Lung CAD can avert malpractice lawsuits

An international team of researchers has determined that a computer-aided detection system can spot lung lesions in routine chest exams read as normal. CAD in this setting can cut perceptual errors, thereby reducing medical malpractice lawsuits, according to the study published in the September issue of Chest.

Nearly 70% of all medical malpractice lawsuits filed against radiologists are related to a missed diagnosis. Perceptual errors account for almost 60% of these cases, according to statistics cited by the researchers.

It is well known that lung lesions are missed on thoracic CT, but the risk increases with the latest generation of multislice CT scanners, which routinely result in more than 200 transaxial images. Consensus or double readings may help to reduce error, but they are labor intensive.

"CAD is rapidly becoming a vital part of contemporary medical imaging," said senior investigator Dr. U. Joseph Schoepf, director of CT research and development at the Medical University of South Carolina, Charleston.

Previous investigations evaluating chest CAD have largely been investigational and applied to highly selected patient subgroups, Schoepf said.

"The current study was aimed at crossing the threshold from science to practice and to show how CAD tools can be beneficially utilized to enhance diagnosis at thoracic CT," he said.

Led by Dr. Kersten Peldschus, a radiologist at Brigham and Women's Hospital, researchers used a prototype of the ImageCheckerCT CAD system by R2 Technology to reassess the results of 100 consecutive patients initially reported as normal at clinical double reading. The other academic center involved was the University of Munich in Germany.

Indications for chest CT were suspected pulmonary embolism (33), lung cancer screening in a high-risk population (28), or follow-up for a cancer history (39).

Two experienced radiologists analyzed each CAD finding and confirmed or dismissed the mark. A third experienced radiologist settled any equivocal calls. The significance of noncalcified, focal lung lesions was classified according to size: high (greater than 10 mm), intermediate (5 to 9 mm), or low (less than 4 mm).

CAD marked 285 image features within all patients. In 33 patients, 53 lesions were deemed significant by readers, and none had been previously reported. Five lesions were of high significance, 21 intermediate, and 27 low. Nineteen patients had at least one lesion of high or intermediate significance.

In the pulmonary embolism group, CAD detected 20 significant lesions (one high, three intermediate) in 12 patients (36%). In the lung cancer screening group, CAD detected nine significant lesions (one high, five intermediate) in eight patients (28.6%). In the cancer history group, CAD detected 24 significant lesions (three high, six intermediate) in 13 patients (33%).

Most of the high and intermediate missed lesions were located in the central areas of the lung, close to adjacent vessels, according to the study. Readers dismissed 125 CAD marks as false positives, for an average of 1.25 per case (range, 0 to 11). Most false positives occurred at vessel crossings or vascular structures that were distorted by motion.

While not statistically significant, CAD found more small nodules in the pulmonary embolism group, possibly due to observer perception. Readers may focus their attention more on vascular structures during the interpretation of studies performed for suspected pulmonary embolism than on nodule detection.

Conversely, CAD detected the lowest per case rate of lesions in the lung cancer screening group. In this group, radiologists are focused on the detection of focal lung lesions.

The authors concluded that the use of CAD as a second reader can offset the inherent variability of human observation by finding oversights that are missed in the original review.

"Our results show that use of CAD tools is capable of improving sensitivity for detection of lung nodules, even for highly skilled observers," Schoepf said.

While previous studies mostly showed that CAD can be used to find more lesions, researchers intentionally used CT scans that were considered normal. Detection of additional lesions in patients with known lung nodules in most instances does not change patient management, while detection of lesions in scans thought to be normal significantly affects patient management.

The study has several drawbacks. It focused on using CAD to reduce only perceptual error and not classification error. It's possible, the authors noted, that some of the 53 lesions reported by CAD originally had been seen by radiologists but deemed not clinically significant. The authors also do not address the ultimate outcome of the CAD-detected lesions.

Schoepf and others are working on developing novel software tools aimed at improving diagnosis in other areas besides chest CT. These include detecting small, peripheral pulmonary emboli and coronary artery stenosis at contrast-enhanced CT.

For more information from the Diagnostic Imaging archives:

Lung CAD makes further improvements

Study validates CAD's impact in early detection of invasive breast cancer

Radiologists puzzle over choices in chest CAD

CAD holds key to screening virtual colonoscopy's future

CAD breaks through in liver imaging