Discrimination of Tumorous Intracerebral Hemorrhage from Benign Causes Using CT Densitometry

Discussion

For acute ICH, we investigated the ability of CT densitometry to discriminate between tumors and nontumorous causes of ICH. With NECT only, tumorous and nontumorous ICHs were differentiated with good diagnostic performance by using histogram analysis of acute ICH attenuation. Among the examined histogram parameters, the 25th and 5th percentile values had excellent diagnostic value with area under the curve values above 0.8. The study population was confined to patients with lobar ICH, in whom secondary ICH is relatively frequent and primary lobar ICH is not negligible, accounting for 15%–25% of hypertensive spontaneous ICH,^1,17 making determining a treatment strategy difficult.

Maximum value, unlike the other parameters, was not significantly different among the 5 groups and had poor diagnostic value. This result was because we set an upper limit of 130 HU for thresholding to draw an ROI; maximum values were fixed at 130 HU in case of hematomas adjacent to the skull causing partial volume artifacts. However, these pixels with partial volume artifacts account for a relatively small portion of the hematoma, so the other histogram parameters minimized this problem. Groups showing high attenuation by cumulative histogram parameters, such as groups 1 and 3, showed relatively lower kurtosis. This result might be because highly attenuated acute hematoma had considerable interspersed low-attenuated voxels that were counted on voxelwise histogram analysis. In this context, the higher kurtosis with wider variance seen for group 5 might be attributable to a higher but variable proportion of tumor tissue with lower attenuation within the hematoma. After dichotomization, a higher trend for kurtosis was observed for tumorous than for nontumorous ICH, though this result was not significant.

As shown in Fig 2, tumorous ICH showed a higher relative frequency of hematoma attenuation at lower Hounsfield unit values than nontumorous ICH, leading to positive skewness and relatively lower Hounsfield unit values of tumorous ICHs for all cumulative histogram parameters except the maximum value. One possible explanation for this trend is that a solid tumor lesion within the hematoma might be clustered at lower Hounsfield unit values, though solid tumor lesions are difficult to detect by visual inspection after excluding patients with definitive tumorous ICH with gross cystic and solid lesions. Our receiver operating characteristic analysis showed the 5th and 25th percentile values to be the best parameters for differentiating the attenuation of hematomas between tumorous and nontumorous ICHs. We propose that histogram analysis is a useful tool for detecting hidden solid tumor lesions within acute ICH, which might be helpful in establishing appropriate imaging follow-up and avoiding delays in long-term management of tumorous ICHs. Vascular malformation, another common cause of secondary ICH,^6,7 showed relatively higher attenuation, similar to that of primary ICH, and thus could be discriminated from tumorous ICH. One explanation for this result is that rupture of abnormal vessels is the main pathophysiology of acute bleeding. Therefore, as in primary ICH, a solid lesion occupies a smaller proportion in vascular malformations than in tumors.

In the primary ICH group composed of groups 1 and 2, hematoma attenuation was lower in group 2 with antithrombotics, though this result was significant only for the minimum and 5th percentile values. Five of 13 patients with antithrombotic medication showed blood-fluid levels in our study, and all 5 were on anticoagulants (warfarin). Lower attenuation at minimum and 5th percentile values might be because of this fluid-blood interface, which has been reported to result from uncondensed blood and can be seen in acute ICH of patients treated with anticoagulants within the first 12 hours.^18,19 The lower and heterogeneous attenuation of primary ICH in patients treated with antithrombotics made differentiating primary ICH from tumorous ICH difficult. Therefore, clinicians should consider the possibility of misdiagnosis when examining acute ICH in patients with antithrombotic medication. In patients treated with anticoagulants, the fluid-blood interface uniquely seen in these patients might be helpful in diagnosis.

In routine clinical settings, prognosis and management of secondary ICH depend on the underlying etiology of ICH, while those of primary ICH depend on hematoma growth, initial ICH volume, and expansion into the ventricles.²⁰ Contrast-enhanced CT or MR imaging is performed when secondary ICH, especially tumorous ICH, is suspected. NECT is often undertaken serially to confirm resolution of primary ICH. However, high attenuation on CT and varying T1 intensity on MR imaging for acute hematoma can mask enhancing lesions of ICH on CT and MR imaging.^11,12 Previous studies^13,14 showed that dual-energy CT improves detection of enhancement within highly attenuated hemorrhage, but dual-energy CT or even single-energy contrast-enhanced CT and MR imaging are not always used as the initial imaging technique in patients with acute ICH. Serial follow-up with NECT takes time to confirm complete ICH resolution and might cause diagnostic delay. Being suspicious of tumor bleeding before complete ICH resolution might be helpful in determining appropriate imaging follow-up strategies and avoiding delays in long-term management of tumorous ICH, though acute management would not differ. In this context, our study has value in that NECT is a standard technique initially performed on patients with acute ICH in routine clinical settings. To our knowledge, this is the first study investigating the use of NECT for discriminating tumorous and nontumorous ICHs by using CT densitometry.

This study had several limitations. First, the CT protocols were heterogeneous because we included CT scans obtained at other hospitals in the study population. The voltage used in CT scans obtained at other hospitals was similar to that used for CT at our hospital, which was from 100 to 120 kV. The heterogeneous CT protocols probably did not affect our results because attenuation of a lesion changes depending on the voltage used in CT, and attenuation of the lesions would be similar if the voltages used in CT are similar. In addition, attenuation of solid lesions and hemorrhage does not change substantially as voltage increases, in contrast to the attenuation of iodine and calcification.¹³ A second limitation was that the ROIs in our study were subjectively drawn and could be regarded as cumbersome. However, semiautomatic methods of Hounsfield unit value thresholding has been used in previous studies^16,21 and partially mitigated this limitation. In our experience, it took, on average, 8–10 minutes for an experienced neuroradiologist to segment and analyze a hematoma. Third, only a small number of patients with tumorous ICHs (groups 4 and 5) were included. In addition, the heterogeneity of primary malignancies in group 4 (secondary ICH with brain metastasis) is related to the heterogeneous cellularity of brain metastasis. Hence, it is likely for hypercellular brain metastasis to cause less low attenuation in hematomas and false-negatives in the histogram analysis we proposed. This small number and heterogeneity of patients resulted from the strict exclusion criteria, including multiple hemorrhagic lesions, gross solid or cystic lesions, and the initial CT scan obtained >24 hours after symptom onset because we tried to include clinically confusing cases only. Nonetheless, the strict exclusion criteria can also be a limitation of our study because this study does not reflect the unselected patients who present with acute ICH in the emergency department. Fourth, the number of patients taking antithrombotic medications was relatively small, confined to the primary ICH group. Subsequently, the effect of antithrombotic medication was not fully stratified in all groups. This grouping was because all patients with secondary ICH taking antithrombotic medication were excluded due to other exclusion criteria. Nonetheless, patients taking antithrombotic medication tended to have lower attenuation in acute primary ICH in our study. If this is the case for tumorous ICH, attenuation of tumor ICH with antithrombotic medication would be more likely opposite of expectations for primary ICH; thus, antithrombotic medication would not affect the diagnostic performance of CT densitometry. However, the subgroup of patients taking antithrombotic medications included patients with both antiplatelets and anticoagulants, which also can be a limitation of our study. Patients treated with anticoagulants are more prone to have unclotted components in their hematomas, which might have contributed more to the lower hematoma attenuation, than patients taking antiplatelets.^22,23 Fifth, no validation has been performed in this study, which needs further investigation.