RSNA: Image manipulation tools make for inconsistent findings in the NICU

CHICAGO—The ability to manipulate chest images in PACS worsened already poor consistency in reporting between radiologists and neonatologists, according to a poster presentation Nov. 27 at RSNA 2012. In an interview with Health Imaging, the lead researcher called for a larger multi-center study to quantify the extent of the problem and suggested operational and technical approaches to address the issue.

Denise A. Castro, MD, of the Hospital for Sick Children in Toronto, and colleagues sought to determine the consistency of chest x-ray impressions among radiologists and neonatologists and to evaluate the effect of image manipulation on report impression.

The study focused on 60 pairs of frontal neonatal chest x-rays, with a pair consisting of a recent and prior exam. Three radiologists and three neonatologists reviewed all 60 sets of images and independently provided an impression of lung disease as “improved,” “no change” or “worsen.”

The physicians interpreted the images a second time after images were windowed/manipulated as needed.

Castro and colleagues reported 100 percent agreement between all six observers in 5 percent of cases with image manipulation, a drop from 13 percent agreement with no manipulation. The three radiologists agreed 100 percent on the impression in 35 percent of cases with manipulation and 42 percent of the cases without image manipulation. The three neonatologists agreed 100 percent in 13 percent of cases with manipulation and 30 percent of cases without manipulation.

“The ability to manipulate images results in a decrease in interpretation consistency in both groups,” wrote Castro and colleagues. The researchers called for new methods to standardize and equalize x-ray image appearance and to synchronize windowing of images with previous exams.   

Castro shared two approaches to address the issue with Health Imaging in an email. The first focuses on improved techniques to standardize patient positioning and radiation exposure. However, she explained this could be difficult with very ill infants.

The second answer is more technical. “Development of a hardware/software that can automatically equalize the appearance of two radiographs that need to be compared is desirable and, we believe, ultimately required. This potentially can be at the cassette level, where a standard small plate of varying densities could be implanted along the margin,” Castro wrote.

According to Castro, existing software allows the plate to equalize the appearance of a current x-ray with a prior exam, which should facilitate comparison and also synchronizes windowing to save time. “[This] should make it easier and quicker to compare studies and should help eliminate some of the variability in report impressions that can have a drastic effect on patient care,” Castro said. She explained the challenge stems from implanting the technology into a cassette.

Alternatively, the technologist could carry the plate and place it during the exam. With this approach, the equalization/comparison software could be part of any PACS.

Castro called for multi-center study to confirm and determine the extent of the problem. She noted some institutions may have developed internal processes to address the problem. At the same time, vendors need to be aware of the problem, she said.