A trial for EBT3 film without batch-specific calibration using a neural network.

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate dose verification is critical in radiation therapy.
• EBT3 films are widely used for dose measurement but often require batch-specific calibration.
• Developing a batch-independent calibration method is highly desirable.

Purpose of the Study:

• To establish an artificial neural network (ANN) method for converting EBT3 film pixel values to absorbed dose without batch-specific calibration.
• To evaluate the performance and generalization capability of the developed ANN method.

Main Methods:

• A feed-forward ANN model was developed using Pytorch.
• EBT3 films from different batches were scanned, and their pixel values were used to train the ANN.
• Radiation therapy planning system (TPS) calculated dose maps served as ground truth labels for the ANN model.
• Cross-validation was performed to assess the method's generalization.

Main Results:

• The ANN method demonstrated good agreement between converted and TPS-calculated dose maps.
• Mean square errors (MSEs) for sliding window irradiation were below [Formula: see text] for trained and tested batches.
• Gamma analysis (3%, 3 mm) for intensity-modulated radiotherapy (IMRT) films showed >97.5% agreement for trained and >97.0% for tested batches.

Conclusions:

• It is feasible to develop a batch-independent ANN method for converting EBT3 film pixel values to absorbed dose.
• The established ANN method shows good performance and generalization, applicable to different batches and irradiation techniques.
• This approach can simplify dosimetry procedures and enhance the reliability of dose verification in radiotherapy.