Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.
1.1Simulation from tissue parameters
What ifWhat if a fixed magnetic resonance signal equation generated the synthetic image from measured tissue parameters?
Claim language“a trained machine-learning model”
The scanner acquires baseline relaxation maps and an early series of images during contrast uptake. A compartment model fits the uptake and washout rates for each voxel and estimates the contrast concentration at the requested time. A fixed signal equation converts those concentrations and the baseline relaxation values into image intensities for a specified repetition time and flip angle.
1.2Spatial gradient training targets
What ifWhat if training pairs included spatial gradients of the target image as the second target representation?
Claim language“in frequency space”
For each target image, centered finite differences compute three directional derivatives on the voxel grid using the physical voxel spacing. The trainer applies the same operator to the predicted image and minimizes both the voxel intensity error and the directional derivative error.
1.3Discriminator scores native image patches
What ifWhat if the second training term used a discriminator score on image patches sampled directly from the synthetic image?
Claim language“at least part of a transformed synthetic representation (SRT)”
The generator minimizes absolute intensity error against the paired target image on the registered voxel grid. A discriminator learns to distinguish acquired post-contrast patches from generated patches and provides a second training signal through its scores on the generated patches.
1.5Image fitting with precomputed maps
What ifWhat if a numerical solver received the new images and fitted them using tissue maps generated offline?
Claim language“into the trained machine-learning model”
An offline decoder generates a fixed library of anatomical tissue maps before the examination. At examination time, a solver registers candidate maps to the measured baseline image and fits their mixture weights against early post-contrast frames. A signal simulator uses the fitted maps to render the requested time point.
Section 2 · machine-generated
Variations
Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.
Other field of use
Delayed contrast imaging of porous rock
An industrial X-ray computed tomography scanner acquires registered volumes of porous rock cores before and after a radiopaque tracer enters the pore network. Training uses the baseline and early tracer volumes as inputs and a delayed tracer volume as the target. The training objective combines voxel error with complex Fourier coefficient error between the predicted and measured delayed volumes. The trained model processes baseline and early scans of another core and stores the estimated delayed volume for pore occupancy analysis.
Component substitution
Cosine transform for frequency targets
An orthonormal three-dimensional discrete cosine transform produces the frequency-space representations of the measured late-phase target and the synthetic volume. The transform operates on the same registered voxel grid along all three spatial axes. Training minimizes the sum of mean absolute voxel error and a weighted mean absolute cosine coefficient error, with larger weights on coefficients above one-quarter of the highest spatial frequency along any axis. The trained model generates a voxel volume directly from the early input images.
Parameter or scale variation
Overlapping tiles for volumetric synthesis
A resampler places the input and late-phase target volumes on a grid with 1 mm isotropic spacing. Training uses corresponding tiles of 96 × 96 × 48 voxels and computes both voxel error and Fourier coefficient error, applying the same separable taper to the prediction and target before the Fourier transform. At inference, the model processes tiles with a 24-voxel overlap along each axis. A weighted average blends the overlapping predictions with weights that decrease toward interior tile edges.
Combination with existing systems
Synthetic series in a radiology archive
A synthesis service connects to a radiology image archive and retrieves baseline and early post-contrast series associated with the same examination. The service aligns the series and runs a model trained with voxel-space and frequency-space errors against measured late-phase images. It writes the synthetic volume as a derived series with a separate series identifier, the output voxel geometry, and references to the source series. The archive stores the result under the source examination and makes it available to the review workstation.
Automated or data-driven variant
Automatic selection of early contrast frames
A segmentation model identifies a reference artery in a dynamic contrast series, and software measures the median arterial intensity in each frame. The software estimates baseline variability from pre-injection frames and selects the first frame whose arterial intensity exceeds the baseline mean by three standard deviations, along with the baseline average. The same selector prepares training and inference inputs, while a fixed delayed frame supplies the voxel target and its Fourier transform supplies the frequency target during training. At inference, the synthesis model uses the selected inputs to generate the delayed-phase image.
Prediction variability across independently trained models
Five synthesis models train on separate bootstrap samples of complete examinations, using voxel-space and frequency-space errors against delayed-phase targets. At inference, all five models process the same registered input images on a common voxel grid. The service computes the voxelwise mean as the synthetic volume and the voxelwise sample standard deviation as a map of prediction variability. It stores both volumes on the common grid so their values align during review.
Specific version of a part
Atlas synthesis from measurements
Age, height, body mass, and external body dimensions select and scale a labeled anatomical atlas from a library. A contrast kinetics simulator uses the dose, injection rate, and elapsed time to assign signal intensities to the atlas tissues. The resulting image depicts the scaled atlas anatomy.
Scalar prediction with image rendering
The model estimates the blood-pool concentration at the requested time as one scalar. For a vascular-region view, the simulator applies that value within a vessel mask derived from an acquired image and computes voxel intensities from baseline relaxation values and acquisition settings.
Three-dimensional residual image prediction
A three-dimensional encoder-decoder receives registered pre-contrast and early post-contrast volumes as two channels. It predicts a voxelwise intensity residual that is added to the pre-contrast volume to produce the synthetic image.
Complex fourier target volumes
Each training case contains a pre-contrast input volume and a registered late post-contrast target volume. A three-dimensional discrete Fourier transform converts the target volume into a second target stored as separate real and imaginary coefficient arrays.
Voxel and fourier losses
The trainer computes mean absolute voxel error between the synthetic and acquired post-contrast volumes. It adds a weighted mean squared difference between their complex coefficients after applying the same unitary three-dimensional Fourier transform to both volumes.
Synthetic series transfer and storage
The workstation packages the synthetic slices into a separate image series with voxel spacing, slice positions, and an indicator that identifies the series as synthetic. It stores the series locally and sends it over an encrypted connection to an image archive.
Section 3 · quoted from USPTO records
Other work
Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.
1.1providing a trained machine-learning model (MLMt); the trained machine-learning model (MLMt) having been trained by means of training data (TD) to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object a synthetic representation (SR) of the examination region of the examination object; and
As a result, in addition to the program's 110 ability to query the Contrast Scorecard database 113, 114 for relevant examples, the database 113, 114 can be used by the program 110 to create a simulation device which shows how image quality and the contrast “fingerprint” changes as the different technical variables are modified.
A · US 2011/0282194 A1 · ¶ 143
1.2providing a trained machine-learning model (MLMt); the training data (TD) comprising for each examination object of a multiplicity of examination objects i) at least one input representation (R1, R2) of the examination region of the examination object, ii) a target representation (TR) of the examination region of the examination object and iii) a transformed target representation (TRT); the at least one input representation (R1, R2) representing the examination region of the respective examination object in a first period of time before or after administration of a contrast agent; the target representation (TR) representing the examination region of the respective examination object in a second period of time after the administration of the contrast agent; and the transformed target representation (TRT) representing at least part of the examination region of the respective examination object in frequency space, if the target representation (TR) represents the examination region of the respective examination object in real space, or in real space, if the target representation (TR) represents the examination region of the respective examination object in frequency space;
1.3providing a trained machine-learning model (MLMt); the training of the machine-learning model (MLMt) comprising reducing differences between i) at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) between at least part of a transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT);
1.4receiving at least one input representation (R1*, R2*) of the examination region of a new examination object, the at least one input representation (R1*, R2*) of the examination region of the new examination object representing the examination region in a first period of time before and/or after administration of the contrast agent;
In another case, a more specialized protocol is used which sequentially obtains CT images during different phases of contrast administration (arterial, venous, and excretory phases).
A · US 2011/0282194 A1 · ¶ 68
A projection-image-ascertaining facility for the automatic generation of synthetic projections of an examination object from at least one three-dimensional data set acquired by way of a medical imaging system, the projection-image-ascertaining facility comprising: an input interface for receiving the three-dimensional data set; a computing unit configured for ascertaining position information relating to an arrangement of structures of the examination object based on the thre
B · US 10,445,904 B2 · claim 19
1.5inputting the at least one input representation (R1*, R2*) of the examination region of the new examination object into the trained machine-learning model (MLMt);
1.6receiving a synthetic representation (SR*) of the examination region of the new examination object from the machine-learning model (MLMt); and
In this simulation example, a technologist could select from a number of options (e.g., Contrast Agent) and see how the images, pulmonary arterial opacification, and fingerprint are modified by the program 110 as different contrast agents are selected.
A · US 2011/0282194 A1 · ¶ 143
The three-dimensional data set is used as a basis for determining position information relating to the arrangement of structures of the examination object and at least one synthetic projection based on the position information.
B · US 10,445,904 B2 · abstract
1.7outputting and/or storing the synthetic representation (SR*) of the examination region of the new examination object and/or transmitting the synthetic representation (SR*) of the examination region of the new examination object to a separate computer system.
However, it is in particular possible for the synthetic projections P depicted also to be displayed on the screen 16.
B · US 10,445,904 B2 · ¶ 80
An empty cell means only that no passage was reproduced there.
Statements in these documents
However, the prior art does not focus on the quantitative data analyses which are derived from the time-activity curve generated through contrast administration, nor how this data is correlated with the clinical data.
A · US 2011/0282194 A1 · ¶ 7
This technical data is in turn correlated with the clinical data from the Contrast Scorecard to collectively produce a comprehensive and longitudinal database which tracks all patient, stakeholder, exam, contrast, technology, and institutional data related to the administration of medical imaging contrast agents.
A · US 2011/0282194 A1 · ¶ 8
The entire process is relatively time-consuming and the quality and speed are heavily dependent upon the experience of the respective diagnosing practitioner.
B · US 10,445,904 B2 · ¶ 7
It is accordingly an object of the invention to provide a method and a projection image generating apparatus which overcome the above-mentioned and other disadvantages of the heretofore-known devices and methods of this general type and to provide a simpler method for the generation of synthetic projections and a projection-image-ascertaining facility suitable for this purpose.
B · US 10,445,904 B2 · ¶ 8
Cite aspatent.fyi, US 2026/0354780 A1, claim 1, first published 2026-11-19T03:31:31-05:00, https://patent.fyi/US20260354780/, SHA-256 1c2e0babc063841a9712bd14e9d888adcef4dc3cf058cdba9536797ce07de44f