SEGMENTING CT PROSTATE IMAGES USING POPULATION AND PATIENT-SPECIFIC STATISTICS FOR RADIOTHERAPY

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiotherapy Planning

Background:

• Accurate prostate segmentation is crucial for effective radiotherapy.
• Existing methods often struggle with intra-patient anatomical variations.
• Intensity and gradient features may not be sufficiently distinctive for segmentation.

Purpose of the Study:

• To develop a robust and accurate deformable model for prostate segmentation from CT images.
• To improve segmentation by incorporating both population and patient-specific shape statistics.
• To address the challenge of intra-patient variation in prostate shape during radiotherapy.

Main Methods:

• A modified Scale-Invariant Feature Transform (SIFT) local descriptor was employed for enhanced feature characterization.
• An online training approach was utilized to build patient-specific shape statistics.
• The deformable model integrates population and patient-specific statistical information.

Main Results:

• The proposed method demonstrated robust and accurate performance in prostate segmentation.
• The modified SIFT descriptor proved more distinctive than traditional features.
• The online training approach effectively captured critical intra-patient shape variations.

Conclusions:

• The developed deformable model is suitable for clinical application in prostate segmentation.
• The integration of SIFT features and patient-specific statistics enhances segmentation accuracy.
• The method offers a promising solution for improving radiotherapy planning through precise prostate delineation.