MRI-based pseudo CT synthesis using anatomical signature and alternating random forest with iterative refinement...

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Machine Learning in Medicine

Background:

• Magnetic resonance imaging (MRI) lacks the electron density information required for accurate radiotherapy treatment planning.
• Computed tomography (CT) provides essential electron density data but involves ionizing radiation and may not always be feasible for serial imaging.
• Integrating MRI and CT data is crucial for enhancing radiotherapy precision and patient safety.

Purpose of the Study:

• To develop and validate a learning-based method for synthesizing patient-specific pseudo computed tomography (PCT) images from routine magnetic resonance imaging (MRI).
• To assess the feasibility and reliability of the proposed PCT synthesis technique for MRI-based radiotherapy treatment planning.

Main Methods:

• A learning-based approach involving a training and synthesizing stage was employed.
• Patch-based features were extracted from MRI data, and feature selection identified informative anatomical signatures.
• A sequence of alternating random forests, trained via an iterative refinement model, was used for PCT synthesis.

Main Results:

• Quantitative assessment using mean absolute error, peak signal-to-noise ratio, and normalized cross-correlation showed high accuracy for both brain and pelvic data.
• The method demonstrated reliable synthesis of PCT images from MRI, comparable to ground truth CT scans.
• Specific accuracy metrics were reported for brain ($$, $$, $$) and pelvic ($$, $$, $$) datasets.

Conclusions:

• The developed learning-based method effectively synthesizes pseudo CT images from MRI data.
• The proposed PCT synthesis technique is feasible and reliable for potential use in MRI-based radiation treatment planning.
• This approach offers a promising non-ionizing radiation alternative for radiotherapy planning.