研究目的
To solve the multi-material decomposition problem in spectral CT by adopting deep learning technique to enlarge the receptive field instead of considering the neighborhood of pixel only.
研究成果
CNN shows its effectiveness to solve multi-material decomposition problem, reducing the MSE of the result by 1~2 orders comparing to direct decomposition. However, there are some problems remaining for further research, such as obtaining training data in real CT systems and increasing the robustness of CNN.
研究不足
The study mentions the need for further research on obtaining training data in real CT systems and improving the robustness of CNN by training with reconstructed images at different spectral.
1：Experimental Design and Method Selection:
A convolutional neural network (CNN) is built to solve the MMD problem. The network is trained with simulated reconstruction images of spectral CT.
2：Sample Selection and Data Sources:
Training data is generated from 15 phantoms similar to each other, and another 100 different phantoms are used as the test set.
3：List of Experimental Equipment and Materials:
Spectral CT equipment with a photon counting detector is used. The X-ray tube spectrum is divided into energy bins, and an ART algorithm is used for reconstruction.
4：Experimental Procedures and Operational Workflow:
The image is reconstructed and discretized into a grid, then cut into patches for training the CNN. The decomposition result is obtained by scanning the image with CNN patch by patch.
5：Data Analysis Methods:
MSE is used to evaluate the quality of the decomposition results.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
