Imaging Angiogenesis Genotype of Glioblastoma by Radiomic Features of Multi-modality MRI
Chia-Feng Lu1,2,3, Fei-Ting Hsu4, Li-Chun Hsieh4, Yu-Chieh Jill Kao1,2, Hua-Shan Liu4,5, Ping-Huei Tsai2,4, Pen-Yuan Liao4, and Cheng-Yu Chen1,2,4

1Translational Imaging Research Center, College of Medicine, Taipei Medical University, Taipei, Taiwan, 2Department of Radiology, School of Medicine, Taipei Medical University, Taipei, Taiwan, 3Department of Biomedical Imaging and Radiological Sciences, National Yang-Ming University, Taipei, Taiwan, 4Department of Medical Imaging, Taipei Medical University Hospital, Taipei, Taiwan, 5Graduate Institute of Clinical Medicine, Taipei Medical University, Taipei, Taiwan

Synopsis

The multi-modality and multi-radomic-feature MRI may provide a more efficient regression model for imaging gene expressions than the conventional radiogenomic approach.

Background and Purpose

Glioblastoma (GBM) is one of the most common and aggressive malignant brain tumor resulting in a poor prognosis after diagnosis. Recently, imaging gene expression of GBM by using the clinical MRI shows the possibility to noninvasively probe the tumor characteristics.1 Previous studies showed that the perfusion maps of cerebral blood volume (CBV) can be a surrogate image biomarker of angiogenesis, the hallmark of GBM that generates blood vessels and facilitates tumor progression.2,3 In addition to the hemodynamic CBV, angiogenesis-related physiologic changes in GBM, such as the vascular leakiness, cellularity, and tumor appearance, can be characterized by several other MR modalities. This study aims to determine whether the multivariate analyses based on the multi-modality MRI can further improve the efficacy in assessing the gene expression by using the MR radiomic features.



Materials and Methods

This study is approved by the local Institutional Review Board and the written informed consents were given before the patient enrollment. Four patients (2 men and 2 women) with primary GBM were recruited in this study. Preoperative clinical MR data, including post-contrast T1-weighted (T1+C), T2 FLAIR, dynamic susceptibility contrast (DSC) perfusion-weighted, and diffusion-weighted imaging (DWI), with default parameters were acquired from each patient using a 1.5 T MRI scanner (GE Signa HDxt, USA) in Taipei Medical University Hospital. Relative CBV (rCBV) maps with correction for vascular leakage and K2 permeability maps were derived from DSC images,4 and the apparent diffusion coefficient (ADC) was calculated from DWI. Three site-specific tissue specimens located in tumor enhancement, necrosis, and edema respectively were extracted from each patient using MRI-based stereotaxic biopsy before surgical resection. Three of overall 12 samples were excluded from the subsequent microarray RNA profiling because of the insufficient RNA quality. Three hundred and thirty-eight genes annotated as positive regulation of angiogenesis function based on the Gene Ontology database5 (GO term, GO:00001525) were selected (Table 1), and the normalized log2 mean of the selected genes was used to measure the gene expression.

In order to acquire the site-specific imaging features from different MR modalities, image co-registration and interpolation reference to the T1+C images were first employed to ensure the concordance of imaging location and spatial resolution (0.43 x 0.43 x 6.00 mm3) between imaging modalities. The region of interests (ROI) in tumor enhancement, necrosis, and edema were then determined respectively by manual delineations with intensity thresholding on T1+C and T2 FLAIR images (Fig.1). Three categories, including intensity-based, geometry-based, and textural features, with overall 55 quantitative features can be derived from each ROI and in each imaging modality based on the previous described radiomics approach.6 Finally, the multivariate linear regression was applied to estimate the fitting efficacy of gene expression based on the radiomic features of MRI modalities, including T1+C, rCBV, K2, and ADC maps. The coefficient of determination, R2, was calculated to assess how well the imaging features of multi-modality MRIs fit the gene expression.

Results & Discussion

New vessel formations during tumor angiogenesis increase local blood supply to promote tumor proliferation. These rapidly formed vessels are characterized by the disrupted structure of blood brain barrier,7 leading to the vascular leakage revealed by the hyperintensity in T1+C images or K2 maps. After the co-registration between imaging modalities, the enhancing ROI defined by the hyperintensity tumor regions in T1+C images showed high cellularity depicted by low ADC values and high rCBV values (Fig. 1). The necrotic regions exhibited opposite phenomena with low cellularity (high ADC values) and low rCBV values (Fig. 1). The results of multivariate linear regressions demonstrated that using all 4 modalities (i.e., T1+C, rCBV, K2, and ADC) as regressors can identify the best fitting model with the highest coefficients of determination, R2, in estimating gene expression by imaging features (Fig. 2). This finding suggested that using multi-modality MRI associated with physiologic changes in angiogenesis (Fig. 2E) can better predict the relevant gene expressions than using single parametric CBV map (Fig. 2B). Table 2 lists the number of genes that exhibited significantly improved R2 values based on the multi-modality approach for each radiomic category. It is worth pointing out that the utilized radiomic features can also be a critical factor on fitting efficacy. Both intensity-based and textural features achieved sufficient R2 and significant R2 improvement using multi-modality MRI approach in assessing gene expressions (Fig. 2 and Table 2). Future studies with larger sample size are warrant to investigate the benefit of combing multi-modality and multi-radiomic-feature MRI in imaging genotype in GBM.

Acknowledgements

This study was funded in part by the Taipei Medical University (TMU103-AE1-B20) and the Ministry of Science and Technology (MOST 104-2221-E-038-007, MOST 104-2314-B-038 -051 -MY3), Taipei Medical University Hospital (104TMU-TMUH-23, 104TMUH-SP-02), and Health and Welfare Surcharge of Tobacco Products supported for Comprehensive Cancer Center of Taipei Medical University (MOHW104-TDU-B-212-124-001), Taipei, Taiwan.

References

1. Jamshidi N, Diehn M, Bredel M, et al. Illuminating radiogenomic characteristics of glioblastoma multiforme through integration of MR imaging, messenger RNA expression, and DNA copy number variation. Radiology, 2013;270(1): 1-2.

2. Tykocinski E S, Grant R A, Kapoor G S, et al. Use of magnetic perfusion-weighted imaging to determine epidermal growth factor receptor variant III expression in glioblastoma. Neuro-oncology, 2012;14(5):613-623.

3. Jain R, Poisson L, Narang J, et al. Correlation of perfusion parameters with genes related to angiogenesis regulation in glioblastoma: a feasibility study. American Journal of Neuroradiology, 2012;33(7):1343-1348.

4. Bjornerud A, Sorensen A G, Mouridsen K, et al. T1-and T2*-dominant extravasation correction in DSC-MRI: Part I—theoretical considerations and implications for assessment of tumor hemodynamic properties. Journal of Cerebral Blood Flow & Metabolism, 2011;31(10):2041-2053.

5. Ashburner M, Ball C A, Blake J A, et al. Gene Ontology: tool for the unification of biology. Nature genetics, 2000;25(1):25-29.

6. Aerts H J, Velazquez E R, Leijenaar R T, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nature communications, 2014;5.

7. Gilbert M R. Renewing interest in targeting angiogenesis in glioblastoma. The Lancet Oncology, 2014;15(9):907-908.


Figures

Table 1 The selected genes with up-regulated angiogenesis function

Figure 1 The co-registered multi-modality MRI, including the (A) post-contrast T1-weighted image (T1+C), (B) T2 FLAIR image, (C) apparent diffusion coefficient (ADC) map, (D) leakiness-corrected relative cerebral blood volume (rCBV), and (E) K2 permeability map, for a patient with primary GBM. The selected ROIs of tumor enhancing (red), necrotic (blue), and edema (green) in (F) were determined by manual delineations with intensity thresholding on T1+C and T2 FLAIR images.


Figure 2 The coefficients of determination, R2, for assessing the fitting efficacy of gene expression by using the MRI modalities of (A) T1+C; (B) rCBV; (C) T1+C, rCBV; (D) T1+C, rCBV, K2; (E) T1+C, rCBV, K2, ADC. Each row of matrix represents a selected gene and each column stands for a radiomic feature belonging to one of the three categories, (a) intensity-based, (b) geometry-based, and (c) textural features.




Table 2 The number of genes with improved fitting efficacy estimated
by multi-modality MRI compared to that by the rCBV map



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1348