Development of a Model for Predicting Defects in Radiation Shielding Aprons Using Machine Learning

Abstract

This study attempted to conveniently and accurately determine the radiation images of the normal lead shielding fabric and the cracked lead shielding fabric using a convolutional neural network to confirm cracks or cracks in the radiation protective apron. Normal and cracked radiation images were determined through a synthetic neural network constructed using a MATLAB by dividing them into lead shielding fabrics of 0.125mm and 0.25mm lead equivalents. The results were analyzed using loss function, accuracy, confusion matrix, and Spearman's rho, and since the method of detecting defects in the lead apron for radiation protection in the current clinical practice is not simple and accurately organized, the applicability in clinical practice was confirmed by making a high-level judgment using deep learning.