a lot of hope wrapped up in computer assisted detection (CAD) - from both physicians and patients. Increasing computer speed has facilitated both software development and higher volumes of detailed images that radiologists and other specialists thus need assistance in reading. That's where CAD comes in - and that's what we're looking at in this month's cover story - "CAD Makes Its Mark on Diagnosis."
All types of CAD struggle to overcome the innate challenge of achieving high sensitivity (detecting a high proportion of significant lesions) while maintaining high specificity (keeping false positives low). Vendors vie with these interlocked concepts to see how many false positives will be tolerated by a busy radiologist and in which applications. When a balance is struck, CAD achieves productivity tool status. CAD "nirvana" means reducing oversight or detecting cancers at an earlier stage - coupled with efficiency of interpretation and boosts in radiologist productivity (while concentrating their viewing time most effectively on images in which cancer is present).
Breast CAD has had the longest time to mature since its 1998 introduction and is doing just that, maturing. Performance is proving equally successful with analog and digital mammograms. Breast CAD systems come in different flavors for analog and digital mammograms and breast MRI and breast ultrasound images. New systems are more sophisticated in their ability to detect a wider variety of microcalcifications, masses, densities and architectural distortions (although effectiveness dwindles in the same order these characteristics are listed). They also are beginning to help physicians to visually prioritize the marks.
Lung and thoracic CAD for x-ray and CT were next on the scene, looking to detect more pulmonary nodules earlier (now, only 16 percent of lung cancers are found in the early, treatable stages) and with more ease. The newcomer, colon CAD, looks at virtual colonography exams seeking to improve detection accuracy for polyps and reduce the number of false positives. In fact, a recent study found it could out-do even experienced readers, with CAD identifying 81 percent of polyps vs. 70 percent detection by the expert readers.
In the research realm, some of which were detailed at the recent RSNA meeting in Chicago, algorithms are being developed to analyze brain MRI lesions, quantify left ventricular heart volumes during CTA, characterize subcutaneous melanomas via CT, find cancers in PET images and even identify diseases such as emphysema and pulmonary fibrosis or pulmonary emboli. But even today, CAD is taking medicine to the next level - diagnosing more cancers, earlier and better.