A New Method on Kerma Estimation in Mammography Screenings.

Area of Science:

• Medical Physics
• Artificial Intelligence in Healthcare
• Radiological Imaging

Background:

• Mammography is a key breast cancer screening tool, raising concerns about patient radiation dose.
• Breast tissue sensitivity necessitates careful management of radiation exposure during mammography.

Purpose of the Study:

• To estimate the entrance surface air kerma (ESAK) before patient irradiation in mammography.
• To develop a predictive model for patient radiation dose in mammography.

Main Methods:

• Utilized a phantom to collect mammography data (ESAK, Kvp, mAs, HVL, filter/target type).
• Trained a MultiLayer Perceptron (MLP) neural network using the Levenberg-Marquardt backpropagation algorithm.
• Estimated ESAK using the trained MLP model.

Main Results:

• The optimal neural network configuration used 35 neurons, achieving a 95.7% regression coefficient.
• The model demonstrated a Mean Squared Error (MSE) of 0.437 mGy.
• The prediction accuracy reached 95.2%, with 4.8% of output range changes accounted for.

Conclusions:

• Neural networks enable accurate pre-irradiation ESAK estimation for mammography patients.
• The proposed method shows a 4.3% difference compared to solid-state dosimeter measurements.
• This neural network approach offers superior accuracy compared to Monte Carlo simulations for ESAK prediction.