The aim of this study to evaluate the role of frequency-selective nonlinear blending (FS-NLB) for the detectability of brain metastases with contrast-enhanced computed tomography (CECT) using magnetic resonance imaging (MRI) as standard of reference.
A retrospective patient data search at our institution yielded 91 patients who underwent both brain CECT and MRI for screening of brain metastases (n = 173) between 2014 and 2016 (mean time interval, 29 ± 37 [malignant: 15 ± 16/benign: 42 ± 47] days). A recently introduced FS-NLB postprocessing technique was applied to CECT images. Two readers interpreted all CT images in an independent fashion. The conventional, linear blending (LB) CT images were evaluated first. After a washout period, the same readers evaluated the FS-NLB CT images. The standard of reference was established by a consensus interpretation of the brain MRI studies. Outcome variables included determination of best performing FS-NLB settings, region of interest (ROI)–based calculation of contrast-to-noise ratios (CNRs), size, and number of brain metastases. Based on the number of metastases, we classified patients in 5 therapeutically relevant categories (0, no metastasis; 1, singular metastasis; 2, less than 4 metastases; 3, >4 and <10 metastases; 4, >10 metastases). Statistical comparison and diagnostic performance tests were applied.
A center of 47 Hounsfield units (HU), delta of 5 HU, and slope of 5 resulted in the best delineation of hyperdense brain metastases, whereas for hypodense brain metastases, a center of 32 HU, delta of 5 HU, and slope of 5 showed best delineation. Frequency-selective nonlinear blending significantly increased CNR in hyperdense cerebral metastases (CECT: 9.11 [6.9–10.9], FS-NLB: 18.1 [11.9–22.8]; P < 0.0001) and hypodense cerebral metastases (CECT: 6.3 [5.2–8], FS-NLB: 17.8 [14.5–19.7]; P < 0.0001). Sensitivity, specificity, negative predictive values, positive predictive values, and accuracy for LB, and FS-NLB were 40%, 98%, 99%, 31%, and 52%, and 62%, 94%, 97%, 40%, and 69%, respectively. Magnetic resonance imaging, LB, and FS-NLB classification of metastatic patients were group 0 (47, 47, 46), group 1 (14, 8, 11), group 2 (16, 12, 15), group 3 (8, 7, 8), and group 4 (6, 4, 6).
Frequency-selective nonlinear blending postprocessing of CECT significantly increases the detection of brain metastases over conventional CECT; however, the sensitivity remains lower than MRI. Frequency-selective nonlinear blending is slightly inferior in the categorization of patients into therapeutically relevant groups, when compared with MRI.
From the Departments of *Diagnostic and Interventional Radiology, and
†Diagnostic and Interventional Neuroradiology, University Hospital of Tübingen, Tübingen, Germany; and
‡Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, MD.
Received for publication July 2, 2018; and accepted for publication, after revision, August 15, 2018.
Conflicts of interest and sources of funding: Jan Fritz received institutional research support from Siemens Healthcare USA, DePuy, Zimmer, Micorsoft, and BTG International. Jan Fritz is a scientific advisor of Siemens Healthcare USA, Alexion Pharmaceuticals, and BTG International. Jan Fritz received speaker's honorarium from Siemens Healthcare USA. Jan Fritz has shared patents with Siemens Healthcare and Johns Hopkins University. Marius Horger received institutional research funds and speaker's honorarium from Siemens Healthineers, and is a scientific advisor of Siemens Healthcare GmbH Germany. Georg Bier holds shares of Bayer AG. There was no funding received for this study.
Correspondence to: Malte N. Bongers, MD, Department of Diagnostic and Interventional Radiology, University Hospital of Tübingen, Hoppe-Seyler-Str. 3, 72076 Tübingen, Germany. E-mail: firstname.lastname@example.org.