Statistical Model Based Image-domain Multiscale Dual-material Decomposition For Dual-energy CT

Dual-energy CT(DECT)imaging plays an important role in advanced CT imaging applications because of its basis material decomposition capability.Direct signal decomposition via matrix inversion is very sensitive to noise in the projection or image domain,mainly because the energy spectra of linear attenuation coefficients of basis materials are strongly overlapped with each other in the diagnostic x-ray energy range.Besides,medical images contain rich structure and texture information,and the single-scale DECT method cannot achieve a clear distinction in the distribution of the object boundaries and noise behavior of CT images.In this work,we propose to introduce the concept of multi-scale space to the DECT material decomposition,implementing material decomposition in each scale space to obtain an excellent balance between noise and spatial resolution.High-and low-energy CT images are decomposed into the scale space based on the Gaussian linear transformation kernel firstly.And it uses the cost function based on penalized likelihood estimation principle containg of the data fidelity term to restrict the integrality and the regularization term to strengthen the smoothness.The data fidelity term is derived from a negative log-likelihood function,and the statistical weight is determined by a data-based method accounting for the noise variance of high-and low-energy CT images.The decomposed CT images of the same scale are jointly decomposed into two material-specific images using stastical model based method with adjustable penalty parameter for each scale.The final material-specific images are generated by accumulating the material-specific images throughout all scales.The proposed method is evaluated on phantoms and clinical patient data.The results show that the single-scale DECT and the multi-scale DECT methods both substantially reduce the noise in decomposed material images compared with the image-domain direct inversion one.In the line-pair slice of Catphan(?)600 phantom study,the proposed method reduces noise standard deviation(STD)by 93%and 98%for bone and soft-tissue images,respectively.Under a fair comparison of the same noise STD,the proposed multi-scale DECT method with a lower scaling factor(k =20)increases the spatial resolution of soft-tissue image by 28%,while a higher scaling factor(k = 80)increases it by 51%,compared with the single-scale DECT one.In the contrast-rod slice of Catphan(?)600 phantom study,the proposed method reduces noise STD by 97%and 98%for iodine and teflon images,respectively.For a fair comparison,the proposed multi-scale DECT method achieves a lower electron density measurement error compared with the single-scale DECT one.Specifically,multi-scale DECT method with a lower scaling factor(k = 20)reduces the root mean square(RMS)of the average percent error by over 17%,i.e.,from 8.9%to 7.4%.Higher scaling factor(k = 80)further decreases the RMS of the average percent error by 25%,i.e.,from 8.9%to 6.7%.In the pelvis patient data study,the proposed method reduces noise STD by 87%and 92%for bone and soft-tissue images.For a fair comparison,the proposed multi-scale DECT method with a lower scaling factor(k =20)increases the spatial resolution of soft-tissue image by 9%,while a higher scaling factor(k = 80)increases it by 14%,compared with the single-scale DECT one.In the head-and-neck patient data study,the proposed method reduces noise STD by 56%and 71%for bone and soft-tissue images,respectively,since the noise of high-and low-energy CT images is relatively weak.Overall,compared with the image domain direct inversion method,the single-scale DECT and the multi-scale DECT ones reduce the noise STD in decomposed material images by 87%averagely.What’s more,the multi-scale DECT method retains anatomical structure and quantification accuracy in the decomposed material images faithfully.