Reconstruction of volume averaging effect-free continuous photon beam profiles from discrete ionization chamber array...

Area of Science:

• Medical Physics
• Radiotherapy
• Machine Learning Applications

Background:

• The ICProfiler (ICP) ionization chamber array has limited spatial resolution and suffers from volume averaging effects (VAE), restricting its use in precise photon beam profile measurements.
• Accurate photon beam profiles are crucial for effective radiotherapy planning and delivery.

Purpose of the Study:

• To investigate the feasibility of using machine learning, specifically artificial neural networks (ANNs), to reconstruct continuous, VAE-free photon beam profiles from discrete ICP measurements.
• To overcome the inherent limitations of the ICP detector for precise beam characterization.

Main Methods:

• Photon beam profiles (6 MV) were measured using an ICP at various field sizes and depths.
• Discrete ICP measurements were interpolated to create continuous profiles.
• Artificial neural networks (ANNs) were trained to restore the penumbra, with both plane-specific and combined models evaluated.

Main Results:

• ANNs significantly reduced the penumbra width difference (PWD) between reconstructed and reference profiles, from millimeters to approximately 0.1 mm.
• Reconstructed profiles showed excellent agreement with reference data across different depths and field sizes.
• A combined ANN effectively restored both in-plane and cross-plane beam profiles.

Conclusions:

• Simple ANNs are feasible for reconstructing VAE-free continuous photon beam profiles from discrete ICP measurements.
• Machine learning offers a viable solution to enhance the accuracy of beam profile characterization using the ICP.
• A single combined ANN can successfully reconstruct both in- and cross-plane profiles, simplifying the process.