Stationary multi-source AI-powered tomography (SMART) image reconstruction method and apparatus
In one embodiment, there is provided a dynamic multi-source image reconstruction apparatus. The apparatus includes a first reconstruction stage, a second reconstruction stage, and a refinement stage. The first reconstruction stage is configured to receive an input data set including a group of data frames. Each data frame corresponds to a respective time step. Each data frame includes a number of projection data sets. Each projection data set corresponds to a respective source-detector pair of a stationary multi-source tomography system. The first reconstruction stage is further configured to reconstruct a first intermediate image based, at least in part, on the group of data frames. The second reconstruction stage is configured to receive a selected data frame and to reconstruct a second intermediate image with a constraint of the first intermediate image as prior. The refinement stage is configured to refine the second intermediate image to produce a three-dimensional output image.
1 . A dynamic multi-source image reconstruction apparatus, wherein the dynamic multi-source image corresponds to dynamic computed tomography (CT) imaging captured by using a plurality of source-detector pairs configured to capture projection data, the apparatus comprising:
a first reconstruction stage configured to receive an input data set comprising a group of data frames, each data frame of the group of data frames corresponding to a respective time step, each data frame comprising the group of data frames comprising a number of projection data sets, each projection data set of the number of projection data sets corresponding to a respective source-detector pair of a stationary multi-source tomography system, the first reconstruction stage further configured to reconstruct a first intermediate image based, at least in part, on the group of data frames;
a second reconstruction stage configured to receive a selected data frame from the group of data frames and to reconstruct a second intermediate image based on the selected data frame with a constraint an image space of the first intermediate image as prior; and
a refinement stage configured to refine the second intermediate image to produce a three-dimensional output image that corresponds to the dynamic multi-source image.
2 . The apparatus of claim 1 , further comprising a preprocessing stage configured to arrange at least a portion of the data frames from the group of data frames in a chronological order or an order based on image structural similarity.
3 . The apparatus of claim 1 , wherein at least one of the first reconstruction stage, the second reconstruction stage and the refinement stage comprises a tensor dictionary.
4 . The apparatus of claim 1 , wherein training at least one of the first reconstruction stage, the second reconstruction stage or the refinement stage is unsupervised.
5 . The apparatus of claim 1 , wherein the first reconstruction stage is configured to implement a compressed sensing-based reconstruction, the compressed sensing-based reconstruction accommodating a structure and an intensity of a plurality of different time frames, and wherein the refinement stage is configured to implement a deep prior refinement, the deep prior refinement facilitating removal of residual image artifacts without relying on a ground truth.
6 . The apparatus of claim 5 , wherein the deep prior refinement is performed based, at least in part, on a deep image prior determined by a deep convolutional network without a ground truth.
7 . The apparatus of claim 1 , wherein the first reconstruction stage is configured to implement a deep prior prediction, the second reconstruction stage is configured to implement a sparsity group prior reconstruction to provide sparsity regularization, and the refinement stage is configured to implement a temporal sensing reconstruction corresponding to the respective time step of the selected data frame.
8 . A method for dynamic multi-source image reconstruction, the dynamic multi-source image corresponding to dynamic computed tomography (CT) imaging captured by using a plurality of source-detector pairs configured to capture projection data, the method comprising:
receiving, by a first reconstruction stage, an input data set comprising a group of data frames, each data frame within the group of data frames corresponding to a respective time step, each data frame within the group of data frames comprising a number of projection data sets, each projection data set within the number of projection data sets corresponding to a respective source-detector pair of a stationary multi-source tomography system;
reconstructing, by the first reconstruction stage, a first intermediate image based, at least in part, on the group of data frames;
receiving, by a second reconstruction stage, a selected data frame;
reconstructing, by the second reconstruction stage, a second intermediate image with a constraint of the first intermediate image as prior; and
refining, by a refinement stage, the second intermediate image to produce a three-dimensional output image that corresponds to the dynamic multi-source image.
9 . The method of claim 8 , further comprising arranging, by a preprocessing stage, at least a portion of the data frames from the group of data frames in a chronological order or an order based on image structural similarity.
10 . The method of claim 8 , wherein at least one of the first reconstruction stage, the second reconstruction stage and the refinement stage comprises a tensor dictionary.
11 . The method of claim 8 , wherein training at least one of the first reconstruction stage, the second reconstruction stage or the refinement stage is unsupervised.
12 . The method of claim 8 , wherein the first reconstruction stage is configured to implement a compressed sensing-based reconstruction accommodating a structure and an intensity of a plurality of different time frames, and wherein the refinement stage is configured to implement a deep prior refinement, the deep prior refinement facilitating removal of residual image artifacts without relying on ground truth.
13 . The method of claim 8 , wherein the first reconstruction stage is configured to implement a deep prior prediction, the second reconstruction stage is configured to implement a sparsity group prior reconstruction to provide sparsity regularization, and the refinement stage is configured to implement a temporal sensing reconstruction corresponding to the respective time step of the selected data frame.
14 . A system for dynamic multi-source image reconstruction, wherein the dynamic multi-source image corresponds to dynamic computed tomography (CT) imaging captured by using a plurality of source-detector pairs configured to capture projection data, the system comprising:
a computing device comprising a processor, a memory, an input/output circuitry, and a data store;
a first reconstruction stage configured to receive an input data set comprising a group of data frames, each data frame within the group of data frames corresponding to a respective time step, each data frame within the group of data frames comprising a number of projection data sets, each projection data set within the number of projection data sets corresponding to a respective source-detector pair of a stationary multi-source tomography system, the first reconstruction stage further configured to reconstruct a first intermediate image based, at least in part, on the group of data frames;
a second reconstruction stage configured to receive a selected data frame from the group of data frames and to reconstruct a second intermediate image based on the selected data frame with a constraint in an image space of the first intermediate image as prior; and
a refinement stage configured to refine the second intermediate image to produce a three-dimensional output image that corresponds to the dynamic multi-source image.
15 . The system of claim 14 , further comprising a preprocessing stage configured to arrange at least a portion of the group of data frames from the group of data frames in a chronological order or an order based on image structural similarity.
16 . The system of claim 14 , wherein at least one of the first reconstruction stage, the second reconstruction stage and the refinement stage comprises a tensor dictionary.
17 . The system of claim 14 , wherein training at least one of the first reconstruction stage, the second reconstruction stage or the refinement stage is unsupervised.
18 . The system of claim 14 , wherein the first reconstruction stage is configured to implement a compressed sensing-based reconstruction, the compressed sensing-based reconstruction accommodating a structure and an intensity of a plurality of different time frames, and wherein the refinement stage is configured to implement a deep prior refinement, the deep prior refinement facilitating removal of residual image artifacts without relying on a ground truth.
19 . The system of claims 14 , wherein the first reconstruction stage is configured to implement a deep prior prediction, the second reconstruction stage is configured to implement a sparsity group prior reconstruction to provide sparsity regularization, and the refinement stage is configured to implement a temporal sensing reconstruction corresponding to the respective time step of the selected data frame.
20 . A computer readable storage device having stored thereon instructions that when executed by one or more processors result in the following operations comprising the method according to claim 8 .