Segmentation of prostate zones using probabilistic atlas-based method with diffusion-weighted MR images
Introduction
Prostate cancer (PCa) is the second most common cancer and the fifth leading cause of cancer-related death among men worldwide [1]. Accurate segmentation of prostate gland and its zones on MR images is a challenging task due to the high variability in prostate anatomic structures, lack of clear boundaries in-between its zones, and similar intensity profiles with nearby tissues [2].
Multi-parametric MRI analysis (mpMRI) uses various MRI sequences such as T2-weighted (T2W), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) imaging for evaluation of pathological lesions in prostate. mpMRI has significantly improved the diagnostic accuracy of PCa and has the potential to reduce number of unnecessary biopsies [3]. Prostate consists of three zones: central zone (CZ), transition zone (TZ) and peripheral zone (PZ). CZ and TZ are usually not clearly distinguishable on MR images, recently a study by Vargas et al. [4] showed that in patients with PCa undergoing MR imaging, the CZ was visible at least partially, in only 81-84% of patients. However, these two zones are commonly referred to combined as TZ for the purpose of image processing and analysis [5]. About 70-75% of PCa originate in PZ, and 20-30% in TZ [6]. An internationally accepted scoring and interpretation scheme, Prostate Imaging-Reporting and Data System version 2 (PI-RADS v2) has defined the evaluation criteria for the assessment of PZ and TZ separately because tumor exhibit different characteristics in these two zones [7]. It is important to thus sub-segment the prostate into its zones for any detailed clinical interpretation and quantitative analysis.
Although the accuracy for detection of PCa using mpMRI is good [8], it can be further improved using a reliable segmentation technique along with mpMRI. Manual delineation of tissues on each slice of 3D-MRI acquisition is a tedious and time-consuming task that is also highly subjective, which significantly increases the burden for radiologists. Therefore, an automatic, accurate and reproducible segmentation algorithm can reduce image interpretation time and represent a significant advancement in computer-aided diagnostic (CAD) system and possibly also in treatment planning.
Several segmentation algorithms for the prostate gland have been reported in literature which include region-based approach [9], contour and shape-based approach [10], supervised and unsupervised classification based approach [11,12] and hybrid of these methods [13], [14], [15]. However, only few studies have focused on the sub-segmentation of the prostate into its zones [16], [17], [18] and more recently deep learning has been applied using convolution neural network (CNN) based segmentation approaches [12,[19], [20], [21]]. Most segmentation work has been presented using T2W MR images [18], [19], [20], [21] and only two studies have used DWI [12,22]. PI-RADS v2 recommends DWI sequence as the most important for the interpretation of lesions in PZ and also a key feature in TZ lesions detection [6]. Therefore, using DWI for the segmentation process is beneficial, which might be useful in CAD algorithms of PCa to make the process suitable for clinical use. However, automated segmentation of prostate gland and its zones using DWI is a challenging task due to low signal-to-noise ratio (SNR).
In this paper, the Chan-Vese active contour model [23] along with morphological opening operation has been used for whole prostate gland segmentation, and a probabilistic atlas-based approach has been developed for prostate zonal segmentation. A zone is generated after matching the test subject with the atlas at similar probability level. In this zone, a pixel may contain both types of tissues PZ or TZ; this is called partial volume (PV) zone. For the accurate estimation of PV zone pixels, either belonging to PZ or TZ, a PV correction algorithm has been developed in this study. The aim of this study was to develop a semi-automated and accurate framework to segment prostate gland and its zones simultaneously using DWI.
The main structure of this paper is organized as follows. Methodology is presented in Section 2, the MRI data acquisition in Section 2.1, and data processing in Section 2.2. Prostate gland segmentation method is presented in Section 2.2.1 and atlas-based sub-segmentation of prostate zones in Section 2.2.2. Experimental results are elaborated in Section 3. Discussion and conclusion are presented in Sections 4 and 5.
Section snippets
MRI data
MRI dataset of 18 patients (age = 63 ± 7.4 years/ range = 55-72 years) with biopsy-proven PCa was used in this study with approval from Institutional Ethics Board. All prostate MRI data were acquired at 3T scanner (Ingenia, Philips Health Systems, The Netherlands) between June 2015 and April 2017 using an external phased array body coil. MRI acquisition sequences included axial turbo spin echo (TSE) T2W (TR/TE = 3715/100 ms; slice thickness = 3 mm; field of view (FOV) = 160 × 160 mm2
Results
Experimental results showed that the proposed method is able to segment the prostate gland and its zones, PZ and TZ, with good accuracy in our dataset, Mean ± standard deviation (SD) of DSC of 90.76 ± 3.68%, JC of 83.00 ± 5.78%, and accuracy of 99.42 ± 0.36% for the prostate gland; DSC of 77.73 ± 2.76%, JC of 64.46 ± 3.43%, and accuracy of 82.46 ± 2.22% for the PZ; and DSC of 86.05 ± 1.50%, JC of 75.80 ± 2.10%, and accuracy of 91.67 ± 1.56% for the TZ was observed using the proposed methodology
Discussion
Accurate segmentation of the prostate gland and its zones is a challenging task due to high variability of the prostate anatomic structures. In literature, shape and contour-based methods were proposed as the most appropriate segmentation method for prostate gland in recent times [10,13,26]. In the current study, the region-based active contour model was used for segmentation of prostate gland and morphological opening operation was applied to refine the segmentation result. In Chan Vese active
Conclusion
A semi-automated framework for segmentation of prostate gland and its zones using DWI has been proposed. The proposed probabilistic atlas based prostate zonal segmentation with PV correction algorithm was stable, robust and easy to implement for both our and QIN-PROSTATE-Repeatability dataset. The future work will include this segmentation framework as pre-processing step in the computer-aided automated diagnostic system for PCa localization and treatment planning.
Declaration of Competing Interest
No potential conflict of interest was reported by the authors.
Acknowledgments
The authors would like to acknowledge the support from staffs of IIT Delhi, New Delhi, and AIIMS Delhi, New Delhi for data acquisition. The authors also thank the anonymous reviewers for their insightful suggestions and comments that has helped to improve the manuscript. Authors thank Dr. Siddharth Arora from University of Oxford, UK for his guidance in planning the statistical analysis. D.S. thank the Ministry of Human Resource Development, Government of India for the research fellowship
Funding source
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
References (37)
- et al.
PI-RADS Prostate Imaging – Reporting and Data System: 2015, Version 2
Eur. Urol.
(2016) - et al.
Accuracy of magnetic resonance imaging for local staging of prostate cancer: a diagnostic meta-analysis
Eur Urol.
(2016) - et al.
Accurate prostate volume estimation using multifeature active shape models on T2-weighted MRI
Acad. Radiol.
(2011) - et al.
Fast and accurate segmentation method of active shape model with Rayleigh mixture model clustering for prostate ultrasound images
Comput. Methods Programs Biomed.
(2020) - et al.
Simultaneous segmentation of prostatic zones using active appearance models with multiple coupled level sets
Comput. Vis. Image Underst.
(2013) - et al.
Central zone carcinoma of the prostate gland: a distinct tumor type with poor prognostic features
J. Urol.
(2008) - et al.
Multi-atlas based segmentation using probabilistic label fusion with adaptive weighting of image similarity measures
Comput. Methods Programs Biomed.
(2013) - et al.
Multi-atlas segmentation and correction model with level set formulation for 3D brain MR images
Pattern Recognit.
(2019) - et al.
3D registration of mpMRI for assessment of prostate cancer focal therapy
Acad. Radiol.
(2017) - et al.
Global Cancer Statistics 2018: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries
CA Cancer J. Clin.
(2018)
Prostate MRI segmentation using learned semantic knowledge and graph cuts
IEEE Trans. Biomed. Eng.
Evaluating the role of mpMRI in prostate cancer assessment
Exp. Rev. Med. Dev.
Normal central zone of the prostate and central zone involvement by prostate cancer: clinical and MR imaging implications
Radiology
The abnormal prostate: MR imaging at 1.5 T with histopathologic correlation
Radiology
A comparison of the biological features between prostate cancers arising in the transition and peripheral zones
Br. J. Urol.
Robust atlas based segmentation of the prostate and peripheral zone regions on mri utilizing multiple mri system vendors
Med. Phys.
Three-dimensional prostate segmentation using level set with shape constraint based on rotational slices for 3D end-firing TRUS guided biopsy
Med. Phys.
Automated prostate gland segmentation based on an unsupervised fuzzy C-means clustering technique using multispectral T1w and T2w MR imaging
Information
Cited by (9)
Attention-guided multi-scale learning network for automatic prostate and tumor segmentation on MRI
2023, Computers in Biology and MedicineAutomatic prostate and peri-prostatic fat segmentation based on pyramid mechanism fusion network for T2-weighted MRI
2022, Computer Methods and Programs in BiomedicineCitation Excerpt :Their experimental results show that CNN with patched DeepLab V3+ shows the best performance. Singh et al. [40] proposed a semi-automatic framework with partial volume correction algorithm for simultaneous segmentation of prostate inner gland and peripheral zone. The main method is Chan Vese active contour model combined with morphological open operation.
H-ProSeg: Hybrid ultrasound prostate segmentation based on explainability-guided mathematical model
2022, Computer Methods and Programs in BiomedicineCitation Excerpt :However, manual delineation is time-consuming, tedious, and subjective, resulting in heavy workloads for radiologists and large variability in prostate volume delineation. Despite many advances in the medical imaging field [3,4], and specifically, in prostate segmentation techniques [5,6], accurate prostate segmentation in ultrasound images remains a challenging task for the following reasons: (1) missing boundaries due to shadow artifacts produced by surrounding structures (i.e., the bladder and seminal vesicles), (2) blurry boundaries caused by low contrast between the prostate and neighboring organs, (3) inhomogeneous intensity distribution of the prostate structure, and (4) variability in prostate shape. Fig. 1 shows several challenging prostate segmentation cases using TRUS images.
Machine learning-based analysis of a semi-automated PI-RADS v2.1 scoring for prostate cancer
2022, Frontiers in OncologySegmentation of the prostate, its zones, anterior fibromuscular stroma, and urethra on the MRIs and multimodality image fusion using U-Net model
2022, Quantitative Imaging in Medicine and Surgery