Surrogate-free machine learning-based organ dose reconstruction for pediatric abdominal radiotherapy

Area of Science:

• Medical Physics
• Radiation Oncology
• Machine Learning Applications

Background:

• Accurate 3D radiation dose distribution is crucial for dose-effect modeling in radiotherapy.
• Reconstructing 3D doses for childhood cancer survivors treated before CT scans is challenging due to limited 2D radiographic data.
• Current surrogate anatomy methods lack personalization and yield coarse dose reconstructions.

Purpose of the Study:

• To develop and validate a surrogate-free Machine Learning (ML) based method for reconstructing 3D radiation dose distributions.
• To enable accurate dose-effect modeling for patients treated with radiotherapy in the pre-CT era.
• To improve personalization and accuracy in dose reconstruction for historical radiotherapy cases.

Main Methods:

• Utilized 142 abdominal CT scans of childhood cancer patients and simulated 300 artificial Wilms' tumor plans.
• Extracted anatomical features from digitally reconstructed radiographs and collected historical patient/plan features.
• Employed an evolutionary ML algorithm to link features to derived dose-volume metrics, validated with cross-validation and an independent dataset.

Main Results:

• Achieved mean absolute errors ≤ 0.6 Gy for organs fully within or outside the radiation field.
• Obtained mean absolute errors ≤ 1.7 Gy for [Formula: see text], ≤ 2.9 Gy for [Formula: see text], and ≤ 13% for [Formula: see text] and [Formula: see text] for organs at the field edge, without systematic bias.
• Demonstrated accurate and efficient dose reconstruction comparable to current methods, eliminating the need for surrogate setup.

Conclusions:

• A novel ML-based organ dose reconstruction method accurately predicts dose-volume metrics from patient and plan features.
• The surrogate-free approach enhances personalization and efficiency in reconstructing historical 3D radiation dose distributions.
• This method offers a significant advancement for studying radiotherapy-related adverse effects in long-term childhood cancer survivors.