EuroSafe Imaging 2020
Performed at one institution, Not applicable, Quality assurance, Physics, CT, Radioprotection / Radiation dose, Artificial Intelligence and Machine Learning
T. Higaki, Y. Nakamura, F. Tatsugami, Y. Honda, C. Fujioka, K. Awai
In the conventional dual-energy CT (DECT) imaging with rapid-kV switching, automatic exposure control (AEC) cannot be used  due to limitations of hardware and software. Other DECT acquisition systems such as dual-layer detector CT or dual-source CT can perform the AEC, however, their hardware cost is generally higher than the rapid-kV switching system. If AEC is not available, an optimal radiation dose cannot be delivered to each body portion, especially in a wide range CT scan such as the chest and abdomen. It may result in overexposure on the chest, or insufficient exposure in the abdomen.
Recently, a new rapid-kV switching method that can enable the AEC by using a deep learning technology has been developed. Fig. 1 shows the process flow of the deep learning-based DECT system. Sparse raw data obtained by the rapid-kV switching technique is restored to full view and low noise raw data by processing by a deep convolutional neural network (DCNN). After that, the raw data is reconstructed to image data after material decomposition, and used for various analyses, as in the normal DECT analysis. Fig. 2 shows the training process of the DCNN for raw data restoration. Target data for DCNN training is a full view and low noise raw data. The DCNN is trained to convert sparse, noisy high kV and low kV raw data to full view and low noise raw data. Since the DCNN can correct even the wider gaps introduced by the use of AEC, the deep learning-based rapid-kV switching system can apply AEC despite the rapid-kV switching system.
The use of AEC can be expected to optimize radiation exposure and image quality in whole-body DECT examinations. The purpose of this study is to evaluate the performance of the AEC system of the rapid-kV switching based DECT using a body phantom.