# US 11,906,608 B2 · Automated detection of water-fat swaps in Dixon magnetic resonance imaging > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,906,608 B2 - Document ID: US11906608B2 - Status: Granted - Kind: B2 - Application number: 17/604,454 - Filed: 2020-04-09 - Published: 2022-06-23 - Granted: 2024-02-20 - Earliest claimed: 2019-04-19 - Assignees: Koninklijke Philips N.V. - Applicants: Koninklijke Philips N.V. - Inventors: Karsten Sommer; Steffen Weiss; Holger Eggers - Web page: https://patentgrep.com/patent/US11906608B2 ## CPC - G01R 33/4828: {Resolving the MR signals of different chemical species, e.g. water-fat imaging} - G01R 33/5608: {Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels (image data processing or generation, in general G06T)} - G01R 33/56563: {caused by a distortion of the main magnetic field B0, e.g. temporal variation of the magnitude or spatial inhomogeneity of B0 (G01R33/56509, G01R33/56518, G01R33/56536 take precedence)} - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N 3/09: Supervised learning - G01R 33/56527: {due to chemical shift effects} - G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Disclosed herein is a medical system (100, 300, 500) comprising a memory (110) storing machine executable instructions (120) and a convolutional neural network (122). The convolutional neural network is configured for receiving an initial Dixon magnetic resonance image (124, 126) as input. The convolutional neural network is configured for identifying one or more water-fat swap regions (128) in the initial Dixon magnetic resonance image. The medical system further comprises a processor (104) for controlling the medical system. Execution of the machine executable instructions causes the processor to: receive (200) the initial Dixon magnetic resonance image; and receive (204) the one or more water-fat swap regions from the convolutional neural network in response to inputting the initial Dixon magnetic resonance image into the convolutional neural network. ## Claims ### Claim 1 A medical system comprising: a memory configured to store machine executable instructions and a convolutional neural network previously trained using training data comprising training Dixon magnetic resonance images and training water-fat swap masks, wherein the convolutional neural network is configured to receive an initial Dixon magnetic resonance image as input, wherein the convolutional neural network is configured to identify one or more water-fat swap regions in the initial Dixon magnetic resonance image; and a processor configured to control the medical system, wherein execution of the machine executable instructions causes the processor to: receive the initial Dixon magnetic resonance image; receive the one or more water-fat swap regions from the convolutional neural network in response to inputting the initial Dixon magnetic resonance image into the convolutional neural network; and reconstruct a corrected Dixon magnetic resonance image using the received one or more water-fat swap regions. ### Claim 2 A medical system comprising: a memory configured to store machine executable instructions and a convolutional neural network, wherein the convolutional neural network is configured to receive an initial Dixon magnetic resonance image as input, wherein the convolutional neural network is configured to identify one or more water-fat swap regions in the initial Dixon magnetic resonance image; and a processor configured to control the medical system, wherein execution of the machine executable instructions further causes the processor to: receive training data, wherein each of the training data comprises a training Dixon magnetic resonance image and a training water-fat swap mask, wherein the training water-fat swap mask identifies a ground truth location of the one or more water-fat swap regions; train the convolutional neural network using the training data; receive the initial Dixon magnetic resonance image; and receive the one or more water-fat swap regions from the convolutional neural network in response to inputting the initial Dixon magnetic resonance image into the convolutional neural network. ### Claim 3 A non-transitory computer readable medium product storing machine executable instructions for execution by a processor controlling a medical system and a convolutional neural network previously trained using training data comprising training Dixon magnetic resonance images and training water-fat swap masks, wherein the convolutional neural network is configured for receiving an initial Dixon magnetic resonance image as input, wherein the convolutional neural network is configured for identifying one or more water-fat swap regions in the initial Dixon magnetic resonance image, wherein execution of the machine executable instructions causes the processor to: receive the initial Dixon magnetic resonance image; and receive the one or more water-fat swap regions from the convolutional neural network in response to inputting the initial Dixon magnetic resonance image into the convolutional neural network. ### Claim 4 A method of training a convolutional neural network, wherein the convolutional neural network is configured for receiving an initial Dixon magnetic resonance image as input, wherein the convolutional neural network is configured for identifying one or more water-fat swap regions in the initial Dixon magnetic resonance image, wherein the method comprises receiving training data, wherein each of the training data comprises a training Dixon magnetic resonance image and a training water-fat swap mask, wherein the training Dixon magnetic resonance image comprises the one or more water-fat swap regions, wherein the training water-fat swap mask identifies a ground truth location of the one or more water-fat swap regions; and training the convolutional neural network using the training data. ### Claim 5 (depends on 1) The medical system of claim 1, wherein the corrected Dixon magnetic resonance image is reconstructed using any one of the following: wherein the Dixon magnetic resonance image further comprises a fat image and a water image, wherein the corrected Dixon magnetic resonance image is reconstructed by swapping voxels of the one or more water-fat swap regions between the fat image and the water image; and wherein the corrected Dixon magnetic resonance imaging is reconstructed from the Dixon magnetic resonance imaging data according to a Dixon magnetic resonance image reconstruction algorithm, wherein the Dixon magnetic resonance image reconstruction algorithm is configured to use the one or more water-fat swap regions as constraints and/or for correcting a B0 inhomogeneity estimation. ### Claim 6 (depends on 2) The medical system of claim 2, wherein execution of the machine executable instructions further causes the processor to: receive swap free Dixon magnetic resonance images; and generate the training data by generating one or more synthetic water-fat swap regions in each swap free Dixon magnetic resonance image of the swap free Dixon magnetic resonance images and constructing the training water-fat swap mask. ### Claim 7 (depends on 6) The medical system of claim 6, wherein the training data is at least partially generated by determining at least some of the one or more synthetic water-fat swap regions by: determining one or more random variables; inputting the one or more random variables into a spatially variable function; and thresholding the spatially variable function to determine the one or more water-fat swap regions for the training Dixon magnetic resonance image. ### Claim 8 (depends on 6) The medical system of claim 6, wherein the training data is at least partially generated by: identifying high water-fat zones in the Dixon magnetic resonance image with both a fat content and a water contend above a predetermined water-fat threshold; and selecting at least a portion of the high water-fat zones as the synthetic water-fat swap regions. ### Claim 9 (depends on 6) The medical system of claim 6, wherein the training data is at least partially generated by: receiving a B0 inhomogeneity map for the swap free Dixon magnetic resonance images; identifying high B0 inhomogeneity zones in the swap free Dixon magnetic resonance images by thresholding the B0 inhomogeneity map with a B0 inhomogeneity threshold; and selecting at least a portion of the high B0 inhomogeneity zones as the synthetic water-fat swap regions. ### Claim 10 (depends on 7) The medical system of claim 7, wherein the training data is at least partially generated by: identifying tissue boundary zones in the swap free Dixon magnetic resonance image using an image segmentation algorithm; and selecting at least a portion of the tissue boundary zones as the synthetic water-fat swap regions. ### Claim 11 (depends on 7) The medical system of claim 7, wherein the training data is at least partially generated by: determining a spatially dependent signal to noise map for the Dixon magnetic resonance image; and selecting at least a portion of the synthetic water-fat swap regions for inclusion in the training data using a signal to noise weighting factor determined using the spatially dependent signal to noise map, wherein the signal to noise weighting factor increases as the spatially dependent signal to noise map decreases. ### Claim 12 (depends on 6) The medical system of claim 6, wherein the training data is at least partially generated by: receiving an isocenter location for the swap free Dixon magnetic resonance image; and selecting at least a portion of synthetic water-fat swap regions for inclusion in the training data using a distance weighting factor determined by a distance from the isocenter location, wherein the distance weighting factor increases as the distance from the isocenter location increases. ### Claim 13 (depends on 2) The medical system of claim 2, wherein the training data is at least partially generated by increasing size of the training data by generating additional training ground truth magnetic resonance images by applying an image transformation to both the training Dixon magnetic resonance images as well as the training water-fat swap mask, wherein the image transformation comprises at least one of an image translation, an image rotation, an image deformation, an image flipping transformation, or a mirror image transformation. ### Claim 14 (depends on 1) The medical system of claim 1, further comprising: a magnetic resonance imaging system, wherein the memory further contains pulse sequence commands configured for controlling the magnetic resonance imaging system to acquire magnetic resonance imaging data according to a Dixon magnetic resonance imaging protocol, wherein execution of the machine executable instructions further cause the processor to: acquire the magnetic resonance imaging data by controlling the magnetic resonance imaging system with the pulse sequence commands; and reconstruct the initial Dixon magnetic resonance image using the magnetic resonance imaging data according to the Dixon magnetic resonance imaging protocol. ### Claim 15 (depends on 4) The method of claim 4, further comprising: receiving swap free Dixon magnetic resonance images; and generating the training data by generating synthetic water-fat swap regions in the swap free Dixon magnetic resonance images. ### Claim 16 (depends on 2) The medical system of claim 2, wherein execution of the machine executable instructions further causes the processor to reconstruct a corrected Dixon magnetic resonance image using the received one or more water-fat swap regions. ### Claim 17 (depends on 16) The medical system of claim 16, wherein the corrected Dixon magnetic resonance image is reconstructed by: swapping voxels of the one or more water-fat swap regions between a fat image and a water image included in the Dixon magnetic resonance image; or using one or more water-fat swap regions as constraints and/or for correcting a B0 inhomogeneity estimation to reconstruct the corrected Dixon magnetic resonance image from the Dixon magnetic resonance imaging data according to a Dixon magnetic resonance image reconstruction algorithm. ### Claim 18 (depends on 3) The non-transitory computer readable medium product of claim 3, wherein execution of the machine executable instructions further causes the processor to reconstruct a corrected Dixon magnetic resonance image using the received one or more water-fat swap regions. ### Claim 19 (depends on 18) The non-transitory computer readable medium product of claim 18, wherein the Dixon magnetic resonance image is corrected using any one of the following: wherein the Dixon magnetic resonance image further comprises a fat image and a water image, wherein the corrected Dixon magnetic resonance image is reconstructed by swapping voxels of the one or more water-fat swap regions between the fat image and the water image; and wherein the corrected Dixon magnetic resonance imaging is reconstructed from the Dixon magnetic resonance imaging data according to a Dixon magnetic resonance image reconstruction algorithm, wherein the Dixon magnetic resonance image reconstruction algorithm is configured to use the one or more water-fat swap regions as constraints and/or for correcting a B0 inhomogeneity estimation. ### Claim 20 (depends on 3) The non-transitory computer readable medium product of claim 3, further storing pulse sequence commands configured for controlling a magnetic resonance imaging system to acquire magnetic resonance imaging data according to a Dixon magnetic resonance imaging protocol, wherein execution of the machine executable instructions further causes the processor to: acquire the magnetic resonance imaging data by controlling the magnetic resonance imaging system with the pulse sequence commands; and reconstruct the initial Dixon magnetic resonance image using the magnetic resonance imaging data according to the Dixon magnetic resonance imaging protocol.