Dual-energy CT in the evaluation of intracerebral hemorrhage of unknown origin: differentiation between tumor bleeding and pure hemorrhage

Abstract

Background and purpose: Detection of underlying tumor in patients with unknown-origin acute ICH may be difficult because acute hematoma may mask enhancement of tumor on postcontrast CT. We intended to investigate the clinical utility of DECT in differentiating tumor bleeding from pure ICH.

Materials and methods: Using a dual-source CT scanner, we obtained TNC single-energy and postcontrast DECT scans for 56 patients with unknown-origin spontaneous ICH. From the 2 sets of postcontrast DECT images obtained with different tube energy, EA (equivalent to conventional postcontrast CT), VNC, color-coded iodine overlay, fusion images of iodine overlay and VNC images were produced. The diagnostic performances of fusion, EA, and combined EA and TNC images for detecting underlying tumors were compared.

Results: Of the 56 patients, 17 had primary or metastatic tumors (18 lesions) and 39 had nontumorous ICH. The sensitivities of fusion, EA, and combined EA and TNC images for detecting brain tumors were 94.4%, 61.1%, and 66.7%, respectively, and their specificities were 97.4%, 92.3%, and 89.7%, respectively. The areas under the ROC curves were 0.964, 0.786, and 0.842, respectively. Overall, the diagnostic performance of fusion images was significantly superior to EA (P = .006) and combined EA and TNC (P = .011) images.

Conclusions: DECT may be useful in detecting underlying tumors in patients with unknown-origin ICH.

Fig 1.

Diagram of image postprocessing of DECT for analysis of hematoma. Enhanced average images, which are equivalent to conventional enhanced images, are produced by mixing and averaging the CT numbers of corresponding voxels at a 7:3 ratio of A and B tube images. Using dual-energy material decomposition software, we separated the iodine component from the images obtained from the A and B tubes; this iodine component is shown in iodine-overlay images and displayed in red. Virtual noncontrast images are generated by removing iodine from the source images. Fusion images are obtained by mixing iodine overlay images and virtual noncontrast images, typically at a ratio of 50:50.

Fig 2.

A 35-year-old man with known renal cell carcinoma and pulmonary metastasis. A, True noncontrast CT shows hemorrhage in the right frontal lobe. B, Enhanced average image obtained by mixing 80- and 140-kV images. A small area of enhancement is suspected in the anterior portion of the hematoma (arrow). Iodine overlay image (C) and virtual noncontrast image (D) obtained by postprocessing of 80- and 140-kV images by using dual-energy material decomposition software. The iodine component is displayed in red in the iodine-overlay image, and the enhancing tumor portion is clearly identified (arrow). High attenuation of hemorrhage is not visualized in this image set, but high attenuation of the bony structure is preserved (C). Virtual noncontrast image (D) shows the hematoma clearly, but the image is smoothed compared with the true noncontrast image in A. E, Fusion image obtained by mixing of iodine-overlay and virtual noncontrast images clearly shows an enhancing mass in the anterior portion of the hematoma (arrow). The brain lesion was histologically confirmed as a hemorrhagic metastasis from renal cell carcinoma.

Fig 3.

Mechanism of differentiating hematoma and iodine. Measured voxel values are plotted on the 2-energy scatterplot. CSF, brain parenchyma, and hematoma have similar CT numbers on both 80- and 140-kV images. In contrast, iodine has nearly a 2 times higher CT number on 80 kV images compared with 140 kV. Therefore, voxels in the area of gray shading represent various amounts of iodine content.

Fig 4.

A 37-year-old woman with known hepatocellular carcinoma. A, True noncontrast CT image shows acute ICH in the left frontal lobe. Part of a separate lesion is noted in the left parietal lobe. B, Enhanced average image at the corresponding level fails to show enhancement in the frontal lobe lesion (arrow). The parietal lobe lesion showed irregular rim enhancement in the main body of the hematoma (not shown here). C, DECT fusion image clearly shows the frontal lobe lesion in red, representing the iodine component (arrow). D, The parietal lobe lesion was surgically removed and confirmed as a metastasis, but the frontal lobe lesion was not treated initially. Gadolinium-enhanced MR imaging performed 1 month later clearly shows the lesion in the left frontal lobe; its size has increased (arrow).

Fig 5.

A false-negative result of DECT in a 48-year-old man with known hepatocellular carcinoma. A, True noncontrast CT shows a hematoma in the right occipital lobe. Small dotlike enhancement is noted in the medial aspect of hematoma on an enhanced average image (B) and a fusion image (C). However, no definite mass lesion is identified. D, On the enhanced T1-weighted image, some tubular enhancement is noted (arrow), but a tumorous lesion is not definite. E and F, One-month follow-up axial and sagittal postcontrast MR images show multiple enhancing nodules (arrows) around the hematoma (arrowheads). These strongly suggest the possibility of previous ICH caused by underlying metastatic tumor.

Fig 6.

A false-positive result of DECT in a 50-year-old man with right-sided weakness commencing 3 days earlier. A, True noncontrast CT shows a hematoma in the left frontal lobe superior frontal gyrus. A thin irregular rim-enhancing lesion is shown on the enhanced average image (B) and fusion image (C) (arrows). D, T2-weighted sagittal image shows a multilocular dark-signal-intensity lesion mixed with high signal intensity with a dark-signal-intensity rim, suggesting multiple-stage hemorrhage. On the basis of images B and C, a hemorrhagic tumor was suspected, but the diagnosis was changed after reviewing MR imaging, which suggested the possibility of a cavernous malformation. Diagnosis of a cavernous malformation was confirmed by surgery.

Fig 7.

ROC curves for fusion, EA, and EA + TNC images. Fusion images show the highest performance with an AUC of 0.964.