In a study involving 200 negative and 40 positive TB cases read by five radiologists using an iPad 2 and an LCD monitor, the researchers found no significant differences between diagnostic decisions made on the two displays.
"Mobile displays should provide an effective extension of the radiologist's reach," said medical student Samir Abboud. He presented the research at the RSNA 2011 meeting in Chicago.
In previous pilot research presented at the RSNA 2010 meeting, the University of Maryland team had initially reviewed 30 images in JPEG format on an Apple iPad for the purpose of screening for TB. The current work sought to more thoroughly assess the amount of variability among readers using the same display, the agreement between the consensus of readers for a traditional display versus a mobile display, and the degree to which the readings by an individual reader agreed on both displays.
The team sampled a database of more than 500 cases of TB screening, and randomly selected 200 cases that were called negative and 40 that were called positive, Abboud said. The studies were originally interpreted by a chest fellowship-trained attending radiologist.
The DICOM images were stripped of identifying data and loaded into a viewer in random order. Five radiologists (two chest fellowship-trained attending physicians, two chest fellows, and one fourth-year resident) were asked to characterize the conventional radiographs as either positive or negative based on the suspicion of TB using Apple's iPad 2 with OsiriX HD viewing software and a 27-inch iMac monitor with OsiriX viewing software. They were told to ignore any other suspected findings, Abboud said.
For each subject, the display (either iPad 2 or workstation) that was viewed first was randomly selected. To reduce recall bias, retests on the same image for the alternate display were given at a minimum of one week apart. Reading on both the iPad 2 and the workstation was performed in the same area of the reading room to minimize ambient light variability.
The researchers used mean proportion of agreement (averaged over all images for both displays) to measure variability among readers using the same display. They also determined the proportion of complete agreement for images classified as positive or negative.
The team also calculated both the agreement between the consensus of readers (defined as three or more readers) and the degree to which an individual reader agrees with his or her findings on both displays as a proportion of agreement.
|Variability among readers using same display
|Mean proportion of agreement
|Proportion of cases with complete agreement
|Proportion of positive cases with complete agreement
|Proportion of negative cases with complete agreement
The agreement among the five readers was statistically equal using either display, Abboud said.
In evaluating the agreement between reader consensus for the traditional display versus an iPad, the researchers found that only two cases did not have agreement. One case was called positive on the iPad but negative on the workstation, and one case was called negative on the iPad and positive on the workstation, Abboud said.
- Overall agreement between traditional display and iPad: 0.9916
- Positive-case agreement: 0.9744
- Negative-case agreement: 0.9950
"The consensus of the five readers of the iPad agreed with the consensus of the five readers for the traditional display," Abboud said.
As for the degree to which the readings by an individual reader agreed on both displays, the average overall proportion of agreement was 0.965. The proportion of agreement was 0.8944 on positive images and 0.9790 on negative images.
"The readers basically agreed with themselves well using either display," he said.
On the downside, many readers found the reading experience on the iPad to be a lot slower, Abboud said.
"It took, on average, twice as long to view the dataset on the iPad than on the traditional workstation," he said. "Readers also commented on how much smaller the iPad's screen was compared to the 27-inch monitor."
Overall, however, no significant differences were detected between decisions made on the iPad and the LCD, he said.