WE-E-213CD-05: A Non-Rigid Image Registration Algorithm That Accommodates Organ Segmentation Error.

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Analysis

Background:

• Daily Cone-beam CT (CBCT) imaging is crucial for image-guided radiotherapy (IGRT).
• Organ segmentation accuracy in CBCT can be variable, impacting deformable image registration (DIR).
• Inter-observer variability in organ delineation introduces uncertainty in daily CBCT scans.

Purpose of the Study:

• To develop a novel DIR algorithm using surface matching that incorporates organ delineation uncertainty.
• To improve the robustness of image registration in the presence of segmentation errors in CBCT.
• To leverage a priori knowledge of segmentation uncertainty to guide the registration process.

Main Methods:

• A new DIR algorithm was developed based on surface matching, penalized by estimated inter-observer segmentation uncertainty.
• The algorithm prioritizes reliably delineated organ surfaces while down-weighting uncertain regions.
• Comparison of three registration algorithms: intensity-only (INT), equally-weighted surface and image registration (EWSIR), and uncertainty-weighted surface and image registration.

Main Results:

• The proposed uncertainty-weighted method significantly reduced surface dissimilarity compared to baseline methods.
• The Jacobian of the transformation was closer to one with the proposed method, indicating more accurate volume preservation in the prostate.
• Preliminary evaluation demonstrated improved registration performance.

Conclusions:

• Slice-by-slice manual contouring in prostate radiotherapy exhibits variable spatial accuracy, potentially misguiding surface-matching DIR methods.
• The proposed method explicitly accounts for segmentation uncertainty, unlike traditional methods that assume exact surface matching.
• This uncertainty-aware DIR approach shows promise for enhancing accuracy in image-guided radiotherapy (IGRT).