Bone Suppression and CAD Software Improves Lung Nodule Detection on Chest X-Ray

Three studies reported that using bone suppression and computer-aided detection (CAD) software can significantly improve the detection of lung nodules on chest X-ray images.

CAD performed well on images captured by both portable and upright X-ray machines in the studies, and that most false-positive marks—the marking of portions on the chest X-ray that are not lung nodules—call radiologists’ attention to other significant abnormalities, including serious lung disorders, and to medical devices in or on the chest that radiologists would be able to swiftly disregard. The studies’ findings were presented Radiological Society of North America (RSNA) annual meeting held November 2012 in Chicago (IL, USA). Riverain Technologies’ ClearRead software was used in each of the studies.

Riverain’s ClearRead software suppresses the ribs and clavicles on traditional chest X-ray images, providing radiologists with a clear, unobstructed image of the lungs and exposing abnormalities that may be a sign of disease. The software is easy to install and adopt, requiring no additional imaging equipment or staffing, and no additional radiation exposure to patients. It can be used to immediately enhance images generated by all X-ray machines throughout a hospital or health system.

In a study at Radbound University Nijmegen Medical Center (RUNMC; The Netherlands), five attending radiologists and three radiology residents who were not skilled with the Riverain Technologies (Dayton, OH, USA) software detected more confirmed lung nodules utilizing ClearRead bone suppression 2.4 software than when they evaluated the X-ray images on their own.

The radiologists read 300 chest X-ray images with and without bone suppression (111 X-rays with solitary lung nodules and 189 control cases with no nodules). They identified 79% of the lung nodules with bone suppression, with about 0.3 false positive per image compared to 71% detection of lung nodules and 0.21 false positive per image when they reviewed the X-rays on their own.

“The detection of hard-to-find nodules classified as ‘moderately subtle’ and ‘subtle’ was especially improved,” said Steven Schalekamp, MD, a PhD student in the department of radiology of RUNMC. “The radiologists using the bone-suppression software found 26% of the lung nodules that were missed entirely on conventional X-ray images.”

In a separate RUNMC study involving 300 radiographs and eight readers, the radiologists’ performance also improved when they read the images using Riverain’s CAD software, ClearRead +Detect 5.2, which circles suspected lung nodules on bone-suppressed X-ray images. The radiologists accurately detected 79% of 111 lung nodules with CAD, versus 73% when reviewing the bone-suppressed image without CAD.

The radiologists’ false-positive rate per image was 0.28 when they used CAD and bone suppression combined, and 0.23 when they used bone suppression without CAD. ClearRead +Detect found more than half of the nodules that were missed by the radiologists.
In a University of Chicago Medical Center (IL, USA) study of 608 unselected, consecutive chest X-ray images, (479 X-rays with abnormalities, such as lung nodules, other lung diseases and scars; and 129 X-rays without significant abnormalities), Riverain’s ClearRead +Detect 5.2 CAD software on its own, without radiologist interpretation, detected 72% of lung nodules present, or 82 of 114 nodules. The software detected a majority of the nodules regardless of whether the image had been captured using a portable or upright X-ray machine, and even when other abnormalities, including underlying lung disease, were present.

“CAD found the majority of lung nodules in this large study, which simulates a real-world, clinical setting where the use of portable X-ray images and the presence of other abnormalities in patients’ chests make lung nodule detection challenging,” said Steve Worrell, Riverain’s chief technology officer.

The CAD software had 1.3 false-positives per image (0.7 on X-rays without significant abnormalities and 1.5 on images with abnormalities). Forty-one percent of the false-positives across all radiographs indicated other notable pathologic changes, including fluid in the lungs (edema), pneumonia, a partial or full collapse of a lung (atelectasis), other lung diseases, calcifications, and scars.

Nineteen percent of the false-positive marks were attributed to medical devices in or on the chest, including electrocardiogram leads and catheter ports. As Dr. Schalekamp’s study suggests, the number of false positives actually reported in a clinical environment would be lower because radiologists use their judgment to rapidly reject automatically generated marks that are caused by grossly abnormal tissue or medical devices.

“We expect that after a learning period, radiologists would be able to dismiss false-positive CAD marks very quickly,” Dr. MacMahon said. “We know that small cancers are often visible in retrospect but were not detected on earlier X-rays,” Dr. MacMahon added. “Routine use of bone suppression and CAD on chest X-rays will reduce the chances of a cancer being missed, thus improving the likelihood of diagnosing unsuspected early-stage disease.”

Related Links:
Radbound University Nijmegen Medical Center
Riverain Technologies
University of Chicago Medical Center