Satyajit Maurya1, Rakesh Kumar Gupta2, and Anup Singh1,3,4
1Centre for Biomedical Engineering, Indian Institute of Technology Delhi, New Delhi, India, 2Department of Radiology, Fortis Memorial Research Institute, Gurgaon, India, 3Department of Biomedical Engineering, All India Institute of Medical Sciences, Delhi, New Delhi, India, 4Yardi School of Artificial Intelligence (ScAI), Indian Institute of Technology Delhi, New Delhi, India
Synopsis
Keywords: Tumors (Pre-Treatment), Brain, Blood vessels
Motivation: Echo planar based SWI (SWI-EPI) can provide better contrast of vasculature and higher spatial resolution compared to SWI and at shorter acquisition time. However, its potential in glioma grading has not been explored well.
Goal(s): To evaluate the potential of SWI-EPI for automatic segmentation and quantification of tumor vasculature for glioma grading.
Approach: Tumor vasculature for both SWI and SWI-EPI images were segmented and quantified. T-test and ROC curve analysis was used to determine statistical significance and grading accuracy.
Results: Tumor vasculature features automatically computed from SWI-EPI provided improved glioma grading accuracy compared to conventional SWI based features.
Impact: SWI-EPI offers advantages
over SWI images in terms of image resolution and shorter acquistion time. It
was found to have improved glioma grading accuracy and has the potential to be used as
a routine imaging sequence in clinical settings.
Introduction
Gliomas represent a
heterogeneous group of primary brain tumors presenting diverse
histopathological and molecular characteristics, associated with significant
mortality and morbidity. Due to their heterogeneous nature, accurate glioma
grading is necessary for treatment planning and monitoring disease progression.
HGGs (high grade gliomas) in comparison to LGGs (low grade gliomas) present a
higher degree of angiogenesis, severe necrosis, and micro-hemorrhages. Angiogenesis
plays an important role in the development and progression of tumors. The
degree of angiogenesis in tumors can be an important parameter in determining
the tumor progression.
Several studies have been
conducted that have tried to grade gliomas by indirect, non-invasive
quantification of the degree of angiogenesis. These studies had the downside of
either being semi-quantitative1 or excluding
cases that did not have visible ITSS2 (Intratumoral susceptibility signals).
In this study, we
developed a fully automatic approach for IVV (ITSS vasculature volume) segmentation
and quantification for glioma grading using SWI-EPI imaging sequence. SWI-EPI employs
the echo-planar based acquisition for obtaining the SWI images, whereas the
conventional SWI images are acquired using a gradient echo-based sequence. In
our experiments, it was obsereved that the image resolution of SWI-EPI was
higher than that of the SWI images (Figure 1), with the added advantage of
faster acquisition time.Methods
A total of 68 glioma
patients (15 LGG, 49 HGG) were considered for this study. Routine imaging
sequences like T1-weighted, T2-weighted, 3D FLAIR, and post-contrast T1-weighted
were acquired along with SWI-EPI and SWI sequences.
FLAIR and SWI images were
co-registered with SWI-EPI images. The whole tumor (WT) region was segmented
from the FLAIR images by using an in-house developed tumor segmentation model3. The tumor masks
were applied to the SWI-EPI and the SWI images and the ITSS was segmented using
Otsu thresholding. R2* relaxation maps computed from multi-echo SWI
images were used to obtain the tumor micro-hemorrhage masks by using an
appropriate threshold value. This threshold value was chosen based on R2* maps of
4 non-glioma patients who presented chronic hemorrhage. These hemorrhage masks were then subtracted
from the ITSS masks to obtain just the tumor vasculature, termed IVV (ITSS
vasculature volume). Figure
2 shows the images obtained by the above-mentioned steps.
A new grading feature
nIVV-TVR (normalized IVV-Tumor volume ratio), obtained by dividing IVV with the
WT volume, followed by appropriate normalization was also proposed.
Results
The mean, standard
deviation along with the range of IVV and nIVV-TVR for LGG and HGG for both SWI
and SWI-EPI were separately calculated and are summarized in Table 1. IVV
values were significantly higher for HGGs as compared to LGGs. This was seen
both for the SWI-EPI and the SWI images. The same trend was observed in the
case of nIVV-TVR. It was also observed that IVV and nIVV-TVR values were higher
when using the SWI-EPI images for the same subject. The proposed nIVV-TVR
values using SWI-EPI images showed a higher significance level (P < 0.001)
than the values obtained using SWI images (P = 0.0097).
ROC curve analysis - An
AUC (Area under the curve) of 0.822 as compared to 0.796 was obtained when
comparing the grading accuracy of IVV for SWI-EPI and SWI, respectively.
Comparing the AUCs of SWI-EPI (0.850) and SWI (0.801), it was seen that
there was an improvement in the diagnostic accuracy using SWI-EPI images with
nIVV-TVR as the differentiating metric. Table 2 reports the sensitivity and
specificity values for differentiating between the tumor grades for the SWI-EPI
and the SWI images. Figure 3 shows the ROC AUC curve analysis for
differentiating between the tumor grades using IVV and the proposed nIVV-TVR.Discussion
This study was aimed at evaluating the potential of SWI-EPI towards
grading gliomas into LGG and HGG and compared it with GRE-based SWI. In general,
SWI-EPI provides better contrast for visualizing the brain vasculature.Conclusion
In this study, a fully
automatic approach for tumor vasculature segmentation and qunatification using
SWI-EPI for glioma grading was developed. It was found that SWI-EPI has
potential advantages over SWI and can provide better diagnostic capabilities
for tumor grading.Acknowledgements
I acknowledge fundings from SERB-DST (Project no. CRG/2019/005032) and the PMRF fellowship for supporting this project. I also thank Rakesh Kumar Singh from Fortis Memorial Research Institute for his help in data collection.References
1. Park
MJ, Kim HS, Jahng GH, Ryu CW, Park SM, Kim SY. Semiquantitative assessment of
intratumoral susceptibility signals using non-contrast-enhanced high-field
high-resolution susceptibility-weighted imaging in patients with gliomas:
comparison with MR perfusion imaging. Am J Neuroradiol.
2009;30(7):1402-1408.
2. Bhattacharjee
R, Gupta RK, Patir R, Vaishya S, Ahlawat S, Singh A. Quantitative vs.
semiquantitative assessment of intratumoral susceptibility signals in patients
with different grades of glioma. J Magn Reson Imaging.
2020;51(1):225-233.
3. Maurya
S, Kumar Yadav V, Agarwal S, Singh A. Brain Tumor Segmentation in mpMRI Scans
(BraTS-2021) Using Models Based on U-Net Architecture. In: International
MICCAI Brainlesion Workshop. Springer; 2021:312-323.