Craniopharyngioma and meningioma are two of the most common benign tumors in the sellar or parasellar area. Craniopharyngioma presents approximately 2.5%–4% of the brain tumor. It can be detected at any age; besides, it is the overwhelming major tumor (approximately 90%) of the pituitary region neoplasms in children [
Texture analysis (TA) is a method to describe the voxel-value frequency distribution and the spatial organization of voxel value, which can reflect how each voxel value differs from the neighbor voxel values. During the analysis process, several matrices are used to capture information from clinical images, and each matrix enables the calculation of several heterogeneity descriptors [
We retrospectively searched our institution database to identify all qualified patients. Eligibility criteria for qualified patients were (1) histopathological confirmation; (2) elaborate electrical medical records; (3) diagnostic MR scan before the operation; and (4) tumors in the sellar/parasellar area. Exclusion criteria were (1) history of treatments before the MR scan; (2) history of intracranial disease (e.g., brain trauma, intracranial infection, or other types of brain tumor); and (3) patients with a recurrent brain tumor considering the interference of scar tissue. Finally, 127 qualified patients with craniopharyngioma (
MRI was available in all patients, including contrast-enhanced T1-weighted sequences and T2-weighted sequences. The MR device used is the 3.0 T Siemens Trio Scanner. Contrast-enhanced T1-weighted imaging used gadopentetate dimeglumine (0.1 mmol/Kg) as the contrast agent, acquiring axial, coronal, and sagittal data. The scanning of dynamic enhanced MRI was conducted within 250 s after injection of the contrast agent. Among the 127 patients enrolled in this study, the contrast-enhanced images were available in all patients, while images of T2-weighted imaging (T2WI) were available among 114 patients.
Two neurosurgeons reviewed all MRI scans to extract qualitative MR images features under the supervision of a senior radiologist and a senior neurosurgeon, with whom disagreements were solved by discussion and consultation. We evaluated the following qualitative MR images features based on the clinical experience and the previous studies: (1) signal intensity compared with normal tissue on contrast-enhanced images and images of T2WI, (2) heterogeneity on contrast-enhanced images and images of T2WI, (3) unenhanced area (s), (4) cystic alteration (s) on contrast-enhanced images or images of T2WI, and (5) the presence of air-fluid level. Besides, the size and the location of tumor tissue were also measured and recorded [
LifeX is medical software which reads medical images locally and characterizes tumor heterogeneity. Two neurosurgeons utilized LifeX package (
All statistical analyses were conducted using IBM SPSS Statistics for Windows, Version 22.0 (IBM Corp. Armonk, NY, USA) and MedCalc statistics (MedCalc Software bvba, Acacialaan, Belgium). In statistics processing, we summarized variables based on their classification, the continuous with means and ranges, while the categorical with frequencies and percentages. For the clinical, radiological, and histopathological features, the significant difference between meningioma and craniopharyngioma was examined first with chi-square tests (for categorical variables with enough statistics), Fisher exact tests (for categorical variables with limited statistics), and the Mann–Whitney
As for the texture features, the Mann–Whitney
Finally, the texture features were analyzed with the Mann–Whitney
The characteristics of patients and lesions are summarized in Table
Characteristics of the patient and lesion.
Character | Craniopharyngioma | Meningioma |
---|---|---|
Gender | Male: 37 (58.7) | Male: 18 (28.1) |
Female: 26 (41.3) | Female: 46 (71.9) | |
Age (years) | 31.62 (2∼73) | 49.19 (9∼72) |
Tumor size (mean ± SD (mm)) | 28.86 ± 9.57 | 20.41 ± 5.96 |
Location | Intrasellar: 0 | Intrasellar: 0 |
Intra- and suprasellar: 17 | Intra- and suprasellar: 8 | |
Suprasellar: 46 | Suprasellar: 56 | |
Dural tail sign | 0 | 62 |
Among the seven MR images features we analyzed, four of them were found to be significantly different between craniopharyngioma and meningioma (the
The differences in MR images features between craniopharyngioma and meningioma. Entries in bold were significant.
Qualitative MR features | Craniopharyngioma |
Meningioma |
| |
---|---|---|---|---|
Signal intensity on contrasted images | Hypointense | 0 (0) | 2 (3) | 0.149 |
Isointense | 0 (0) | 0 (0) | ||
Hyperintense | 40 (63) | 32 (50) | ||
Extreme hyperintense | 23 (37) | 30 (47) | ||
Heterogeneity on contrasted images | Homogenous | 7 (11) | 7 (11) | 0.975 |
Heterogeneous | 56 (89) | 57 (89) | ||
Unenhanced area (s) | Presence | 50 (79) | 6 (9) |
|
Absence | 13 (21) | 58 (91) | ||
Signal intensity on T2WI | Hypointense | 0 (0) | 1 (2) |
|
Isointense | 6 (10) | 46 (81) | ||
Hyperintense | 14 (25) | 10 (17) | ||
Extreme hyperintense | 37 (65) | 0 (0) | ||
Heterogeneity on T2WI | Homogenous | 10 (18) | 39 (68) |
|
Heterogeneous | 47 (82) | 18 (32) | ||
Cystic alteration (s) | Presence | 58 (92) | 5 (8) |
|
Absence | 5 (8) | 59 (92) | ||
Air-fluid level | Presence | 7 (11) | 0 (0) | 0.006 |
Absence | 56 (89) | 64 (1) |
T2WI: T2-weighted imaging.
Examples of two cases from the MR images in patients with craniopharyngioma and meningioma. (a) Contrast-enhanced images with craniopharyngioma, (b) a contrast-enhanced image with meningioma, (c) images of T2WI with craniopharyngioma, and (d) an image of T2WI with meningioma.
According to the Mann–Whitney
Boxplot of five independent texture features: (a) HISTO-Skewness, (b) GLCM-Contrast, and (c) GLCM-Dissimilarity on contrast-enhanced images; (d) HISTO-Skewness and (e) GLCM-Contrast on images of T2WI in discriminating craniopharyngioma and meningioma. Craniopharyngioma showed higher HISTO-Skewness, GLCM-Contrast, GLCM-Dissimilarity on contrast-enhanced images, and GLCM-Contrast on images of T2WI, but lower HISTO-Skewness on images of T2WI than craniopharyngioma.
The binary logistic regression on texture features between craniopharyngioma and meningioma.
Texture feature |
|
OR | 95% CI | |
---|---|---|---|---|
Contrast-enhanced images on T1WI | HISTO-skewness |
|
0.410 | 0.242–0.693 |
GLCM-contrast |
|
0.087 | 0.009–0.863 | |
GLCM-dissimilarity | 0.145 | 4.637 | 0.588–36.560 | |
Images of T2WI | HISTO-skewness |
|
2.458 | 1.534–3.940 |
GLCM-contrast | 0.086 | 0.635 | 0.378–1.066 |
Entries in bold were significant. HISTO: histogram-based matrix, GCLM: grey-level co-occurrence matrix, T1WI: T1-weighted imaging, T2WI: T2-weighted imaging, OR: odds ratio, CI: confidence interval.
ROC curves were only performed in three independent predictors (HISTO-Skewness, GLCM-Contrast on contrast-enhanced images, and the HISTO-Skewness on T2WI). AUC of these texture features were all higher than 0.700, which presented their practical value in contrastive analysis. The outcomes are presented in Figure
Receiver operating characteristic (ROC) curves of (a) GLCM-Contrast, (b) HISTO-Skewness on contrast-enhanced images, and (c) HISTO-Skewness on images of T2WI demonstrated promising diagnostic value of the three texture features, of which area under curves (AUC) were all more than 0.700. (d) ROC curves of an integrated model combining GLCM-Contrast and HISTO-Skewness on contrast-enhanced images showed more value in practical diagnosis with higher AUC.
Diagnostic performance of texture features for differentiating craniopharyngioma from meningioma.
Texture parameter | AUC | Standard error | 95% CI | Cutoff point | Sensitivity | Specificity | |
---|---|---|---|---|---|---|---|
Contrast-enhanced images on T1WI | HISTO-Skewness | 0.700 | 0.0491 | 0.612∼0.778 | 0.648 | 87.50 | 55.56 |
GLCM-Contrast | 0.711 | 0.046 | 0.624∼0.788 | 29.444 | 64.06 | 73.02 | |
|
0.776 | 0.043 | 0.693∼0.845 | 0.093 | 79.69 | 69.84 | |
Images of T2WI | HISTO-Skewness | 0.713 | 0.050 | 0.612∼0.793 | −0.308 | 74.14 | 68.97 |
HISTO: histogram-based matrix, GCLM: grey-level co-occurrence matrix, T1WI: T1-weighted imaging, T2WI: T2-weighted imaging, AUC: area under the curve, CI: confidence interval.
Considering the diagnostic value of a single texture feature was not good enough to be taken as a practical parameter, the integrated model was built based on the results of binary logistic regression. The formula of the model is
The ROC curve shown in Figure
According to the Mann–Whitney
Receiver operating characteristic (ROC) curves of (a) GLCM-Contrast, (b) HISTO-Skewness on contrast-enhanced images, and (c) HISTO-Skewness on images of T2WI demonstrated MR images features and texture features were related to each other.
Clinically, there still remained to be a challenge in the contrastive analysis of craniopharyngioma and meningioma in the sellar/parasellar area because craniopharyngiomas may radiologically mimic meningiomas [
Previous studies had shown us a lot about the MR images features of craniopharyngioma and meningioma. Craniopharyngioma is typically a solid-cystic, lobular tumor with calcareous concretions of the intra- and/or suprasellar region, [
Texture analysis has been applied to improving the accuracy in classifying and grading meningioma, but texture analysis on craniopharyngioma has not been reported yet [
Previous studies determined that craniopharyngioma is typically a cystic tumor and the cystic alteration could be regarded as the statistical demarcations between craniopharyngioma and meningioma. Meanwhile, variant protein concentration within the cystic fluid could result in variant signal intensity in MRI. Thus, we carried out an analyzation to detect the possible relationship between texture features and cystic alteration [
Our study had several limitations. First, as a retrospective study, we only included patients with surgically resectable tumors. Second, the potential for selection biases could not be excluded. Third, we were unable to assess other subsequences, especially conventional T1WI. Fourth, the differences in tumor subtypes were not taken into consideration because of the limited number of patients.
MR images features (cystic alteration) and texture features (HISTO-Skewness and GLCM-Contrast on contrast-enhanced images and HISTO-Skewness on images of T2WI) were useful in the contrastive analysis between craniopharyngioma and meningioma. Besides, the two types of features were related to each other. But, more studies are required to verify our results and rectify the defects.
The data used to support the findings of this study are available from the corresponding author upon request.
The authors declare that they have no conflicts of interest.
Zerong Tian and Chaoyue Chen contributed equally to this work.
Supplementary Table 1: the explanation of the selected features. Abbreviations: HISTO: histogram-based matrix; GCLM: grey-level co-occurrence matrix.