Evaluation of the usefulness of deep neural networks in classifying X-ray images according to radiation exposure level for automatic exposure control of digital radiography

In medicine, X-ray imaging has to have stable image quality while minimizing radiation exposure. In recent digital radiography (DR), the radiation exposure and image quality are controlled by employing automatic exposure control (AEC) technology. In such an environment, the technology needs the capability of accurately classifying the level of radiation exposure despite statistical X-ray fluctuations. Therefore, the purpose of this study was to implement a radiation exposure classifier that could objectively determine the level using the deep neural network (DNN) method. Datasets containing 2000 images were used for training and validation of the DNN. This was done by changing the sensitivity of the AEC of the DR system to S200, S400, S800, and S1000. The changes in sensitivity generated four groups of 500 X-ray images each: overexposure, normal exposure (2), normal exposure (1), and underexposure. The results showed a classification accuracy of 93.83% and a loss 38.9%. In conclusion, through deep learning with X-ray images, an objective radiation exposure classifier was implemented and is expected to be used for the AEC of DR systems and to improve diagnostic accuracy.