WE-E-213CD-02: Gaussian Weighted Multi-Atlas Based Segmentation for Head and Neck Radiotherapy Planning

Area of Science:

• Medical Imaging
• Radiotherapy Planning
• Computational Anatomy

Background:

• Accurate segmentation of organs at risk is crucial for Intensity-Modulated Radiation Therapy (IMRT) planning in head and neck cancer.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Multi-atlas segmentation offers a potential solution for automated and consistent contouring.

Purpose of the Study:

• To develop and evaluate a multi-atlas segmentation strategy for head and neck IMRT planning.
• To compare different atlas selection and image registration techniques for improved segmentation accuracy.
• To optimize the fusion strategy for combining information from multiple atlases.

Main Methods:

• A multi-atlas segmentation approach using pairwise demons Deformable Registration (DR) was developed.
• Fixed Number (FN) and Thresholding (TH) atlas selection methods were compared.
• Gaussian Weighted (GW) fusion was employed, adapted for poor soft tissue contrast.
• The method was validated on 31 head and neck CT datasets without pre-clustering.

Main Results:

• The Fixed Number (FN) atlas selection strategy yielded significantly higher Dice Similarity Coefficients (DSC) for most structures compared to single atlas selection.
• Gaussian Weighted (GW) fusion was the optimal strategy for most structures, while average contouring benefited tubular structures.
• High median DSC values were achieved: 0.86 (mandible), 0.80 (spine), 0.51 (optical nerves), 0.81 (eyes), 0.69 (parotids), and 0.79 (brainstem).

Conclusions:

• An efficient multi-atlas segmentation algorithm for CT volumes was developed, utilizing Deformable Registration (DR) and Gaussian Weighted (GW) fusion.
• Fixed Number (FN) atlas selection enhances computational efficiency.
• The strategy's applicability in real clinical settings is supported by the absence of pre-clustering and specific imaging protocols in the database.