Improving oncoplastic breast tumor bed localization for radiotherapy planning using image registration algorithms.

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Anatomy

Background:

• Precise tumor bed localization is critical for effective radiotherapy, minimizing damage to healthy tissues and preventing cancer recurrence.
• Breast cancer treatment presents unique challenges due to soft tissue deformations, complicating accurate tumor bed identification.
• Standard medical image registration methods struggle with the discontinuous deformations and lack of direct correspondence in tumor bed localization.

Purpose of the Study:

• To evaluate the performance of various deformable image registration algorithms for breast tumor bed localization.
• To develop and utilize artificial deformations that accurately model the tumor bed creation process for evaluation.
• To compare B-Splines FFD, Demons, and TV-L1 optical flow algorithms in reconstructing discontinuous deformation fields.

Main Methods:

• Development of artificial deformations simulating tumor bed creation for quantitative assessment.
• Comprehensive evaluation of B-Splines FFD, Demons (symmetric and asymmetric), and TV-L1 optical flow algorithms.
• Assessment metrics included deformation field reconstruction accuracy, target registration error, 3D contour propagation, and expert visual judgment.

Main Results:

• Rigid registration and B-Splines FFD failed to accurately reconstruct the discontinuous deformation fields characteristic of tumor bed changes.
• Symmetric Demons demonstrated superior performance in soft tissue alignment, deformation field reconstruction, and target registration error.
• While visually appealing, B-Splines FFD and TV-L1 optical flow were found inappropriate for breast tumor bed localization due to poor deformation reconstruction.

Conclusions:

• Symmetric Demons algorithm offers the most accurate approach for breast tumor bed localization among the evaluated methods.
• Current practical methods like B-Splines FFD are insufficient for accurately handling the complex, discontinuous deformations involved.
• No evaluated algorithm achieved sufficient accuracy in reconstructing all aspects of the deformation field, highlighting areas for future research.