Tumor Classification Using Blood Arrival Histogram Obtained by Resting-state fMRI
Tianyi Qian1, Yinyan Wang2,3, Kun Zhou4, Yuanyuan Kang4, Shaowu Li2,5, and Tao Jiang2,5

1MR Collaborations NE Asia, Siemens Healthcare, Beijing, China, People's Republic of, 2Neurosurgery, Tiantan Hospital, beijing, China, People's Republic of, 3Beijing Neurosurgical Institute, Capital Medical University, Beijing, China, People's Republic of, 4Siemens Shenzhen Magnetic Resonance Ltd., APPL, Shenzhen, China, People's Republic of, 5Beijing Neurosurgical Institute, Capital Medical University, beijing, China, People's Republic of


In this study, a new post-processing pipeline of resting-statefMRI (rs-fMRI)was proposed for glioma grading, with the feasibility of extracting the timing information of brain perfusion from BOLD signal. The blood arrival time obtained from rs-fMRI shows unevenly distributed perfusion patterns in tumors. A histogram-based analysis method was employed to analyze the non-uniform distribution that could extract the patterns better than the routine method. The proposed pipeline was able to classify between low- and high-grade gliomas.


Perfusion patternsare important clinical indexes for glioma grading1.However, dynamic susceptibility contrast MRrequires the use of exogenous contrast agents.As a non-invasive method, resting-state fMRI (rs-fMRI) was used to identify functional brain regions in brain malignancy lesions. By reflecting the oxygen level in blood fluid, the signal of resting-state fMRI may be also associated with the histopathological classification of the tumor and further guide the adjuvant therapy of brain gliomas2.Here we demonstrate a new post-processing pipeline of rs-fMRI applied for glioma grading, with the feasibility ofextracting the timing information of brain perfusion from BOLD signal. Previous methods, based on averaging theROI values,could notfully demonstrate the non-uniformed tumor structure.As a consequence, a histogram-based analysis method was employed in this study to extract the best perfusion patterns for glioma classification.


A total of 65 patients with histologically confirmed gliomas were included in this study. The MRI exam consisted of both a routine pre-surgical protocol and rs-fMRI. All data was collected on a MAGNETOMPrisma3T MR scanner (Siemens AG, Erlangen, Germany). The parameters of rs-fMRI were as follows: TR=2500 ms, TE=30 ms, flip angle=90°, 43 slices, slice thickness=3 mm, distance factor=20%, FOV=210 ×210 mm2, matrix= 70×70, measurements=100. The fMRI data was pre-processed with a standard pipeline for resting-state data analysis, without regressing out the global signal. After pre-processing, the blood arrival map was obtained using the following steps: 1) Averaging the time series of the whole brain to create the first time series template. 2) For each voxel, the time course was shifted from -6TR to +6TR and correlated with the template at each TR. Each voxel was then labeled as the number of TR thathad the maximal correlation coefficient value. 3) Realigning the time series of all voxels based on their relative blood arrivaltime determined by step2. 4) Averaging the re-aligned time series of the whole brain to create a new global template. 5) Repeating steps 2 and 3 until the number of voxel that had changed their blood arrival value between two iterations is less than 100. The blood arrival time of each subject wasthen converted to z-scores for group analysis. Lesions were manually identified by two neuroradiologists based on T2-weighted MR images from each patient.Thehistogram pattern (HP)wasdefined as the probability density distribution(pdf) of z-score within Lesion ROI after subtraction of the pdf of the corresponding area in the other hemisphere. The control ROI in the All HP were then averaged among each grade (WHO I/II/III/IV) to find the best z-score range for classification. The mean value of different ranges in HPwere then used to test the significance of group differences between low- and high-grade glioma.


Figure 1 shows a case with WHO IV glioma. In the blood arrival time map, red-yellow regions represent early arrivalsand purple regions represent late ones. From the figure, we can see that the arrival time within ROI is significantly earlier than the contralateralarea and the distribution of the early arrival is temporally heterogeneouswithinthe tumor. The average HP of each grade demonstratesa trend of more early perfusion arrival areas and fewerlate arrival areas along with the grade increase (shown in Fig.2). The histogram differences in the range of 2-5 (in z-score) have the most significant patterns for classifying low- and high-grade glioma (p=0.002). Fig.3 shows the box plotof the sum percentage of the BAH in this range within the lesion ROI. There were no significant differences (p>0.05) between LGG and HGG whenaveraging the z-score of the whole lesion area.


High-grade gliomas have higher blood flow than low-grade ones.Although the perfusion pattern could be measured by DSC-MR, a contrast agent-free method could provide a more flexible protocol, especially if the protocol includesa DCE scan before a late-enhancement measurement. Tumor structure is not uniform and usually contains several different grade tumor cells. So simply averaging the perfusion pattern in the lesion area may not reflect the full pattern. The pdf differences between control and lesion areas could more clearly show the perfusion pattern changes and provide more useful information to decide the highest grade of tumor cells in the glioma.


The timing information of blood perfusion in tumor could be detected by computing the time-delay of resting-state BOLD signal. A histogram analysis of this pattern could distinguish between low-grade glioma and high-grade glioma.


No acknowledgement found.


1.Law M, Yang S, Wang H, et al. Glioma grading: sensitivity, specificity, and predictive values of perfusion MR imaging and proton MR spectroscopic imaging compared with conventional MR imaging [J]. American Journal of Neuroradiology, 2003, 24(10): 1989-1998. 2. Zhang D, Johnston J M, Fox M D, et al. Preoperative sensorimotor mapping in brain tumor patients using spontaneous fluctuations in neuronal activity imaged with fMRI: initial experience [J]. Neurosurgery, 2009, 65(6 Suppl): 226.


The T2-weighted Dark Fluid MR image and Time-shift map of one patient with WHO IV glioma. The lesion area shows non-uniform earlier arrival time (in red).

The average blood arrival histogram patterns of each grade of gliomas. The pattern of each subject was obtained by subtracting the histogram of lesion ROIfrom the ROI histogram of the mirrored ROI in the healthy side. The higher the glioma grade, the more voxelsthere are withearlier perfusion arrival time.

Box plot of the different percentages of lesion area with relative blood arrival time ranging from 2 to 5 (in z-score).

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