An Approach for Live Motion Correction for TRUS-MR Prostate Fusion Biopsy using Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Prostate biopsy accuracy is limited by alignment challenges between pre-operative MRI and live TRUS.
• Patient movement and probe pressure cause prostate deformation, impacting image registration.
• Efficient and accurate MR-TRUS alignment is crucial for effective fusion-guided biopsies.

Purpose of the Study:

• To develop a deep learning framework for real-time, accurate alignment in TRUS-MR fusion guided prostate biopsies.
• To simultaneously correct for rigid motion and prostate deformation during live biopsy procedures.
• To reduce computation time and improve the efficiency of MR-TRUS image registration.

Main Methods:

• An end-to-end deep learning network was designed for integrated rigid and deformation correction.
• The framework was trained and validated using 6500 images from 34 subjects.
• The registration pipeline was evaluated on an unseen patient dataset.

Main Results:

• The proposed pipeline achieved a Target Registration Error (TRE) of 2.51 mm after correction.
• The system demonstrated a computation time of 70ms, enabling a rendering rate of 14 FPS.
• The framework is well-suited for live TRUS-MR alignment in clinical settings.

Conclusions:

• The developed deep learning framework significantly improves TRUS-MR image alignment accuracy and efficiency.
• Real-time correction of rigid motion and deformation enhances the reliability of fusion-guided prostate biopsies.
• This technology offers a valuable tool to reduce false negative rates in prostate cancer diagnosis.