There is an increasing call for radiation dose tracking from medical examinations and patient-specific dose management has become a great concern. Especially, since computed tomography (CT) can lead to a significant amount of patient dose, fast and accurate CT dose estimation has become an important issue. For real-time scan protocol optimization and patient-specific dose management in cone-beam CT (CBCT), we introduce a deep-learning approach that estimates the absorbed dose distributions from CT scan data. The deep convolutional neural network model based on U-Net architecture is trained to predict the absorbed dose distribution from CT images. The model is trained in 3 different strategies that utilize datasets in 2D, 2.5D (slice-based), and 3D (image-based) forms. The validation of the proposed method is performed by comparative analysis with the Monte Carlo (MC) simulations for typical dentoalveolar CBCT protocols which consider the anthropomorphic head phantoms as a patient. The proposed approach shows good agreement with the MC method while consuming a significantly lower computational cost. This study will be useful for the development of dental CBCT imaging techniques in terms of patient-specific dose management.
|