Learning Non-rigid Deformations for Robust, Constrained Point-based Registration in Image-Guided MR-TRUS Prostate...

Area of Science:

• Medical Imaging
• Image Registration
• Prostate Cancer Diagnostics

Background:

• Accurate fusion of pre-procedure magnetic resonance (MR) imaging and intra-procedure trans-rectal ultrasound (TRUS) is crucial for image-guided prostate cancer biopsies.
• Current clinical MR-TRUS image fusion relies on segmentation, which is challenging and variable in TRUS, potentially leading to inaccurate biopsy targeting and false negatives.
• Prostate cancer is a leading cause of cancer-related death in men, underscoring the need for improved diagnostic accuracy.

Purpose of the Study:

• To develop and validate a non-rigid surface registration approach for MR-TRUS fusion using a statistical deformation model (SDM).
• To improve the accuracy and robustness of prostate gland registration compared to current clinical methods.

Main Methods:

• A novel non-rigid surface registration method based on a statistical deformation model (SDM) derived from clinical training data was developed.
• The SDM captures intra-procedural deformations for more accurate registration between MR and TRUS images.
• Validation included synthetic experiments with PI-RADS parcellation and clinical landmark data.

Main Results:

• The proposed SDM-based registration achieved a median target registration error of 2.98 mm, significantly outperforming the current clinical method.
• The SDM approach demonstrated robustness against common segmentation errors found in clinical TRUS data.
• Quantitative analysis showed improved volume of interest overlaps using the PI-RADS parcellation standard.

Conclusions:

• The statistical deformation model provides a more accurate and robust method for non-rigid MR-TRUS registration in prostate cancer biopsies.
• This advancement can lead to improved localization of biopsy targets, potentially reducing false-negative cancer detection rates.
• The SDM approach offers a significant improvement over existing segmentation-reliant registration techniques in clinical practice.