# US 2022/0284584 A1 · Computerised tomography image processing > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0284584 A1 - Document ID: US20220284584A1 - Status: Published application - Kind: A1 - Application number: 17/637,279 - Filed: 2020-08-21 - Published: 2022-09-08 - Earliest claimed: 2019-08-23 - Assignees: Oxford University Innovation Limited - Applicants: Oxford University Innovation Limited - Inventors: Regent LEE; Anirudh CHANDRASHEKAR; Vicente GRAU; Ashok HANDA - Web page: https://patentgrep.com/patent/US20220284584A1 ## CPC - G06T 7/0012: {Biomedical image inspection} - G06T 7/11: Region-based segmentation - A61B 6/504: {for diagnosis of blood vessels, e.g. by angiography} - G06T 3/40: Scaling of whole images or parts thereof, e.g. expanding or contracting - G06T 7/174: involving the use of two or more images - G06T 7/30: Determination of transform parameters for the alignment of images, i.e. image registration - G06V 10/764: using classification, e.g. of video objects - G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting - G06V 10/82: using neural networks - G16H 30/40: for processing medical images, e.g. editing - G06T 2207/10081: Computed x-ray tomography [CT] - G06T 2207/20021: Dividing image into blocks, subimages or windows - G06T 2207/20081: Training; Learning - G06T 2207/20084: Artificial neural networks [ANN] - G06T 2207/30048: Heart; Cardiac - G06T 2207/30084: Kidney; Renal - G06T 2207/30092: Stomach; Gastric - G06T 2207/30101: Blood vessel; Artery; Vein; Vascular - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D) - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract Methods for training an algorithm to identify structural anatomical features, for example of a blood vessel, in a non-contrast computed tomography (NCT) image are described herein. The algorithm may comprise an image segmentation algorithm, a random forest classifier, or a generative adversarial network in examples described herein. In one embodiment, a method comprises receiving a labelled training set for a machine learning image segmentation algorithm. The labelled training set comprising a plurality of NCT images, each NCT image of the plurality of NCT images showing a targeted region of a subject, the targeted region including at least one blood vessel. The labelled training set further comprises a corresponding plurality of segmentation masks, each segmentation mask labelling at least one structural feature of a blood vessel in a corresponding NCT image of the plurality of NCT images. The method further comprises training a machine learning image segmentation algorithm, using the plurality of NCT images and the corresponding plurality of segmentation masks, to learn features of the NCT images that correspond to structural features of the blood vessels labelled in the segmentation masks, and output a trained image segmentation model. The method further comprises outputting the trained image segmentation model usable for identifying structural features of a blood vessel in an NCT image. Further methods are described herein for identifying anatomical features from an NCT image, and for establishing training sets. Computing apparatuses and computer readable media are also described herein. ## Claims ### Claim 1 A method for training a machine learning image segmentation algorithm to identify structural features of a blood vessel in a non-contrast computed tomography (NCT) image, the method comprising: receiving a labelled training set for the machine learning image segmentation algorithm, the labelled training set comprising: a plurality of NCT images, each NCT image of the plurality of NCT images showing a targeted region of a subject, the targeted region including at least one blood vessel; and a corresponding plurality of segmentation masks, each segmentation mask labelling at least one structural feature of a blood vessel in a corresponding NCT image of the plurality of NCT images; training a machine learning image segmentation algorithm, using the plurality of NCT images and the corresponding plurality of segmentation masks, to learn features of the NCT images that correspond to structural features of the blood vessels labelled in the segmentation masks, and output a trained image segmentation model; outputting the trained image segmentation model usable for identifying structural features of a blood vessel in an NCT image. ### Claim 2 A method for training a classification algorithm to identify structural features of a blood vessel in a non-contrast computed tomography (NCT) image, the method comprising: receiving a labelled training set for the classification algorithm, the labelled training set comprising: a plurality of NCT images, each NCT image of the plurality of NCT images showing a targeted region of a subject, the targeted region including at least one blood vessel; and a corresponding plurality of segmentation masks, each segmentation mask labelling at least one structural feature of a blood vessel in a corresponding NCT image of the plurality of NCT images; dividing each NCT image and corresponding segmentation mask into sub-volumes, each sub-volume of the corresponding segmentation mask labelling at least one feature in a corresponding sub-volume in the corresponding NCT image; extracting, from each sub-volume of the NCT image, radiomic feature values for a set of radiomic features; training a classification algorithm, using the extracted radiomic feature values for each sub-volume of each NCT image and the corresponding sub-volume of the corresponding segmentation masks, to learn features of the NCT images that correspond to structural features of the blood vessels labelled in the segmentation masks, and output a trained classification model; outputting the trained classification model usable for identifying structural features of a blood vessel in an NCT image. ### Claim 3 (depends on 2) A method according to claim 2, wherein the classification algorithm comprises a random forest classification algorithm. ### Claim 4 A method according to any preceding claim, wherein each segmentation mask has been generated from a corresponding contrast computed tomography (CCT) image, each CCT image corresponding to an NCT image of the plurality of NCT images and showing the features of the blood vessel in the targeted region of the corresponding NCT image. ### Claim 5 A method according to any preceding claim, wherein the method further comprises generating the labelled training set. ### Claim 6 (depends on 23, 24, 25) A method according to any preceding claim, wherein the labelled training set has been established according to any of claims 23 to 25. ### Claim 7 (depends on 1) A method according to claim 1, wherein the machine learning image segmentation algorithm comprises a neural network. ### Claim 8 A method according to any preceding claim, wherein the at least one blood vessel of the targeted region of the non-contrast CT image includes an artery or a vein, or wherein the at least one blood vessel of the targeted region includes the aorta. ### Claim 9 A method according to any preceding claim, wherein the targeted region of the NCT image includes an aortic aneurysm. ### Claim 10 A method according to any preceding claim, wherein the structural features of the blood vessel comprise one or more of: inner lumen, outer wall, intima/media, false lumen, calcification, thrombus, ulceration, atherosclerotic plaques. ### Claim 11 A method according to any preceding claim, wherein each segmentation mask of the plurality of segmentation masks comprises a binary segmentation mask. ### Claim 12 A computer-readable medium having instructions stored thereon which, when executed by one or more processors, cause the one or more processors to implement a method for training a machine learning image segmentation algorithm or to implement a method for training a classification algorithm according to any preceding claim. ### Claim 13 A computer-readable medium having stored thereon computer-readable code representative of a trained image segmentation model for identifying structural features of a blood vessel in a non-contrast computed tomography (NCT) image or having code stored thereon computer-readable code representative of a trained classification algorithm for identifying structural features of a blood vessel in a non-contrast computed tomography (NCT) image. ### Claim 14 (depends on 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) A computing apparatus for training a machine learning image segmentation algorithm or a classification algorithm to identify structural features of a blood vessel in a non-contrast computed tomography (NCT) image, the apparatus comprising: one or more memory units; and one or more processors configured to execute instructions stored in the one or more memory units to perform the method of any of claims 1 to 11. ### Claim 15 A method for identifying structural features of a blood vessel in an unlabelled non-contrast computed tomography (NCT) image, the method comprising: providing the NCT image to a trained image segmentation model, the trained image segmentation model trained to learn features of NCT images that correspond to structural features of blood vessels; generating, using the trained image segmentation model, predicted segmentation data for the provided NCT image, the predicted segmentation data for identifying the features of the blood vessel in the provided NCT image. ### Claim 16 (depends on 1, 15) A method according to claim 15, wherein the trained image segmentation model has been trained according to the method of claim 1. ### Claim 17 (depends on 15, 16) A computer readable medium having stored thereon predicted segmentation data generated using a method according to claim 15 or claim 16. ### Claim 18 (depends on 15, 16) A computing apparatus for identifying structural features of a blood vessel in an unlabelled non-contrast computed tomography (NCT) image, the apparatus comprising: one or more memory units; and one or more processors configured to execute instructions stored in the one or more memory units to perform the method of any of claims 15 to 16. ### Claim 19 A method for identifying structural features of a blood vessel from a non-contrast computed tomography (NCT) image, the NCT image showing a targeted region of a subject, the targeted region including at least one blood vessel, the method comprising: for each of a plurality of sub-volumes of the NCT image: sampling radiomic feature values for a set of radiomic features from the sub-volume; classifying the sub-volume as belonging to a structural feature of a set of structural features based on the sampled radiomic feature values; and identifying, from the classifications of the plurality of sub-volumes of the NCT image, structural features of a blood vessel shown in the targeted region. ### Claim 20 (depends on 19) A method according to claim 19, wherein classifying the sub-volume as belonging to a structural feature of a set of structural features based on the sampled radiomic feature values comprises using a trained classification algorithm. ### Claim 21 (depends on 19) A method according to claim 19, wherein classifying the sub-volume as belonging to a structural feature of a set of structural features based on the sampled radiomic feature values comprises comparing radiomic features with corresponding threshold values. ### Claim 22 (depends on 15, 16, 19, 20, 21) A computer-readable medium having instructions stored thereon which, when executed by one or more processors, cause the one or more processors to implement a method according to any of claims 15-16 or according to any of claims 19-21 to identify structural features of a blood vessel in a non-contrast computed tomography image. ### Claim 23 A method for establishing a labelled training set for training a machine learning image segmentation algorithm or a classification algorithm to identify structural features of a blood vessel in a non-contrast computed tomography (NCT) image, the method comprising: receiving a plurality of NCT images, each NCT image showing a targeted region of a subject, the targeted region including at least one blood vessel; receiving a plurality of contrast computed tomography (CCT) images, each CCT image corresponding to an NCT image of the plurality of NCT images and showing the corresponding targeted region of the subject; adapting the plurality of CCT images to substantially align with the corresponding NCT images; and segmenting the plurality of adapted CCT images to generate a corresponding plurality of segmentation masks, each segmentation mask labelling at least one structural feature of the at least one blood vessel in the corresponding adapted CCT image; wherein the labelled training set includes pairs of NCT images and segmentation masks, each pair comprising a segmentation mask and the NCT image to which the adapted CCT image substantially aligns. ### Claim 24 (depends on 23) A method according to claim 23, wherein adapting the plurality of CCT images comprises orienting the plurality of CCT images to substantially align with the corresponding NCT image. ### Claim 25 (depends on 23, 24) A method according to claim 23 or claim 24, wherein adapting the plurality of CCT images comprises scaling the plurality of CCT images to substantially align with the corresponding NCT images. ### Claim 26 (depends on 23, 24, 25) A labelled training set established according to the method of any of claims 23 to 25. ### Claim 27 (depends on 23, 24, 25) A computer-readable medium having instructions stored thereon which, when executed by one or more processors, cause the one or more processors to implement a method according to any of claims 23 to 25 to establish a training set. ### Claim 28 (depends on 23, 24, 25, 26, 27) A computing apparatus for establishing a labelled training set for training a machine learning image segmentation algorithm or a classification algorithm to identify structural features of a blood vessel in a non-contrast computed tomography (NCT) image for identifying structural features of a blood vessel in an unlabelled non-contrast computed tomography (NCT) image, the apparatus comprising: one or more memory units; and one or more processors configured to execute instructions stored in the one or more memory units to perform the method of any of claims 23 to 27. ### Claim 29 A method for training a generative adversarial network (GAN) to generate a pseudo-contrast computed tomography (PCT) image from a non-contrast computed tomography (NCT) image, the GAN comprising a generator network and a discriminator network, the method comprising: receiving a training set comprising: a plurality of NCT images, each NCT image of the plurality of NCT images showing at least one anatomical structure; and a plurality of contrast computed tomography (CCT) images, each CCT image showing at least one anatomical structure; training the GAN, wherein training the GAN comprises: training the generator network, using the plurality of NCT images and feedback from the discriminator network, to generate PCT images; training the discriminator network, using the generated PCT images and the plurality of CCT images, to classify received images as PCT images or CCT images and to provide feedback to the generator network; and outputting a trained generator model to translate an input NCT image to a PCT image showing at least one anatomical structure. ### Claim 30 (depends on 29) A method according to claim 29, wherein the GAN is a conditional GAN. ### Claim 31 (depends on 29) A method according to claim 29, wherein the GAN is a cycle-GAN. ### Claim 32 (depends on 29, 30, 31) A method according to any of claims 29-31, wherein the generator network comprises a U-NET architecture. ### Claim 33 (depends on 29, 30, 31, 32) A method according to any of claims 29-32, wherein the anatomical structure comprises a blood vessel. ### Claim 34 (depends on 29, 30, 31, 32) A method according to any of claims 29-32, wherein the anatomical structure comprises a kidney. ### Claim 35 (depends on 29, 30, 31, 32) A method according to any of claims 29-32, wherein the anatomical structure comprises a bowel. ### Claim 36 (depends on 29, 30, 31, 32, 33, 34, 35) A computer-readable medium having instructions stored thereon which, when executed by one or more processors, cause the one or more processors to implement a method for training a generative adversarial network (GAN) according to any of claims 29-35. ### Claim 37 A method for identifying anatomical structures in a non-contrast computed tomography (NCT) image, the method comprising: providing the NCT image to a trained generator model, the trained generator model, the generator model trained as part of a generative adversarial network, the generator model trained to translate an input NCT image to a pseudo-contrast computed tomography (PCT) image showing at least one anatomical structure; and generating, using the trained generator model, a PCT image corresponding to the provided NCT image; and identifying, from the PCT image, structural features of the at least one anatomical structure. ### Claim 38 A computer-readable medium having stored thereon computer-readable code representative of a trained generator model to generate a PCT image from an input NCT image showing at least one anatomical structure. ### Claim 39 A method comprising: sending an unlabelled non-contrast computed tomography (NCT) image to a server, the NCT image showing a targeted region of a subject including at least one anatomical structure; and receiving, from the server, predicted segmentation data or a pseudo-contrast computed tomography (PCT) image for the NCT image, the predicted segmentation data or PCT image labelling structural features of the at least one anatomical structure.