Mammogram is considered as the gold standard by Food and Drug Administration (FDA) to screen for breast cancer, which is the second leading cancer worldwide. In spite of its easy access, conventional mammograms won’t find every cancer due to the limited image contrast mechanism. 

The measurement of X-ray beam refraction in breast tissues has been thought to be the next generation screening technique for breast cancer. This new technique is called X-ray phase contrast imaging (XPCI), which provides better soft tissue differentiation and better tumor detections hence.  

However, the use of X-ray interferometry made from gold and silicon gratings sharply reduces the X-ray dose efficiency, i.e., increasing the patient radiation dose. 

Recently researchers from the Shenzhen Institutes of Advanced Technology (SIAT) of the Chinese Academy of Sciences developed a novel XPCI signal extraction technique using the latest deep learning technique, which has shown promising advantages in enhancing signal accuracy and improving the X-ray radiation dose efficiency. 

The study entitle "Enhancing the X-ray differential phase contrast image quality with deep learning technique" was published in IEEE Transactions on Biomedical Engineering journal, one of the well-recognized top three biomedical technology journals. 

The deep convolutional neural network, named as XP-NET, with a special architecture was designed to automatically perform the XPCI signal retrieval and image quality enhancement in a sequence. 

Results showed that the XP-NET was able to improve the phase signal accuracy by over 15% compared with the conventional analytical method. Additionally, both biological specimen and breast phantom studies have demonstrated that the phase images acquired with half the radiation dose, and processed by the XP-NET showed competed image quality to the reference images acquired with the standard radiation dose level. 

The study demonstrated for the first time that the deep learning technique could help to reduce the radiation dose in X-ray phase contrast imaging, enabled automatic signal extraction and post-processing, and provided evidence for the future potential preclinical uses of high quality breast X-ray phase contrast imaging with lower radiation dose levels. 

Media Contact: 
ZHANG Xiaomin 
Email: xm.zhang@siat.ac.cn