Abstract
In digital radiography, image quality is synergistically affected by anatomy-specific examinations, exposure factors, body parameters, detector types, and vendors/systems. However, estimating appropriate exposure factors before radiography with optimized image quality without overexposure or underexposure to patients is difficult. Thus, there is an unmet need to establish a model to predict appropriate mAs for optimizing image quality before radiography. This study aimed to establish a machine learning (ML) model for predicting an appropriate current-time product (mAs) using the target exposure indicator in chest digital radiography. An anthropomorphic chest phantom was used to establish a target exposure indicator which was used to define overexposure and underexposure in the human study. This study enrolled 1,000 (M/F = 915/85) subjects who underwent regular chest radiography. The chest thickness, height, weight, body mass index, mAs, and concomitant reached exposure (REX) were recorded. To construct the prediction model, the dataset was randomly separated into training (80%) and testing (20%) sets by matching their demographic characteristics. Five ML models were trained using the training set with 10-fold cross validation, and the model performance was evaluated using the testing set with correlation coefficients, root-mean-square error, and mean average error. The phantom study showed that the average REX was 355.6 which served as the target exposure indicator. In human study, the comparisons showed that the artificial neural network (ANN) model was the most suitable for predicting both REX and mAs values. The results demonstrated that, on average, the predicted mAs values were 10% lower and 8% higher than the values determined by AEC in the overexposure (REX > 355.6) and underexposure (REX < 355.6) groups, respectively. Moreover, the predicted mAs values were further reduced in all patients when lowering the target REX values. We concluded that the ML approach was feasible for building an artificial intelligence model for predicting appropriate mAs with target exposure indicator for chest digital radiography.