Subtle changes to imaging acquisition parameters can dramatically affect the accuracy of computer-aided detection systems. A pair of studies conducted by researchers at the University of Maryland and the Baltimore VA Medical Center examined two such acquisition parameters: optimal CT reconstruction convolution kernels and the effect of slice thickness.
Subtle changes to imaging acquisition parameters can dramatically affect the accuracy of computer-aided detection systems. A pair of studies conducted by researchers at the University of Maryland and the Baltimore VA Medical Center examined two such acquisition parameters: optimal CT reconstruction convolution kernels and the effect of slice thickness.
While more studies evaluating the effect of CAD on image interpretation are presented every year at the RSNA meeting, surprisingly little attention has been accorded the acquisition parameters themselves, said Dr. Eliot Siegel, chief of radiology and nuclear medicine at the VA Maryland Health Care System.
The parameters that could possibly have a major impact on CAD results include slice thickness, mAs and kVp values, postprocessing filters, and reconstruction kernels.
Siegel's talk Tuesday morning focused on the effect that two reconstruction kernels, soft tissue and bone, would have on lung nodule detection.
The soft-tissue kernel provides a smoother image with increased contrast and decreased noise, but poor edge detection, he said. The bone kernel, on the other hand, provides improved spatial resolution, increased noise, and better edge detection.
Siegel and colleagues tested the two reconstruction parameters on 21 chest CT exams at 120 kVp, 12 mAs, and 0.75 collimation. After using the CAD algorithm to detect nodules, truth was established by a consensus of thoracic radiologists.
Nodules that were determined to be actionable ranged between 3 mm and 3 cm in size. The investigators detected a total of 34 actionable lung nodules. While both kernels had the same 77% sensitivity, the bone kernel was associated with significantly fewer (2.4) false positives compared with the soft tissue kernel (2.8).
"The superior performance of the bone kernel suggests that it should be the preferred technique for research and perhaps even clinical studies," Siegel said.
Siegel's colleague, Dr. Bruce Reiner, director of radiology research at the VA Maryland Health Care System, presented information from the sister study on collimation parameters. That study used the same acquisition parameters as the reconstruction kernel study, but collimation was set at 0.75 mm, 1.5 mm, and 3 mm. The same 21 chest CT exams were studied.
The greatest sensitivity and specificity were seen at the 0.75-mm collimation slice thickness and the lowest at 3-mm collimation - an unexpected finding, Reiner said.
The number of false positives increased with an increase in slice thickness, which can lead to higher costs, increased morbidity, and decreased productivity, he said.
"The accuracy of CAD is dependent on the CT reconstructed slice thickness, and it is best to use narrow collimation and thin reconstruction," Reiner said.
For more information from the online Diagnostic Imaging archives:
CAD products track changes in lung nodules
Questions linger on clinical value of breast CAD
Study Reaffirms Low Risk for csPCa with Biopsy Omission After Negative Prostate MRI
December 19th 2024In a new study involving nearly 600 biopsy-naïve men, researchers found that only 4 percent of those with negative prostate MRI had clinically significant prostate cancer after three years of active monitoring.
Study Examines Impact of Deep Learning on Fast MRI Protocols for Knee Pain
December 17th 2024Ten-minute and five-minute knee MRI exams with compressed sequences facilitated by deep learning offered nearly equivalent sensitivity and specificity as an 18-minute conventional MRI knee exam, according to research presented recently at the RSNA conference.
Can Radiomics Bolster Low-Dose CT Prognostic Assessment for High-Risk Lung Adenocarcinoma?
December 16th 2024A CT-based radiomic model offered over 10 percent higher specificity and positive predictive value for high-risk lung adenocarcinoma in comparison to a radiographic model, according to external validation testing in a recent study.