Accuracy quantification of a deformable image registration tool applied in a clinical setting

Area of Science:

• Medical Imaging
• Radiotherapy Physics

Background:

• Accurate image registration is crucial for radiotherapy planning and replanning.
• Deformable image registration (DIR) tools aim to improve alignment between planning and fraction CT scans.
• Evaluating DIR tool accuracy in clinical scenarios is essential.

Purpose of the Study:

• To assess the accuracy of a commercial deformable image registration tool (Velocity AI) in a clinical context.
• To establish a method for evaluating DIR transformation accuracy against expert-defined references.
• To quantify registration errors and interobserver variations.

Main Methods:

• 16 patients (head and neck, thoracic, abdominal) underwent planning CT and fraction CT scans.
• 30-50 anatomical landmarks were identified on each scan.
• Velocity AI was used for deformable image registration.
• Target registration error (TRE) was calculated, considering interobserver variability (1.2 ± 1.1 mm).

Main Results:

• The DIR tool reduced misalignments from 4-8 mm to 1-4 mm for approximately 50% of landmarks (77% overall).
• Alignment quality showed minor variations across tumor locations (Head and Neck: 79% ≤ 4 mm, Thoracic: 79% ≤ 4 mm, Abdominal: 59% ≤ 4 mm).
• No significant difference in alignment quality was observed between bony and soft tissues.

Conclusions:

• The investigated DIR tool can reduce mean target registration error to a clinically acceptable level for retreatment plan evaluation.
• Physician review of DIR results is still necessary due to the need for further alignment quality improvements.
• The study provides a robust method for evaluating DIR tool performance in clinical practice.