Evaluation of performance of pelvic CT-MR deformable image registration using two software programs

Area of Science:

• Medical Imaging
• Radiology
• Image Processing

Background:

• Deformable image registration (DIR) is crucial for aligning medical images.
• Assessing DIR accuracy between CT and MR images is essential for treatment planning.
• Evaluating open-source and commercial DIR software provides insights into their performance.

Purpose of the Study:

• To compare the accuracy of deformable image registration (DIR) between CT and MR images using Elastix and Velocity AI.
• To evaluate the impact of different DIR parameter settings and MR sequences on registration accuracy.
• To determine the suitability of these DIR methods for clinical applications involving pelvic anatomy.

Main Methods:

• CT, T1-weighted (T1w) MR, and T2-weighted (T2w) MR images from five male patients' pelvic regions were used.
• Deformable image registration (DIR) was performed using open-source Elastix software with varying regularization weights and commercial Velocity AI software.
• Accuracy was evaluated using Dice Similarity Coefficient (DSC) and Mean Distance to Agreement (MDA) for prostate, bladder, rectum, and femoral heads.

Main Results:

• Most DIR algorithms achieved good accuracy (DSC and MDA) for pelvic organs, except for the bladder.
• Elastix showed CT-T1w outperformed CT-T2w for DSC and MDA at lower regularization weights (0, 0.01, 0.1).
• Velocity AI demonstrated no significant differences in DSC and MDA across organs, suggesting sequence-dependent DIR accuracy.

Conclusions:

• DIR accuracy between CT and MR images can vary depending on the MR sequence used.
• Open-source (Elastix) and commercial (Velocity AI) software show different sensitivities to DIR parameters and image sequences.
• Further research is needed to optimize DIR protocols for specific clinical applications and imaging modalities.