Computer-assisted decision making in portal verification--optimization of the neural network approach

Area of Science:

• Medical Physics
• Artificial Intelligence in Medicine
• Radiation Oncology

Background:

• Conventional portal verification relies on radiation oncologist expertise, which is unsustainable with increasing image volumes and time pressures.
• Automating portal verification is crucial for efficient and accurate radiation therapy delivery.
• Artificial intelligence offers a potential solution for decision-making in portal verification.

Purpose of the Study:

• To develop, optimize, and evaluate an artificial intelligence decision-making tool for portal verification using clinical data.
• To create an artificial neural network (ANN) that approximates radiation oncologist assessments of portal images.
• To enhance the efficiency and accuracy of radiation therapy quality assurance.

Main Methods:

• A dataset of 328 electronic portal images from breast irradiations was analyzed.
• A radiation oncologist expert rated treatment setup acceptability on a scale of 0-10.
• A three-layer feedforward artificial neural network (ANN) was trained on feature vectors derived from image data and expert ratings.

Main Results:

• The ANN achieved an RMS discrepancy of 1.05 rating points compared to the expert.
• The ANN correctly identified 100% of unacceptable portals as classified by the expert.
• The ANN misclassified only 6.5% of acceptable portals as unacceptable.

Conclusions:

• The developed ANN portal image classifier demonstrates feasibility as an automated assistant for radiation oncologists.
• The tool can recommend decisions on treatment setup acceptability from portal images.
• This AI approach has the potential to improve the quality assurance process in radiation oncology.