4033
Optimal b-Value for Superpixel based Automatic Prostate Gland Segmentation using Diffusion-Weighted Imaging1Centre for Biomedical Engineering, Indian Institute of Technology Delhi, New Delhi, India, 2Department of Electronics & Communication Engineering, Amity University, Noida, India, 3Department of NMR, All India Institute of Medical Sciences Delhi, New Delhi, India, 4Department of Radiology, All India Institute of Medical Sciences Delhi, New Delhi, India, 5Department of Biomedical Engineering, All India Institute of Medical Sciences Delhi, New Delhi, India
Synopsis
Accurate segmentation of the prostate gland is a challenging task due to the high variability of prostatic anatomic structures. The diagnostic accuracy of diffusion-weighted-imaging (DWI) to detect prostate cancer (PCa) is well established. Proper delineation using DWI of the prostate gland can take an essential part in the computer-aided-diagnosis of PCa. The purpose of this study is to automatic segmentation of prostate-gland using superpixel-based segmentation in DWI and select the optimal b-value of DWI for segmentation. Experimental results provide a close match to radiologist manual delineation and able to determine the optimal b-value of DWI for superpixel-based prostate-gland segmentation.
Introduction
The second most common cancer in men is Prostate cancer (PCa).1 It has been seen that multiparametric magnetic resonance imaging (mpMRI) is convenient for the detection of PCa at an early stage, its involve T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI) and dynamic contrast-enhanced(DCE) MRI.2 Numerous studies3,4 have explored the diagnostic performance of DWI in detecting PCa with widely varied sensitivity and specificity. Hence incorporating DWI for segmentation is advantageous for computer-aided-diagnosis(CAD) of PCa. To save the time and minimize the operator-induced error,accurate and automatic segmentation is essential.5 We investigate a fully automated Simple-Linear-Iterative-Clustering(SLIC)6 superpixel-based method for segmentation of prostate-gland using different b-values DWI and determine the optimal b-value for superpixel-based segmentation.Methods
MRI Data Acquisition: MRI dataset for 14 patients with PCa, were retrospectively used for this study. All prostate MRI examinations were performed on 3T MRI system (Ingenia,Philips,Netherlands) using an external phased array body coil. MR sequences included axial turbo- spin-echo (TSE) T2W (TR/TE= 3715/100ms; slice-thickness=3mm;field of view (FOV) =160×160mm2; matrix-size=400×400) and axial-echo-planar DWI(TR/TE=5521/75ms; FOV=177.6×177.6mm2; matrix-size=176×176;slice-thickness=4mm; with four b-values of 0,500,800 and 1500s/mm2).
Superpixel based prostate-gland segmentation: Conventional T2W images were also used for segmentation of prostate-gland however contrast found very-poor compare to DWI. Therefore we have used DWI for prostate-gland segmentation in this study. In the pre-processing step, normalization, morphological erosion-operation,contrast-enhancement, and median-filtering was performed. The SLIC based superpixel method was implemented and evaluated to segment each slice in order to obtain superpixels.6 In the SLIC algorithm,we have found the number of desired equally sized superpixels is 30 and the maximum number of iterations is 100 for our dataset. Then each group of superpixel segmented image was replaced with their mean intensity,creating the six clusters. Five-level thresholding and binarization were done for the generation of the mask of prostate-gland. For the prostate-gland segmentation, we multiplied binary-mask with original images.
Data-Processing: Data were processed using in-house build algorithms with MATLAB-R2017a. Manual segmentation of prostate-gland was performed by an expert-radiologist(with more than 10 years of expertise in prostate-imaging) and used as ground-truth for calculating the performance of segmentation in terms of Dice-coefficient(DC) and Jaccard-index(JI). Here,DC=2.(|A∩B|/(|A|+|B|)) and JI=(|A∩B|/|AUB|), where A and B are areas delineated by radiologist and segmentation algorithm respectively.
Results
The performance of superpixel segmentation of prostate-gland for 14 patients was DC:87.71±6.41% and JI:77.32±8.50% at b=0 s/mm2; DC:78.15±9.30% and JI:64.62±12.30% at b=500 s/mm2; DC:77.53±9.31% and JI:64.25±10.12% at b=800 s/mm2; and DC:74.20±7.73% and JI:59.50±9.23% at b=1500 s/mm2 DWI. DC and JI were observed to decrease with increase in b-values, as shown in Figure-2.Discussion
SLIC superpixel method is an automated method that requires few input parameters, yielding a lower under-segmentation error, control on superpixel numbers with lower computational cost and increases memory efficiency. With our proposed approach it is possible to segment the prostate-gland with approximately similar DC (~88%) using DWI as compared to the previous study (89%) using structural T2W MRI.7 DWI is important in the diagnosis of PCa thus it is useful to directly perform the segmentation on DW images. Our study found that the values of DC and JI values decrease with increasing b-values. Low b-value(b=0s/mm2) DWI is shown to have a reasonably good accuracy compared to higher b-values for prostate gland segmentation using superpixel method. This is possible mostly because b-value=0s/mm2 has a better signal to noise ratio and contrast compared to surrounding tissues against higher b-values. Sub-segmentation of the prostate to its finer zones as central, peripheral and transition zone, needs to be further optimized. In the future, we can sub-segment the prostate into zones using SLIC based superpixel segmentation.Conclusion
A superpixel based methodology for automatic segmentation of prostate-gland has been developed using different b-values of DWI. Lower b-values DWI is optimal for segmentation of prostate using SLIC algorithm.Acknowledgements
This work is supported by IIT Delhi, India and AIIMS New-Delhi, India. We thanks to Dr Vijay Kubihal for providing clinical inputs. DS was supported with the research fellowship fund from Ministry of Human Resource Development, Government of India.References
1.Ferlay J, Soerjomataram I, Dikshit R, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. International Journal of Cancer. 2015;136(5): E359-E386. 2.Wang R, Wang H, Zhao C, et al. Evaluation of Multiparametric Magnetic Resonance Imaging in Detection and Prediction of Prostate Cancer. PLoS ONE. 2015;10(6):e0130207.
3.Tan CH, Wei W, Johnson V, et al Diffusion-weighted MRI in the detection of prostate cancer: meta-analysis. AJR Am J Roentgenol. 2012;199:822–829.
4.Wu LM, Xu JR, Gu HY, et al. Usefulness of diffusion-weighted magnetic resonance imaging in the diagnosis of prostate cancer. Acad Radiol. 2012;19:1215–1224.
5.Aslian H ,Sadeghi M, Mahdavi HR et al. Magnetic resonance imaging-based target volume delineation in radiation therapy treatment planning for brain tumors using localized region-based active contour. J Radiat Oncol Biol Phys. 2013;87(1):195-201.
6.Achanta R., Shaji A., Smith A, et al. SLIC Superpixels Compared to State-of-the-Art Superpixel Methods. IEEE Transactions on analysis and machine intelligence, 2012; 34(11).
7.Tian Z, Liu L., Zhang Z, et al. Superpixel-based segmentation for 3D prostate MR images. IEEE Transactions on Medical Imaging. 2015;35(3):791-801
Figures
