US-12622665-B2 - Adaptive dual-energy x-ray imaging using pre-calibrated weighting factors

US12622665B2US 12622665 B2US12622665 B2US 12622665B2US-12622665-B2

Abstract

Methods for adaptive dual-energy imaging comprise calibrating a fitting model and implementing the calibrated model. Calibrating the model comprises: acquiring high and low energy images of a step phantom, generating regions of interest with overlapping materials, and determining an average intensity for each region of interest in each of the images; and determining a model material cancellation weighting factor and a model noise cancellation weighting factor for each of a first material and a second material for each region of interest. The weighting factors are fit to a fitting model. Implementing the calibrated model comprises: acquiring high and low energy images of a subject and generating maps of a subject-specific material cancellation weighting factor and a subject-specific noise cancellation weighting factor according to the fitting model; and applying the maps of the subject-specific material cancellation weighting factor and the subject-specific noise cancellation weighting factor to the images of the subject.

Inventors

Mike SATTARIVAND
Ivan ROMADANOV

Assignees

DALHOUSIE UNIVERSITY

Dates

Publication Date: 20260512
Application Date: 20230815

Claims (20)

1 . A method for processing x-ray images, the method comprising: obtaining higher energy (HE) and lower energy (LE) x-ray images of a subject, determining intensity values of pairs of corresponding aligned pixels of the HE and LE images respectively, for each of the pairs of corresponding pixels, using the intensity values of the pair to determine first and second material cancellation weighting factors and a noise cancellation weighting factor, creating a first dual energy (DE) x-ray image by combining the HE and LE x-ray images by log subtraction of the intensity values of the pairs of corresponding aligned pixels of the HE and LE x-ray images using the first material cancellation weighting factor corresponding to the pair of corresponding aligned pixels to yield the logarithm of a corresponding pixel value for the first DE x-ray image; creating a second dual energy (DE) x-ray image complementary to the first DE x-ray image by combining the HE and LE x-ray images by log subtraction of the intensity values of the pairs of corresponding aligned pixels of the HE and LE x-ray images using the second material cancellation weighting factor corresponding to the pair of corresponding aligned pixels to yield the logarithm of a corresponding pixel value for the second DE x-ray image; and processing pixels of the first DE x-ray image to reduce noise using the noise cancellation weighting factor corresponding to each of the pixels and the second DE x-ray image.
2 . The method according to claim 1 wherein processing the pixels of the first DE x-ray image to reduce noise comprises generating a noise cancellation image by convolving a logarithm of the second DE x-ray image with a high pass filter and for each of the pixels of the first DE x-ray image subtracting from the logarithm of the corresponding pixel value of the first DE x-ray image a corresponding pixel value of the noise cancellation image weighted by the corresponding noise cancellation weighting factor.
3 . The method according to claim 1 wherein processing the pixels of the first DE x-ray image to reduce noise comprises, computing: ln( DE ACNR )=ln( DE )−ω A (ln( DE C )* h HPF ) where DE ACNR is an array of the pixel values of a noise-reduced version of the first DE x-ray image, DE is an array of the pixel values of the first DE x-ray image, ω A is the corresponding noise cancellation weighting factor, DE C is an array of the pixel values of the second DE x-ray image and *h HPF denotes a convolution with a high-pass filter.
4 . The method according to claim 1 wherein one of the first and second material cancellation weighting factors is a bone cancellation weighting factor and the other one of the first and second material cancellation weighting factors is a soft tissue cancellation weighting factor.
5 . The method according to claim 1 wherein using the intensity values of the pair to determine the first and second material cancellation weighting factors comprises inputting the intensity values of the pair into first and second fitted models which relate pairs of corresponding intensity values of the LE and HE x-ray images to the first and second material cancellation weighting factors respectively.
6 . The method according to claim 5 wherein the first and second fitted models comprise first and second calibration functions that each take as arguments a pair of an intensity value from the LE x-ray image and a corresponding intensity value from the HE x-ray image and the method comprises receiving the first and second material cancellation weighting factors as outputs of the first and second calibration functions respectively.
7 . The method according to claim 5 wherein the first and second fitted models are respectively embodied in first and second lookup tables wherein using the intensity values of the pair to determine the first and second material cancellation weighting factors comprises using the intensity values of the pair as keys for the first and second lookup tables and receiving the first and second material cancellation weighting factors as outputs of the first and second lookup tables respectively.
8 . The method according to claim 5 wherein the first and second fitted models are obtained by: obtaining HE and LE x-ray images of a phantom comprising a plurality of different regions, each of the regions of the phantom comprising a first material, a second material or overlapping first and second overlapping materials; identifying the regions of the phantom in the HE and LE x-ray images of the phantom; determining an average intensity for each of the identified regions in each of the HE and LE x-ray images of the phantom; determining a model material cancellation weighting factor for each of the first material and the second material for each of the regions of the phantom; and fitting the model material cancellation weighting factor for the first material for the regions of the phantom as a function of the average intensities of the HE and LE x-ray images of the phantom for the regions of the phantom to provide the first fitted model; and fitting the model material cancellation weighting factor for the second material for the regions of the phantom as a function of the average intensities of the HE and LE x-ray images of the phantom for the regions of the phantom to provide the second fitted model.
9 . The method according to claim 8 wherein one of the first and second materials is a bone mimicking material and the other one of the first and second materials is a soft tissue mimicking material.
10 . The method according to claim 8 wherein determining the model material cancellation weighting factor for each of the first and second materials is based on achieving a contrast to noise ratio (CNR) of zero between regions in which the first and second materials overlap and regions comprising only the first or second material respectively.
11 . The method according to claim 8 wherein the phantom is a step phantom comprising slabs of soft tissue mimicking material and bone mimicking material wherein each of the regions has a corresponding thickness of the soft tissue mimicking material or a corresponding thickness of the bone mimicking material or a corresponding thickness of the soft tissue mimicking material or a corresponding thickness of the bone mimicking material.
12 . The method according to claim 1 wherein using the intensity values of the pair to determine the noise cancellation weighting factor comprises inputting the intensity values of the pair into a first fitted noise cancellation model which relates pairs of corresponding intensity values of the LE and HE x-ray images to the noise cancellation weighting factor.
13 . The method according to claim 12 wherein the first fitted noise cancellation model comprises a first noise cancellation calibration function that takes as arguments a pair of an intensity value from the LE x-ray image and a corresponding intensity value from the HE x-ray image and the method comprises receiving the noise cancellation weighting factor as an output of the first noise cancellation calibration function.
14 . The method according to claim 12 wherein the first fitted noise cancellation model is embodied in a first noise cancellation lookup table and wherein using the intensity values of the pair to determine the noise cancellation weighting factor comprises using the intensity values of the pair as keys for the first noise cancellation lookup table and receiving the noise cancellation weighting factor as outputs of the first noise cancellation lookup table.
15 . The method according to claim 12 wherein the first fitted noise cancellation model is obtained by: obtaining HE and LE x-ray images of a phantom comprising a plurality of different regions, each of the regions of the phantom comprising a first material, a second material or both the first and second materials overlapping; identifying the regions of the phantom in the HE and LE x-ray images of the phantom; determining an average intensity for each of the identified regions in each of the HE and LE x-ray images of the phantom; determining a first model noise cancellation weighting factor corresponding to the first material for each of the regions of the phantom by an anti-correlated noise reduction (ACNR) method; and fitting the first model material cancellation weighting factors for the first material for the regions of the phantom as a function of the average intensities of the HE and LE x-ray images of the phantom for the regions of the phantom to provide the first fitted noise cancellation model.
16 . The method according to claim 15 further comprising providing a second fitted noise cancellation model, the second fitted noise cancellation model being generated by: determining a second model noise cancellation weighting factor corresponding to the second material for each of the regions of the phantom by the ACNR method; and fitting the second model noise cancellation weighting factors for the second material for the regions of the phantom as a function of the average intensities of the HE and LE x-ray images of the phantom for the regions of the phantom to provide the second fitted noise cancellation model.
17 . The method according to claim 15 wherein determining the first model noise cancellation weighting factors comprises generating a first dual energy (DE) x-ray image of the phantom by combining the HE and LE x-ray images of the phantom by log subtraction of the intensity values of the pairs of corresponding aligned pixels of the HE and LE x-ray images using the first material cancellation weighting factor corresponding to the pair of corresponding aligned pixels to yield the logarithm of a corresponding pixel value for the first DE x-ray image of the phantom and selecting values for the first model noise cancellation weighting factors to maximize the signal-to-noise (SNR) ratio for each region of the phantom in the first DE x-ray image of the phantom.
18 . The method according to claim 15 wherein the phantom is a step phantom comprising slabs of soft tissue mimicking material and bone mimicking material wherein each of the regions has a corresponding thickness of the soft tissue mimicking material or a corresponding thickness of the bone mimicking material or a corresponding thickness of the soft tissue mimicking material or a corresponding thickness of the bone mimicking material.
19 . A method for dual energy x-ray imaging comprising: obtaining higher energy (HE) and lower energy (LE) x-ray images of a subject, based on pixel intensities of the HE and LE x-ray images, generating patient specific maps of material cancellation ω ST,Bn and noise cancellation ω A ST,Bn weighting factors and combining the LE and HE x-ray images to yield a dual-energy (DE) x-ray image using the material cancellation ω ST,Bn and noise cancellation ω A ST,Bn weighting factors.
20 . Apparatus for dual energy (DE) x-ray imaging comprising: first and second fitted material cancellation models respectively corresponding to first and second materials, each of the first and second fitted material cancellation models comprising an input for receiving an intensity value for a pixel of a higher energy (HE) x-ray image and an input for receiving an intensity value for a pixel of a lower energy (LE) x-ray image and configured to output a material cancellation weighting factor corresponding to intensity values presented at the inputs; and a data processor configured to: process higher energy (HE) and lower energy (LE) x-ray images of a subject to obtain intensity values of pairs of corresponding aligned pixels of the HE and LE images respectively, for each of the pairs of corresponding pixels, using the intensity values of the pair as inputs to each of the first and second fitted material cancellation models to obtain corresponding first and second material cancellation weighting factors; and creating a first dual energy (DE) x-ray image by combining the HE and LE x-ray images by log subtraction of the intensity values of the pairs of corresponding aligned pixels of the HE and LE x-ray images using the first material cancellation weighting factor corresponding to the pair of corresponding aligned pixels to yield the logarithm of a corresponding pixel value for the first DE x-ray image; and creating a second dual energy (DE) x-ray image complementary to the first DE x-ray image by combining the HE and LE x-ray images by log subtraction of the intensity values of the pairs of corresponding aligned pixels of the HE and LE x-ray images using the second material cancellation weighting factor corresponding to the pair of corresponding aligned pixels to yield the logarithm of a corresponding pixel value for the second DE x-ray image.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application is a continuation of Patent Cooperation Treaty (PCT) application No. PCT/CA2022/050244 having an international filing date of 18 Feb. 2022, which in turn claims priority from, and for the purposes of the United States of America the benefit under 35 U.S.C. § 119 of, U.S. application No. 63/151,552 filed 19 Feb. 2021. All of the applications in this paragraph are hereby incorporated herein by reference. TECHNICAL FIELD This invention pertains to dual-energy x-ray imaging. Some embodiments of the invention relate to dual-energy x-ray imaging methods which use a phantom for calibrating a model. Some embodiments relate to dual-energy x-ray imaging methods which use pre-calibrated models for generating x-ray images. Some embodiments of the invention relate to apparatus for dual energy x-ray imaging. BACKGROUND Projection x-ray imaging is commonly used in both diagnostic radiography (e.g. to diagnose a cancer lesion) and in image guided radiation therapy (IGRT) to enable precision therapy by identifying the tumor position. Image acquisition can be performed with a variety of techniques such as through the use of a linear accelerator (LINAC) or room mounted kV systems, which produce either volumetric (e.g. cone beam computed tomography) or planar images. The benefits of x-ray imaging can often be limited by overlapping anatomical noise in the projection images, thus obscuring the region of interest (e.g. a tumor position). This may particularly be the case where the region of interest overlaps with bony anatomy. This results in reduced alignment and/or tracking accuracy. One technique aimed at solving this problem is dual-energy (DE) imaging. DE imaging allows for material specific (bone or soft tissue) images to be obtained. DE imaging requires two radiographs obtained with x-ray beams of different spectra. Typically, this is achieved by acquiring images with different x-ray tube potentials. Such radiographs are referred to as high energy (HE) and low energy (LE) images. DE images are obtained by performing a logarithmic subtraction of individual energy images, where one of the images is multiplied by a weighting factor ωST,Bn, which is also referred to as a material cancellation weighting factor (where ST and Bn stand for cancelling bone and soft-tissue, respectively). Mathematically, this can be expressed as: ln(DE)=ln(HE)−ωST,Bn ln(LE) (1) Equation (1) provides for the complete cancellation of a material only with the assumption of a monoenergetic beam, usually referred to as simple log subtraction (SLS). In this case, the optimal weighting factor ωST,Bn is equal to the ratio of linear attenuation coefficients at different energies μHE/μLE for bone or soft tissue. This weighting factor is assumed constant across the image. Clinically used x-ray sources are polychromatic, which results in non-uniform beam hardening, due to different attenuation through anatomical structures of various thicknesses. Therefore, the weighting factors ωST,Bn are desirably optimized in a way to provide full cancellation of a signal of the undesired material (e.g. bone or soft tissue). However, due to the above-mentioned phenomena, it is impossible to completely negate the signal from the cancelled material if a constant weighting factor is used across the entire image. This results in a DE image with artifacts and reduced image quality. Another important factor in DE images is noise. DE images typically have noise contributions from both HE and LE images. There are a variety of possible methods for reducing noise, such as simple smoothing of the high energy image and anti-correlated noise reduction (ACNR), for example. ACNR utilizes the anti-correlation of noise on the material-specific images. For example, noise on the bone only image is anti-correlated to the noise on the soft tissue only image. Mathematically, the ACNR algorithm can be expressed as: ln(DEACNR)=ln(DE)−ωA(ln(DEC)*hHPF) (2) where DE is the DE image obtained using Equation (1), ωA is the ACNR weighting factor, DEC is the complimentary material image (e.g. bone for soft-tissue and vice versa), and*hHPF denotes a convolution with a high-pass filter. However, since the noise cancellation weighting factor ωA is assumed to be constant across the image, this method does not take into account spatial variations of the noise arising from various quantum noise across the image due to different attenuating material thicknesses. Methods have been proposed which involve applying spatially varying weighting factors based on a priori CT scans of patients in order to provide the material distribution of the imaged area. A priori CT scans are not always available and accurate image registration is time consuming and can be problematic if patient anatomy changes relative to the CT images. Despite the work that has been done in the field of DE imaging to date there remains a need for DE imaging technologies which improve on t