Comparison of Synthetic Computed Tomography Generation Methods, Incorporating Male and Female Anatomical Differences,...

Area of Science:

• Medical Physics
• Radiotherapy Oncology
• Medical Imaging

Background:

• Magnetic resonance imaging (MRI)-only radiotherapy planning requires synthetic computed tomography (sCT) generation.
• Existing sCT research often overlooks large pelvic regions and female-specific anatomy.
• Accurate sCT is crucial for effective treatment planning in complex anatomical sites.

Purpose of the Study:

• To apply and validate four popular sCT generation methods for MRI-only planning.
• To assess sCT accuracy for male and female anorectal and gynecological neoplasms.
• To compare sCT methods against conventional computed tomography (CT) for Hounsfield Unit (HU) estimation and plan dosimetry.

Main Methods:

• Four sCT generation methods (bulk density, tissue class, hybrid atlas, deep learning) were applied to paired MRI and CT scans from 40 patients.
• Dosimetric accuracy was evaluated using dose difference, dose volume histogram (DVH) parameters, and 3D gamma analysis.
• Hounsfield Unit (HU) estimation accuracy was assessed by mean error and mean absolute error.

Main Results:

• All tested sCT methods demonstrated high dosimetric agreement with CT (median dose difference <1.0%).
• Deep learning and hybrid atlas methods showed superior HU estimation accuracy.
• Mean 3D gamma dose agreement was high across all methods (99.8% at 3%/2 mm).

Conclusions:

• All sCT generation methods provide comparable, high dosimetric accuracy for pelvic MRI-only radiotherapy planning.
• The selection of an sCT technique can be based on departmental resources and image guidance needs.
• Minimal impact on dosimetric accuracy allows flexibility in choosing sCT methods.