A Local ROI-specific Atlas-based Segmentation of Prostate Gland and Transitional Zone in Diffusion MRI

Junjie Zhang, Sameer Baig, Alexander Wong, Masoom A. Haider, Farzad KhalvatiJournal of Computational Vision and Imaging Systems 2016 · DOI

Abstract

Segmentation of prostate and related anatomic structure, such as transitional zone, in medical images facilitates prostate cancer detection, as well as a number of other clinical practices. In this paper, we propose a semi-automatic local ROI-specific atlas-based segmentation (LABS) method to segment prostate gland and transitional zone in diffusion magnetic resonance images. Inspired by a sequential registration-based segmentation method, the proposed method further reduces the amount of user intervention and focuses on the vicinity of prostate for atlas matching and atlas-to-target registration by specifying the bounding boxes of prostate gland on key slices of volume images. We evaluated the method on an atlas database with the 100 cases by performing a leave-one-out study. Our proposed method produced favorable outcomes with an average Dice similarity coefficient of 0.85±0.03 for prostate gland and 0.77±0.06 for transitional zone segmentations, which indicates the effectiveness of the proposed method and its potential to be used in relevant clinical applications.

Introduction

Accurate segmentation of the prostate gland and its transitional zone (TZ) is a critical prerequisite for effective prostate cancer diagnosis and radiation therapy planning. Traditionally, this task is performed manually by clinicians, a process that is both time-consuming and subject to significant inter-observer variability.

While automated atlas-based methods exist, they often struggle with the inherent variability of MRI scans, which can differ significantly based on the scanner, field of view, and individual patient anatomy. Global registration techniques frequently fail because the algorithm is "distracted" by anatomical noise from surrounding organs. To address this, we propose a Local ROI-specific Atlas-based Segmentation (LABS) method that focuses computational resources on the immediate vicinity of the prostate.

Methodology

The LABS pipeline is designed to minimize user effort while maximizing registration accuracy through a localized approach. The method consists of four primary stages:

  1. User-specified Bounding Boxes (BBs): Instead of complex manual tracing, a user draws simple bounding boxes around the prostate on key slices (top, middle, and bottom). These are interpolated to create a 3D Volume of Interest (VOI).
  2. Atlas Selection: The system queries a database to find the reference scan that most closely matches the target's VOI based on image correlation and volume ratios.
  3. ROI-based Registration: Affine registration is performed exclusively within the specified bounding boxes. This localized focus prevents registration errors caused by anatomical variations outside the prostate.
  4. Post-processing: The system applies the transformation to the atlas labels and performs a final scaling step to ensure the segmented mask fits precisely within the user's original bounding boxes.
Overview of the LABS segmentation pipeline

Experimental Setup and Results

The study utilized a dataset of 100 patients' diffusion-weighted MRI (dMRI) scans. Validation was performed using a leave-one-out cross-validation strategy. Performance was measured using the Dice Similarity Coefficient (DSC), where 1.0 represents a perfect match with expert manual segmentation.

Quantitative performance (DSC) comparison across different configurations

Key results from the evaluation include:

  • Prostate Gland: Achieved a DSC of 0.85 ± 0.03 when using 5 user-specified bounding boxes and post-processing.
  • Transitional Zone: Achieved a DSC of 0.77 ± 0.06.
  • Impact of User Input: Increasing the number of user-provided bounding boxes from 3 to 5 improved the prostate segmentation accuracy by approximately 5%.
  • Post-processing Efficacy: The final scaling step improved the DSC by 4–7% across different configurations, demonstrating that fine-tuning the atlas result to the user's specific ROI is vital for high accuracy.

Discussion and Conclusion

The LABS method demonstrates that restricting the computer's focus to a local region-of-interest significantly improves the robustness of atlas-based segmentation in MRI. By incorporating a "human-in-the-loop" approach for initial localization, the system bypasses the common pitfalls of global registration.

A primary limitation remains the lower accuracy in segmenting the transitional zone compared to the whole gland, likely due to the ambiguous boundaries of internal structures. Future work will involve testing the algorithm on more diverse datasets from multiple institutions to further validate the generalizability of this local approach across different scanner types and clinical settings.

Summary updated

Abstract

Segmentation of prostate and related anatomic structure, such as transitional zone, in medical images facilitates prostate cancer detection, as well as a number of other clinical practices. In this paper, we propose a semi-automatic local ROI-specific atlas-based segmentation (LABS) method to segment prostate gland and transitional zone in diffusion magnetic resonance images. Inspired by a sequential registration-based segmentation method, the proposed method further reduces the amount of user intervention and focuses on the vicinity of prostate for atlas matching and atlas-to-target registration by specifying the bounding boxes of prostate gland on key slices of volume images. We evaluated the method on an atlas database with the 100 cases by performing a leave-one-out study. Our proposed method produced favorable outcomes with an average Dice similarity coefficient of 0.85±0.03 for prostate gland and 0.77±0.06 for transitional zone segmentations, which indicates the effectiveness of the proposed method and its potential to be used in relevant clinical applications.
BibTeX
@article{zhang2016local,
  author = {Junjie Zhang and Sameer Baig and Alexander Wong and Masoom A. Haider and Farzad Khalvati},
  title = {{A Local ROI-specific Atlas-based Segmentation of Prostate Gland and Transitional Zone in Diffusion MRI}},
  journal = {Journal of Computational Vision and Imaging Systems},
  year = {2016},
  doi = {10.15353/vsnl.v2i1.113},
}
Figure