Multi-scale image fusion for x-ray grating-based mammography

Abstract

X-ray phase contrast imaging (PCI) can provide high sensitivity of weakly absorbing low-Z objects in medical and biological fields, especially in mammography. Grating-based differential phase contrast (DPC) method is the most potential PCI method for clinic applications because it can works well with conventional X-ray tube and it can retrieve attenuation, DPC and dark-field information of the samples in a single scanning. Three kinds of information have different details and contrast which represent different physical characteristics of X-rays with matters. Hence, image fusion can show the most desirable characteristics of each image.In this paper, we proposed a multi-scale image fusion for X-ray grating-based DPC mammography. Firstly, non-local means method is adopted for denoising due to the strong noise, especially for DPC and dark-field images. Then, Laplacian pyramid is used for multi-scale image fusion. The principal component analysis (PCA) method is used on the high frequency part and the spatial frequency method is used on the low frequency part. Finally, the fused image is obtained by inverse Laplacian pyramid transform. Our algorithm is validated by experiments. The experiments were performed on mammoDPC instrumentation at the Paul Scherrer Institut in Villigen, Switzerland. The results show that our algorithm can significantly show the advantages of three kinds of information in the fused image, which is very helpful for the breast cancer diagnosis.

HIDE: Hierarchical iterative decoding enhancement for multi‐view 3D human parameter regression

“…Compared to the single‐view method PyMAF, 4 HIDE improves by 9.6 mm on MPJPE and by 5.7 mm on PA‐MPJPE. Compared to the multi‐view method MMT, 31 HIDE improves by 3.5 mm on PA‐MPJPE. After adding the video stream condition, both MPJPE and PA‐MPJPE metrics are further slightly improved.…”