RSNA: Software automatically & accurately detects abnormal chest findings
CHICAGO—A bag of words model for automatically detecting abnormal x-ray findings based on patterns from previous cases demonstrated strong accuracy in a recent clinical trial, offering a promising method for prioritizing urgent cases and enhancing radiology workflow, according to a study presented Nov. 28 at the 96th annual scientific meeting of the Radiological Society of North America (RSNA).

Increasing imaging orders and a shortage of radiologists, on top of the need to diagnose abnormal emergent findings as quickly as possible, led Michal Sharon, MD, and co-researchers at the Chaim Sheba Health Center in Tel Hashomer to develop a bag of words (BoW) software program to automatically detect abnormal chest findings in emergency room x-rays.

The BoW software functions by breaking down and coding small sections of chest films. The codes are then plotted on a histogram and compared to a database of graphed x-ray findings. The software searches for similarities in the plots to identify abnormal pathologic findings, including pleural effusion, cardiac/mediastinal enlargement and lung opacities, as well as normal x-rays.

The software was used to review 114 consecutive anteroposterior supine chest x-rays from emergency room cases, with a consensus reading by two radiologists serving as the gold standard.

The BoW software detected 38 out of 46 abnormal findings, demonstrating a sensitivity of 83 percent and a specificity of 91 percent. The program's negative- and positive-predictive values were measured at 89 percent and 86 percent, respectively.

The program detected cardiomegaly with pronounced accuracy, yielding a specificity of 85 percent and a sensitivity of 91 percent. Overall detection of individual pathologies was low, however, with sensitivity below 59 percent, while specificity for individual pathologies averaged much higher at 96 to 100 percent.

Sharon highlighted the potential importance of the program for alerting emergency physicians to abnormal findings and accelerating workflow, especially given the high frequency of chest x-rays in emergency rooms.

"We showed that the use of the BoW model for automatic detection of abnormal chest film is feasible," concluded Sharon. "Such software might be used in the future for prioritization of urgent cases in overcharged hospitals, thus improving medical treatment."