A multiresolution prostate representation for automatic segmentation in magnetic resonance images

Area of Science:

• Medical imaging
• Radiotherapy
• Computational anatomy

Background:

• Accurate prostate delineation is crucial for effective radiotherapy, minimizing side effects.
• Manual segmentation on MRI is time-consuming and prone to observer variability.
• Automated tools are needed to improve efficiency and consistency in prostate delineation.

Purpose of the Study:

• To present a novel multiresolution representation for prostate segmentation.
• To develop a method for segmenting new prostates by combining similar cases from a dataset.
• To improve the accuracy and efficiency of prostate delineation in radiotherapy planning.

Main Methods:

• A multiresolution representation is used to identify the most similar prostates in a dataset.
• Salient points from a region of interest characterize prostate boundaries.
• Segmentation is achieved by combining registered expert delineations of similar prostates using a weighted patch-based strategy.
• Morphological operations are applied for contour smoothing.

Main Results:

• The proposed method achieved averaged Dice coefficients of 82% ± 0.07 and 83% ± 0.06 on two public datasets.
• Performance was evaluated against expert manual segmentation using a leave-one-out scheme.
• The approach demonstrated competitive results compared to state-of-the-art atlas-based methods.

Conclusions:

• The multiresolution representation effectively captures local and global spatial features for identifying similar prostates.
• The developed strategy offers a promising, easily adaptable tool for clinical annotation support.
• This method has the potential to streamline radiotherapy planning workflows.