A feasibility study to predict 3D dose delivery accuracy for IMRT using DenseNet with log files.

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Machine Learning in Healthcare

Background:

• Intensity-modulated radiation therapy (IMRT) requires rigorous quality assurance (QA) for accurate dose delivery.
• Three-dimensional (3D) gamma analysis is a standard method for IMRT QA, but can be time-consuming.
• Log files from IMRT delivery contain rich data that could potentially be leveraged for faster QA.

Purpose of the Study:

• To investigate the feasibility of using DenseNet, a deep learning model, to establish a 3D gamma prediction model for IMRT.
• To predict gamma passing rates (GPRs) directly from machine log files, bypassing traditional time-intensive analysis.
• To assess the accuracy and efficiency of this predictive model for patient-specific QA.

Main Methods:

• A DenseNet model was developed using 55 IMRT plans (367 fields).
• Log files containing gantry angle, monitor units (MU), multi-leaf collimator (MLC), and jaw positions were collected.
• Log files were converted into MU-weighted fluence maps as input, with GPRs under four gamma criteria (3%/3mm, 3%/2mm, 2%/3mm, 2%/2mm) as output.

Main Results:

• The 3D GPR prediction model demonstrated accuracy, with Mean Absolute Errors (MAE) ranging from 1.41 to 3.54 and Root Mean Square Errors (RMSE) from 1.85 to 4.40 across different gamma criteria.
• A significant correlation (P < 0.01) was found between predicted and measured GPRs.
• Model accuracy was comparable between validation and test sets, and higher in the high GPR group.

Conclusions:

• A 3D GPR prediction model for patient-specific IMRT QA was successfully established using DenseNet and log file data.
• This model shows potential as an auxiliary tool to enhance the accuracy and efficiency of IMRT dose validation.
• The findings suggest a feasible approach for expedited and reliable IMRT QA.