Computer aided detection (CADx) programs offer lots of potential to help radiologists avoid missing important findings, but the benefits vary between applications. Two recent studies, however, have given high marks to CADx programs for identifying pulmonary embolisms (PEs) and breast lesions.
The first study, published in the January issue of the American Journal of Roentgenology, tested a prototype PE-CADx program (Philips Healthcare) for its ability to spot PEs missed on initial interpretation. Researchers from the University of Maryland School of Medicine in Baltimore retrospectively assessed CT angiography studies in order to find 53 studies with at least one overlooked acute PE.
The PE-CADx program demonstrated a high effectiveness, finding a PE in 77.4 percent of cases, including correctly marking at least one PE in all 23 cases with multiple PEs missed on initial interpretation.
Concern about false-positives has dampened enthusiasm for CADx in the detection of PEs, and the current study did have some of these as well. An average of 3.8 false-positive marks per case were present, though the authors noted that a skilled interpreter would likely be able to recognize and dismiss these marks. Also, results showed that higher image quality lessened the likelihood of false-positives.
Shifting from detection of PE to breast lesions, the second study, published online Dec. 17 in Magnetic Resonance Imaging, showcased the benefits of combining kinetic information and morphological features of 3D breast MRI to create an algorithm for CADx.
Researchers from Taipei City Hospital and National Taiwan University developed the algorithm for tumor segmentation and characterization that included an integrated color map that combined kinetic and area under the curve color maps to detect potential lesions. Morphological features including shape and texture where acquired with dynamic contrast-enhanced MRI. Volume of plasma, energy, rate constant, entropy and compactness were the five features that, when combined, provided the algorithm with the best performance.
The authors evaluated the algorithm using a cohort featuring 63 benign and 69 malignant biopsy-proven lesions. Results showed an accuracy of 91.7 percent, sensitivity of 91.3 percent and a specificity of 92.1 percent.
These studies provide more examples of the power of CADx programs as tools that can be incorporated into daily workflow to provide an assist to interpreters.
Editor – Health Imaging