System and method for automatically predicting inversion time in magnetic resonance imaging

Abstract

Systems and methods for contrast scanning are provided. In one example, a system includes a display device (234), one or more processors (204), and a memory (206) storing instructions executable by the one or more processors to obtain (304) a series of TI scout images, each TI scout image of the series of TI scout images obtained with a different inversion time; process (408) at least a portion of each TI scout image of the series of TI scout images to generate a similarity metric map that depicts a degree of statistical similarity between each pair of TI scout images of the series of TI scout images in a region of interest (ROI); predict (418, 422) an inversion time or an inversion time range based on the similarity metric map; and display (308) the predicted inversion time or the inversion time range on the display device.

Description

Technical Field

[0001] This specification relates generally to medical imaging. More specifically, this disclosure relates to contrast-enhanced magnetic resonance imaging. Background Technology

[0002] Magnetic resonance imaging (MRI) is a medical imaging modality that creates images of the inside of the human body without using X-rays or other ionizing radiation. An MRI scan typically consists of a series of radio frequency (RF) excitation pulses and magnetic field gradient pulses, which are emitted at specific timings in a specific sequence to prepare for contrast and encode spatial information into signals to generate an image. To enhance certain anatomical features, some MRI scans may include the administration of contrast agents to the subject being imaged. Summary of the Invention

[0003] In one example, a system includes a display device, one or more processors, and a memory storing instructions executable by the one or more processors to obtain a series of TI positioning images, each of which is obtained using a different inversion time; processing at least a portion of each of the TI positioning images in the series to generate a similarity metric map depicting the degree of statistical similarity between each pair of TI positioning images in the series within a region of interest (ROI); predicting an inversion time or a range of inversion times based on the similarity metric map; and displaying the predicted inversion time or the range of inversion times on the display device.

[0004] It should be understood that the above description of the invention is provided to present a series of concepts further described in the detailed embodiments in a simplified form. This is not intended to identify key or essential features of the claimed subject matter, the scope of which is uniquely defined by the claims following the detailed embodiments. Furthermore, the claimed subject matter is not limited to specific implementations that address any shortcomings pointed out above or in any part of this disclosure. Attached Figure Description

[0005] This disclosure will be better understood by referring to the following description of non-limiting embodiments, in which:

[0006] Figure 1 This is a block diagram of an MRI apparatus according to one embodiment of the present disclosure;

[0007] Figure 2 schematically shown Figure 1 An exemplary scan control device for an MRI apparatus;

[0008] Figure 3 This is a high-level flowchart of a method for contrast scanning according to an embodiment of the present disclosure;

[0009] Figure 4 This is a flowchart of a method for predicting reversal time from a series of positioning images according to an embodiment of the present disclosure;

[0010] Figure 5 It schematically shows the following based on Figure 4 The method is used to generate cropped image frames from a series of localized images using a region of interest (ROI) detector;

[0011] Figure 6 It schematically shows the following based on Figure 4 The method is used for Figure 5 The process of cropping image frames and using a mask predictor to predict the time reversal;

[0012] Figure 7 It shows that according to Figure 5 and Figure 6 The process generates an exemplary similarity metric graph from a series of localized images;

[0013] Figure 8 It shows from Figure 7 The selected prediction reversal time range in the similarity metric graph;

[0014] Figure 9 The first exemplary series of positioning images and predicted reversal time are shown;

[0015] Figure 10 A second exemplary series of positioning images and predicted reversal time are shown;

[0016] Figure 11 This is a flowchart illustrating an exemplary method for training a mask predictor according to an embodiment of the present disclosure; and

[0017] Figure 12 It schematically shows the following based on Figure 11 The method is used for training a mask predictor. Detailed Implementation

[0018] The following description relates to medical imaging, and more specifically to contrast-enhanced magnetic resonance (MR). Contrast-enhanced MR imaging (MRI) scans involve administering a paramagnetic contrast agent, such as a gadolinium-based contrast agent, to the imaging subject to enhance contrast in various tissues of the subject. For example, contrast-enhanced MRI scans can be deployed to visualize myocardial scarring / area ischemia following a myocardial infarction. These scans, known as delayed-enhanced cardiac MRI or delayed-enhanced myocardial (MDE) MRI, allow visualization of fibrotic / scarring areas because the scarring area exhibits a larger accumulation of contrast agent than adjacent normal tissue, which is represented as a high-intensity signal area with a shorter longitudinal relaxation (T1) time.

[0019] Typically, MDE scans are performed using an inversion recovery (IR) gradient echo (GE) sequence 10 to 15 minutes after contrast agent (e.g., gadolinium) administration. In the IR sequence, the IR pulse inverts the longitudinal magnetization of different tissue types. Magnetization recovery differs in each tissue type. For example, the primary tissue types of interest in cardiac MRI are typically normal myocardium, infarcted myocardium (infarction), and blood. The recovery rate of a particular tissue can be expressed as the reciprocal of its longitudinal recovery time T1. A time delay, called inversion time (TI), is inserted between the IR pulse and the data acquisition sequence, such that the magnetization of one of these tissues (e.g., normal myocardium in a specific case of cardiac MRI) is approximately zero at acquisition time. This is often referred to as “zeroing” a particular tissue and can generate improved T1 imaging contrast between the zeroed tissue and other tissue types, such as infarcted myocardium.

[0020] During MDE scanning, the operator can set the reversal time for the tissue to be zeroed. However, the optimal reversal time can vary from patient to patient. To determine the appropriate reversal time for an individual, a series of TI localizations is obtained, where each image in the series has a progressively larger TI. In practice, TI selection is typically performed through visual examination and by choosing the optimal TI from the TI localization series. This approach depends on the skill of the technician or physician in selecting the optimal reversal time, and such skilled clinicians may not be readily available outside of specialized centers. Furthermore, motion artifacts or metallic artifacts can make the selection of the optimal TI challenging. Therefore, consistent selection of the reversal time is often a significant issue, especially when different technicians select different TIs. In practice, this problem is often a common cause of patient recalls, delaying diagnosis and reducing scanning efficiency.

[0021] Therefore, methods for automatically selecting the optimal reversal time for patients based on TI positioning series have been developed. One such method utilizes a sliding window approach to define subgroups of image frames within a TI positioning series. These subgroups are processed using a deep learning-based image classifier to identify spatial features within the image group that allow the subgroup to be separated into “before” and “after” categories corresponding to points corresponding to the location of the optimal image within the TI positioning series. However, current methods for automatically selecting reversal time suffer from various drawbacks. As an example, different imaging facilities may utilize different schemes to acquire TI positioning series, resulting in varying numbers of image frames between scans. This necessitates determining the optimal reversal time multiple times via a sliding window, which is inefficient. Furthermore, disease-related anomalies and imaging-related artifacts (motion, metallicity) can cause variations in imaging data and thus variations in TI selection. Therefore, robust algorithms should be able to address such variations in the data. If a machine learning-based approach is used to select the TI, all such variations (including different cardiac views) must be explicitly included in the training pool, which expands the required training data pool to a level where assembling and labeling the training data may be impractical. Furthermore, deep learning models for recognizing spatial features in images are typically large and require significant computational resources.

[0022] Therefore, according to the implementation disclosed herein, the aforementioned problem can be addressed by locating the region to be evaluated to determine the optimal TI point in the TI localization image (and generating cropped frames of the localized region only), then converting the cropped frames to a probability density function (PDF) and calculating the differences between frames as a combination of measures of the selected statistical distance. This method projects image frame data into a more interpretable measurement space. Measurements in this space focus on changes between frames, highlighting key transitions or zeroing points more clearly than image frames, while being less sensitive to imaging-related artifacts (such as noise, metallic artifacts, etc.) in the image frames. The baseline ground truth TI location can be highlighted as a binary mask. The task of any AI solution is to segment this mask and use the mask's location along the temporal direction to generate the optimal TI. A shallow encoder-decoder network is employed to segment the unit region around the TI point in the temporal statistical distance map. This now becomes a “one-way” optimal TI determination method; where the measurement space is used to determine the TI in the blood pool and myocardium, regardless of the subjectivity of the TI-Scout scheme, the cardiac view, artifacts in the myocardial and blood pool regions, and / or changes in patient condition.

[0023] The method disclosed herein includes the following steps. First, deep learning is used to coarsely locate the cardiac region on each TI localization image (which may be a maximum intensity projection (MIP) image) to create clipped time frames. Statistical distance fraction maps (referred to herein as similarity metric maps) are generated, which effectively measure the affinity between clipped TI frames, and any changes in these maps are likely to occur near the optimal TI region. Automated analysis of these maps provides approximate regions of “phase changes,” including the zeroing-to-initial recovery from the blood pool to the myocardium. This analysis provides an interpretable similarity signal across frames and helps in understanding the model’s decision-making process. This interpretability is often lacking in models using imaging data, where the relationship between inputs and predictions may be less transparent. Finally, the TI estimate from the above process is combined with a systematic pulse sequence design (PSD) to provide the user with the best possible TI time. A PSD is a set of instructions to be executed by the MRI device to perform a scan, and therefore combining the TI estimate with the systematic PSD may include generating a PSD for an MDE scan, which includes the TI between IR pulses and the data acquisition sequence, and the PSD is used to instruct the MRI device to perform the MDE scan. Therefore, once TI is predicted, it is input as a value into the PSD used for MDE scanning.

[0024] Figure 1 An exemplary MRI apparatus is shown that can be used to acquire images of an imaging subject during contrast-enhanced MRI scans. The MRI apparatus may include scan control devices, such as… Figure 2 The scanning control device is configured to process data from an MRI unit to form images, predict optimal reversal times based on the TI positioning series obtained using the MRI unit, command actions of the MRI unit (e.g., start TI positioning scan acquisition, command the reversal time to be used in MDE acquisition, start MDE acquisition, etc.), and display a scanning interface so that the user can monitor the contrast agent level of the imaging subject during the scan, set scan prescription parameters, confirm or select the optimal reversal time, etc. The scanning control device can implement various methods to facilitate contrast scanning, such as... Figure 3 and Figure 4 The method shown includes prediction of the optimal reversal time in the contrast scan. Figure 5 and Figure 6 The process is illustrated schematically. In some examples, the method discussed in this paper can identify the range of inversion times rather than predicting a single inversion time, such as... Figure 7 and Figure 8 As shown. Figure 9 and Figure 10 The image shows TI localization images that can be obtained using an MRI device and predicted TI frames identified according to the method disclosed herein. The predicted inversion time can be determined by... Figure 11 The mask predictor is generated by training the method.

[0025] Figure 1 An MRI apparatus 10 (e.g., an MRI system) is illustrated, comprising a static magnetic field magnet unit 12, a gradient coil unit 13, an RF coil unit 14, an RF body coil unit 15 (e.g., a volume coil unit), a transmit / receive (T / R) switch 20, an RF driver unit 22, a gradient coil driver unit 23, a data acquisition unit 24, a controller unit 25, a patient bed or examination table 26, a data processing unit 31, a scan control device 32, and a display unit 33. In some embodiments, the RF coil unit 14 is a surface coil, which is a local coil typically placed near the anatomical structures of interest of the subject 16. Herein, the RF body coil unit 15 is a transmitting coil that transmits RF signals, and the local surface of the RF coil unit 14 receives MR signals. Therefore, the transmitting body coil (e.g., the RF body coil unit 15) and the surface receiving coil (e.g., the RF coil unit 14) are independent but electromagnetically coupled components. The MRI apparatus 10 sends electromagnetic pulse signals to a subject 16 placed in an imaging space 18 that forms a static magnetic field to perform a scan to obtain magnetic resonance signals from the subject 16. One or more images of the subject 16 can be reconstructed based on the magnetic resonance signals obtained by the scan.

[0026] The static magnetic field magnet unit 12 includes, for example, a toroidal superconducting magnet mounted within a toroidal vacuum container. The magnet defines a cylindrical space surrounding the subject 16 and generates a constant main static magnetic field B0.

[0027] The MRI apparatus 10 also includes a gradient coil unit 13 that forms a gradient magnetic field in the imaging space 18 to provide three-dimensional positional information for magnetic resonance signals received by the RF coil array. The gradient coil unit 13 includes three gradient coil systems, each generating a gradient magnetic field along one of three spatial axes perpendicular to each other, and generating gradient fields in each of the frequency encoding direction, phase encoding direction, and slice selection direction, depending on the imaging conditions. More specifically, the gradient coil unit 13 applies a gradient field in the slice selection direction (or scan direction) of the subject 16 to select slices; and the RF body coil unit 15 or the local RF coil array can send RF pulses to the selected slices of the subject 16. The gradient coil unit 13 also applies a gradient field in the phase encoding direction of the subject 16 to perform phase encoding of the magnetic resonance signals from the RF pulse-excited slices. Then, the gradient coil unit 13 applies a gradient field in the frequency encoding direction of the subject 16 to perform frequency encoding of the magnetic resonance signals from the RF pulse-excited slices.

[0028] RF coil unit 14 is configured, for example, to surround the imaging region of subject 16. In some examples, RF coil unit 14 may be referred to as a surface coil or receiving coil. In the static magnetic field space or imaging space 18 in which the static magnetic field B0 is formed by static magnetic field magnet unit 12, RF coil unit 15 sends RF pulses as electromagnetic waves to subject 16 based on control signals from controller unit 25, thereby generating a high-frequency magnetic field B1. This excites proton spins in the slice to be imaged in subject 16. RF coil unit 14 receives electromagnetic waves generated as magnetic resonance signals when the proton spins thus excited in the slice to be imaged in subject 16 return to alignment with the initial magnetization vector. In some embodiments, RF coil unit 14 may transmit RF pulses and receive MR signals. In other embodiments, RF coil unit 14 may be used only to receive MR signals without transmitting RF pulses.

[0029] The RF body coil unit 15 is configured, for example, to surround the imaging space 18 and generate RF magnetic field pulses within the imaging space 18 that are orthogonal to the main magnetic field B0 generated by the static magnetic field magnet unit 12 to excite the nucleus. The RF body coil unit 15 is fixedly attached to and connected to the MRI apparatus 10, unlike the RF coil unit 14, which can be disconnected from the MRI apparatus 10 and replaced by another RF coil unit. Furthermore, while local coils (such as the RF coil unit 14) can only send or receive signals to or from a local area of ​​the subject 16, the RF body coil unit 15 typically has a larger coverage area. For example, the RF body coil unit 15 can be used to send or receive signals to or from the whole body of the subject 16. Using only receiving local coils and transmitting body coils provides uniform RF excitation and good image homogeneity, at the cost of higher RF power deposited in the subject. For transmit-receive local coils, the local coil provides RF excitation to the region of interest and receives the MR signal, thereby reducing the RF power deposited in the subject. It should be understood that the specific use of the RF coil unit 14 and / or the RF body coil unit 15 depends on the imaging application.

[0030] When operating in receive mode, T / R switch 20 selectively connects RF body coil unit 15 to data acquisition unit 24, and when operating in transmit mode, it selectively connects the RF body coil unit to RF driver unit 22. Similarly, when RF coil unit 14 operates in receive mode, T / R switch 20 selectively connects RF coil unit 14 to data acquisition unit 24, and when the RF coil unit operates in transmit mode, it selectively connects the RF coil unit to RF driver unit 22. When both RF coil unit 14 and RF body coil unit 15 are used for a single scan, for example, if RF coil unit 14 is configured to receive MR signals and RF body coil unit 15 is configured to transmit RF signals, T / R switch 20 can direct control signals from RF driver unit 22 to RF body coil unit 15 while simultaneously directing the received MR signals from RF coil unit 14 to data acquisition unit 24. The coil of RF coil unit 15 can be configured to operate in transmit-only mode or transmit-receive mode. The coil of RF coil unit 14 can be configured to operate in transmit-receive mode or receive-only mode.

[0031] The RF driver unit 22 includes a gate modulator (not shown), an RF power amplifier (not shown), and an RF oscillator (not shown), which drive an RF coil (e.g., RF body coil unit 15) and generate a high-frequency magnetic field in the imaging space 18. Based on a control signal from the controller unit 25 and using the gate modulator, the RF driver unit 22 modulates the RF signal received from the RF oscillator into a signal with a predetermined timing and a predetermined envelope. The RF signal modulated by the gate modulator is amplified by the RF power amplifier and then output to the RF body coil unit 15.

[0032] The gradient coil driver unit 23 drives the gradient coil unit 13 based on control signals from the controller unit 25, thereby generating a gradient magnetic field in the imaging space 18. The gradient coil driver unit 23 includes three systems (not shown) of driver circuits corresponding to the three gradient coil systems included in the gradient coil unit 13.

[0033] The data acquisition unit 24 includes a preamplifier (not shown), a phase detector (not shown), and an analog-to-digital converter (not shown) for acquiring the magnetic resonance signal received by the RF coil unit 14. In the data acquisition unit 24, the phase detector uses the output of the RF oscillator from the RF driver unit 22 as a reference signal to perform phase detection on the magnetic resonance signal received from the RF coil unit 14 and amplified by the preamplifier. The phase-detected analog magnetic resonance signal is then output to the analog-to-digital converter for conversion into a digital signal. The resulting digital signal is then output to the data processing unit 31.

[0034] The MRI apparatus 10 includes an examination table 26 for placing a subject 16 thereon. The subject 16 can be moved inside and outside the imaging space 18 by moving the examination table 26 based on control signals from the controller unit 25.

[0035] The controller unit 25 includes a computer and a recording medium on which a program to be executed by the computer is stored. When executed by the computer, the program causes various parts of the device to perform operations corresponding to a predetermined scan. The recording medium may include, for example, a ROM, floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, or non-volatile memory card. The controller unit 25 is connected to the scan control device 32 and processes the operation signals input to the scan control device 32, and also controls the inspection table 26, RF driver unit 22, gradient coil driver unit 23, and data acquisition unit 24 by outputting control signals to them. The controller unit 25 also controls the data processing unit 31 and the display unit 33 based on the operation signals received from the scan control device 32 to obtain the desired image.

[0036] The scanning control device 32 includes user input devices such as a touchscreen, keyboard, and mouse. The operator uses the scanning control device 32 to, for example, input data as an imaging scheme and set the area to be processed in the imaging sequence. Data regarding the imaging scheme and the area to be processed in the imaging sequence is output to the controller unit 25.

[0037] The data processing unit 31 includes a computer and a recording medium on which a program to be executed by the computer to perform predetermined data processing is recorded. The data processing unit 31 is connected to the controller unit 25 and performs data processing based on control signals received from the controller unit 25. The data processing unit 31 is also connected to the data acquisition unit 24 and generates spectral data by applying various image processing operations to the magnetic resonance signal output from the data acquisition unit 24.

[0038] Display unit 33 includes a display device and displays images on the display screen of the display device based on control signals received from controller unit 25. Display unit 33 displays images, for example, of input items related to operation data input by the operator from scanning control device 32. Display unit 33 also displays two-dimensional (2D) slice images or three-dimensional (3D) images of the subject 16 generated by data processing unit 31.

[0039] During an MRI scan performed using the MRI apparatus 10, a subject can be positioned within the imaging space 18, and an acquisition protocol can be executed to acquire the subject's MR signal. The acquisition protocol may include multiple pulse sequences, wherein in each pulse sequence, one or more RF pulses applied via the RF body coil unit 15 are used to prepare contrast, and the gradient coil unit 13 is controlled to spatially encode the resulting MR signal. The spatially encoded MR signal received by the RF coil unit 14 is digitized and stored in k-space. Therefore, k-space data or k-space datasets can refer to the raw MR signal before processing into an image. In some examples, the raw MR signal may be filled with a line in k-space for each pulse sequence (also referred to as a repetition time). In other examples, the raw MR signal may be filled with a line in k-space for each echo, wherein more than one echo is generated for each pulse sequence / repetition time. k-space data may also be referred to herein as imaging data or MR data.

[0040] refer to Figure 2 This illustration shows a scan control device 202 configured to control scan parameters for an MRI scan. In some embodiments, the scan control device 202 is integrated into the MRI apparatus 10. For example, the scan control device 202 may be provided as a scan control device 32 within the MRI apparatus 10. In some embodiments, at least a portion of the scan control device 202 is located at a device (e.g., an edge device, server, etc.) communicatively coupled to the MRI apparatus 10 via a wired and / or wireless connection. In some embodiments, at least a portion of the scan control device 202 is located at a separate device (e.g., a workstation), for example, capable of communicating with the controller unit of the MRI apparatus. The scan control device 202 may be operatively / communically coupled to a user input device 232 and a display device 234. In some examples, the user input device 232 may be a user input device of the scan control device 32, as explained above. Similarly, the display device 234 may also be the display unit 33 of the MRI apparatus 10.

[0041] The scan control device 202 includes one or more processors, such as processor 204, configured to execute machine-readable instructions stored in non-transitory memory 206. Processor 204 may be single-core or multi-core, and the program executing on this processor may be configured for parallel or distributed processing. In some embodiments, processor 204 may optionally include individual components distributed across two or more devices that may be remotely located and / or configured for coordinated processing. In some embodiments, one or more aspects of processor 204 may be virtualized and executed by remotely accessible networked computing devices configured in a cloud computing configuration.

[0042] The non-transitory memory 206 can store the reconstruction module 208, the scan control module 210, and the scan interface 212. The reconstruction module 208 can be configured to reconstruct an image from k-space data. In some examples, the reconstruction module 208 may be... Figure 1 The data processing unit 31 integrates the data processing unit and the scanning control device into a single device. However, in other examples, the reconstruction module 208 may be omitted, and the scanning control device 202 may communicate with the data processing unit 31 to obtain an image for display.

[0043] The scan control module 210 can be configured to send commands to the MRI apparatus (e.g., to the controller unit 25) to control aspects of a scan performed by the MRI apparatus. In some examples, the scan control module 210 can control aspects of the scan based on user input that can be received via the scan interface 212. For example, the scan control module 210 can process a series of TI localization images to predict the inversion time or inversion time range. Therefore, the scan control module 210 can store a region of interest (ROI) detector 211 and a mask predictor 213, which can be deployed to predict the inversion time, as will be explained in more detail below. The ROI detector 211 can be a deep learning-based model, such as a neural network (e.g., an encoder-decoder network). The mask predictor 213 can also be a deep learning-based model, such as a shallow encoder-decoder network. The scan interface 212 can display the predicted inversion time or the predicted inversion time range. The scan interface 212 may include a scan prescription display panel via which the user can set parameters for the scan (e.g., the selected inversion time for an MDE scan).

[0044] In some embodiments, the nontransitory memory 206 may include components located at two or more devices that are remotely located and / or configured for coordinated processing. In some embodiments, one or more aspects of the nontransitory memory 206 may include remotely accessible networked storage devices configured in a cloud computing configuration.

[0045] User input device 232 may include one or more of a touchscreen, keyboard, mouse, touchpad, motion-sensing camera, or other devices configured to enable a user to interact with and manipulate data within scan control device 202. In one example, user input device 232 may enable a user to select a scan protocol, adjust scan prescription settings, select or adjust angiography tracking areas, and initiate, pause, and adjust scans.

[0046] Display device 234 may include one or more display devices utilizing virtually any type of technology. In some embodiments, display device 234 may include a computer monitor and may display MR images, including images reconstructed by reconstruction module 208. Display device 234 may be integrated into the same package as processor 204, nontransitory memory 206, and / or user input device 232, or may be a peripheral display device and may include a monitor, touchscreen, projector, or other display device known in the art that enables a user to view MRI images generated by the MRI system and / or interact with various data stored in nontransitory memory 206.

[0047] It should be understood that Figure 2 The scan control device 202 shown is for illustration and not for limitation. Another suitable image processing system may include more, fewer, or different components.

[0048] Figure 3 This is a flowchart illustrating an advanced method 300 for contrast scanning according to one embodiment of the present disclosure. Method 300 may be combined with... Figure 1 The MRI device 10 utilizes Figure 2 The scanning control device 202 is used to implement this. Method 300 can be based on data stored in non-transitory memory and controlled by a processor (such as...). Figure 2 The instructions executed by the non-transitory memory 206 and processor 204 of the scanning control device 202 are implemented.

[0049] At point 302, an angiography scan is initiated upon command. For example, an angiography scan may be initiated in response to a user selection of a "Scan" button displayed on a display device. Once the operator of the MRI apparatus has been authenticated, a scan protocol interface is displayed, where the operator can instruct on which an angiography scan to be performed on the imaging subject. In some examples, via the scan protocol interface, the operator can select a specific angiography-enhanced scan protocol, such as delayed myocardial enhancement (MDE) scan, enter information about the imaging subject, etc. When the operator is ready to initiate the scan (e.g., the imaging subject is in the aperture of the MRI apparatus), the operator can select a scan button, which triggers the initiation of the scan and activates the scan interface.

[0050] At 304, TI-localization acquisition is performed, and a series of TI-localization images are reconstructed based on MR data acquired using the MRI apparatus during TI-localization acquisition. This series of TI-localization images can be reconstructed via reconstruction module 208. The series of TI-localization images from the TI-localization acquisition can be viewed via scan interface 212. TI-localization acquisition can be performed after a predetermined delay following the administration of a contrast agent to the imaging subject (e.g., after 3 minutes, after 8 minutes, or after 10 minutes). In at least some examples, the contrast agent can be a gadolinium-based contrast agent. TI-localization acquisition can be performed according to a suitable protocol / prescription (e.g., Look-Locker, Modified Look-Locker Inversion Recovery (MOLLI), or another sequence), where the TI changes for each image acquisition (e.g., gradually increasing in length). TI-localization acquisition can be gated (e.g., via an electrocardiogram sensor) so that each acquisition is performed within the same cardiac circulation. A series of TI localization images may include a suitable number of images, such as in the range of 5 to 100 images, and may span a suitable inversion time range, such as from 10 ms to 1000 ms. TI localization images may be in a suitable view, such as a short-axis view of the mid-ventricle, a long-axis view of the four chambers, or another suitable view.

[0051] At 306, method 300 includes performing automatic TI selection using a series of TI positioning images. The following is about... Figure 4 The automatic TI selection is described in more detail. In short, automatic TI selection involves locating the heart region in each image of a series of TI localization images using a ROI detector; cropping each TI localization image in the series based on the located heart region (removing image data outside the located / detected heart region); calculating the probability density function (PDF) for each cropped image; calculating one or more similarity metric maps based on the PDF of the cropped images; and predicting the optimal TI based on the one or more similarity metric maps, which can be fed as input to a mask predictor trained to identify the optimal TI from the one or more similarity metric maps. For some TI localization images, the mask predictor may not be able to identify a single optimal TI. Therefore, in some examples, instead of predicting a single TI point, the possible range of TIs is predicted by processing the similarity metric maps.

[0052] Therefore, at 308, method 300 includes optionally displaying a predicted TI point (e.g., as predicted by a mask predictor) or a TI range (e.g., when the mask predictor cannot identify a single TI point). The predicted TI point or TI range can be displayed via scanning interface 212. For example, a series of TI positioning images can be displayed, and the predicted TI point can be displayed by highlighting a TI positioning image acquired using the predicted TI point, or the TI range can be displayed by highlighting a TI positioning image acquired using TIs within the TI range. In other examples, the predicted TI point or TI range can be displayed as a value (e.g., 350 ms or 200 ms to 350 ms). At 310, method 300 includes optionally receiving a user selection of the TI point. In some examples, the user can confirm the predicted TI point from an automatic TI selector / mask predictor, or the user can select a TI point from a TI range. In some examples, users can input TI values ​​into a scan prescription interface displayed on the scan interface, or select TI values ​​from a menu on the scan prescription interface, which includes various user interface elements (e.g., menus, buttons, text boxes) through which users can set prescriptions for subsequent diagnostic scans. In further examples, when a predicted TI point is identified, the predicted TI point can be automatically input into the scan prescription interface and / or the predicted TI point can be included in the system PSD used for MDE scans.

[0053] At 312, method 300 includes performing an MDE scan using a predicted or selected TI point (e.g., a TI point selected by the user or a predicted TI point when automatically entered into the PSD for scanning). The MDE scan may include controlling the MRI apparatus to perform MDE acquisition using an inversion recovery (IR) gradient echo (GE) sequence, where the selected TI point is the delay time between the IR pulse and the data acquisition (e.g., the center of the data acquisition window). The MDE scan may include acquiring multiple short-axis and / or long-axis views of the heart, where each slice / view obtained using the IR-GE sequence has the selected TI point. The MDE scan may be performed immediately after TI-localization acquisition (allowing for user selection of the time and TI point for reconstructing the TI-localization image) or within a predetermined time (e.g., 5 minutes) from the TI-localization acquisition to ensure adequate angiography during the MDE scan. At 314, the MDE image reconstructed from the MDE acquisition may be displayed via a scanning interface. In some examples, the MDE image may be further processed and / or stored in long-term memory (e.g., in an image archiving and communication system operatively coupled to the scan control apparatus). Then, method 300 ends.

[0054] Figure 4 This is a method for predicting reversal time from a TI positioning image according to an embodiment of the present disclosure. Method 400 may be combined with Figure 1 The MRI device 10 utilizes Figure 2 The scanning control device 202 is used to implement this. Method 400 can be based on data stored in non-transitory memory and controlled by a processor (such as...). Figure 2 The instructions executed by the non-transitory memory 206 and processor 204 of the scan control device 202 are implemented. In some examples, method 400 may be executed as part of method 300, such as at 306 of method 300.

[0055] At 402, method 400 includes inputting each of a series of TI localization images as input to a ROI detector (e.g., ROI detector 211). This can be as described above regarding... Figure 3 The interpretation involves acquiring a series of time-inversion (TI) localization images, and therefore each TI localization image can be acquired at a different time-inversion (TI) but can depict the same anatomical structures (e.g., the heart) in the same view (e.g., a short-axis view) and at the same phase of the cardiac cycle. The ROI detector can be a neural network trained to identify regions of interest in the heart within each TI localization image. For example, the TI localization image may include the heart but may also include adjacent anatomical structures. The ROI detector identifies cardiac ROIs in each TI localization image and outputs a cropped frame for each TI localization image that includes only the cardiac ROI, with image data outside the cardiac ROI removed. The cardiac ROI may include all aspects of the heart visible in the TI localization image, or may include only certain features of the heart. At 404, method 400 includes receiving the cropped frame as output from the ROI detector. In this way, the ROI detector can receive TI localization images as input and can generate cropped frames corresponding to the TI localization images as output, which include the cardiac ROI from the TI localization images and exclude image data from the TI localization images outside the cardiac ROI. Each TI positioning image in a series of TI positioning images can be input into the ROI detector, resulting in the generation of a cropped frame for each TI positioning image in the series. Each cropped frame can have the same size and shape.

[0056] It should be understood that the ROI detector is not trained to segment or otherwise identify individual cardiac features, such as the septum, endocardium, ventricular cavity, etc. Instead, the ROI detector can be trained using training data pairs, where each training data pair includes training TI localization images and a baseline ground truth crop frame containing only cardiac ROIs from the training TI localization images. Each training TI localization image may have the same view, such that the ROI detector can be trained to identify cardiac ROIs in only one view. In such examples, when the trained ROI detector is used during inference (e.g., as part of method 400), the ROI detector can be selected from multiple ROI detectors based on the view of the TI localization image. In other examples, the ROI detector can be trained using training TI localization images in different views, such that one ROI detector can identify cardiac ROIs in any of multiple different views of the TI localization image.

[0057] At position 406, a probability density function (PDF) is generated for each cropped frame to form multiple PDFs, each PDF corresponding to the corresponding cropped frame (and therefore the corresponding TI positioning image). The PDF can be computed using suitable methods, such as histogram-based methods. For example, the PDF for a given cropped frame could be a normalized histogram comprising multiple bins, where each bin represents a pixel intensity value, and the height of each bin is the probability that the pixel intensity value occurs within the cropped frame.

[0058] At point 408, one or more similarity metric maps are computed based on multiple PDFs. Each similarity metric map visualizes the level of similarity between every two PDFs across the multiple PDFs, where the selected similarity metric is used to compute the similarity level. For example, a first similarity metric (such as Hellinger distance) can be used to compute a first similarity metric map, and a second similarity metric (such as Jensen-Shannon divergence) can be used to compute a second similarity metric map. Hellinger distance emphasizes the distributional overlap between two PDFs, while Jensen-Shannon divergence captures information about the divergence between two PDFs, but both similarity metrics provide a comprehensive view of the signal variations between cropped frames. The Hellinger distance (d) between the two PDFs P and Q (of the two cropped frames) is... H It can be calculated according to the following formula 1:

[0059]

[0060] Jensen-Shannon divergence (d) of two PDFP and Q values ​​(from two cropped frames) JS It can be calculated according to the following formula 2:

[0061]

[0062] To generate a similarity metric map using Hellinger distance, the Hellinger distance between every two PDFs from multiple PDFs is calculated. Each Hellinger distance can be visualized in the similarity metric map as a color or grayscale value based on that Hellinger distance. For example, a lower Hellinger distance (which indicates that two PDFs, and therefore two cropped frames, are similar) can be visualized with a darker color or darker grayscale value (e.g., black for the lowest Hellinger distance, followed by dark blue and dark purple or darker gray), and a higher Hellinger distance can be visualized with a lighter color or lighter grayscale value (e.g., purple to light blue to light green to yellow, where yellow is used for the highest Hellinger distance, or medium to light gray, where white or almost white is used for the highest Hellinger distance). The similarity metric map can be generated from points (e.g., pixel blocks) for each combination of PDFs, each point having a color or grayscale value assigned based on the Hellinger distance of that combination. A similarity metric map using Jensen-Shannon divergence can be generated in a similar manner. An example of a similarity metric graph using Hellinger distance and Jensen-Shannon divergence is shown in Figure 6 It is shown in the figure and described in more detail below.

[0063] At position 410, the similarity metric map is fed as input to the mask predictor (e.g., mask predictor 213). The mask predictor is trained to output a predicted mask indicating the location of the predicted TI point based on the input similarity metric map. The following section discusses... Figure 11Additional details regarding the training of the mask predictor are provided. In some examples, only one similarity metric map may be input as input to the mask predictor (e.g., generated using a first similarity metric or a second similarity metric). In other examples, two similarity metric maps may be input as input to the mask predictor (e.g., one generated using the first similarity metric and the other using the second similarity metric), and the two similarity metric maps may be input as input on different channels. Therefore, at 412, method 400 includes receiving a predicted mask as output from the mask predictor. The predicted mask may have the same dimensions / scale as the similarity metric map and may include points with different color or grayscale values ​​than the remaining points of the mask to indicate that the point is a predicted TI point. However, in some examples, the predicted mask may include artifacts or background points, and therefore, in order to be able to identify TI points from the predicted mask, the predicted mask may undergo post-processing, as indicated at 414. Post-processing may include adaptive thresholding to remove background noise, maximum connectivity component identification, determination of the maximum spot / point, and centroid placement based on the maximum connectivity component. The centroid can be a TI point. For example, since the prediction mask and the similarity metric map have the same resolution, the centroid can identify regions of the similarity metric map where transitions / signal changes occur, which can be mapped back to the cropped frame / TI localization image.

[0064] At 416, method 400 determines whether a single TI point can be predicted from the prediction mask. If multiple centroids are placed after post-processing or if centroids cannot be placed, a single TI point may not be predictable from the prediction mask. Furthermore, if the mask predictor cannot output a prediction mask, a single TI point may not be predictable from the prediction mask. If a single TI can be predicted from the prediction mask, method 400 proceeds to 418 to predict the TI point based on the prediction mask. Then, as described above regarding... Figure 3 The predicted TI points can then be displayed to the user, automatically entered into the scan prescription interface, and / or automatically included in the system PSD used for MDE scanning. Method 400 can then conclude.

[0065] However, if a single TI point cannot be predicted from the prediction mask, method 400 proceeds to 420 to process the similarity metric map, and as shown at 422, the TI range is identified from the processed similarity metric map. This processing can be performed by applying an Otsu threshold to the similarity metric map and identifying symmetrical bands about the diagonal. The width of the band provides the region of interest, and therefore the TI range. The TI range can then be displayed to the user, allowing the user to select TI points from the TI localization image corresponding to the TI range, as described above. Figure 3 As explained. Method 400 can be completed.

[0066] Figure 5A process 500 for generating cropped image frames from a series of TI positioning images using a ROI detector (e.g., ROI detector 211) is illustrated schematically. Process 500 schematically depicts aspects of method 400, specifically, inputting the TI positioning images to the ROI detector and outputting cropped frames from the ROI predictor. Figure 5 As shown, a series of TI positioning images 502 of the imaging subject are obtained. During the same cardiac cycle phase, each TI positioning image in the series 502 images images the same anatomical region in the same view / axis. The only difference between the TI positioning images in the series is the TI used in the pulse sequence during image acquisition. The TI (Time Inversion) can be gradually increased during the acquisition of the TI positioning images. The TI positioning images are numbered, with the first TI positioning image (image 1) having the shortest TI and the last TI positioning image (image 26 in the illustrated example) having the longest TI. For example, the TI of each TI positioning image is known and stored in memory along with the corresponding TI positioning image. Therefore, the series TI positioning images 502 are temporally ordered based on the TI from shortest to longest. However, it should be understood that other ways of ordering the TI positioning images in the series are possible (such as a reverse order from longest to shortest TI), as long as the ordering is consistent.

[0067] Each TI localization image is fed as input into ROI detector 504 (which is a non-limiting example of ROI detector 211). (See above regarding...) Figures 2 to 4 As explained, the ROI detector 504 can be a neural network trained to identify cardiac ROIs in each input TI positioning image and output cropped frames of each input TI positioning image. For example, the encoder of the ROI detector can extract features from the cropped frames into a feature space, and the decoder can process the extracted features to generate a segmentation mask for the cardiac ROI. In at least some examples, the cardiac ROI may include the myocardium and blood pool. Thus, after all images of a series of TI positioning images 502 have been input into the ROI detector 504, a set of cropped frames 506 is obtained. Each cropped frame in the set of cropped frames 506 corresponds to a TI positioning image from the series of TI positioning images 502. For example, the seventh cropped frame 508 corresponds to the seventh TI positioning image 510 in the series of TI positioning images. For example, the seventh cropped frame 508 is the cardiac ROI of the seventh TI positioning image 510. The set of cropped frames 506 is ordered similarly to the series of TI positioning images. By locating the heart ROI and cropping the image to include only the heart region, the cropped frames (which are used to calculate a similarity metric map from which the best TI point is predicted) are standardized, resulting in consistent performance regardless of changes in field of view, patient size, or view.

[0068] Figure 6 A process 600 for predicting the inversion time from cropped image frames using a similarity metric and a mask predictor is illustrated schematically. Process 600 schematically depicts aspects of method 400, specifically generating a similarity metric map from a set of cropped frames, and generating TI point predictions from a prediction mask output by the mask predictor. Figure 6 As shown above, based on the information regarding... Figure 4 The described method processes a set of cropped frames 506 to generate two similarity metric maps. For example, each cropped frame is processed to generate a PDF of that cropped frame, forming multiple PDFs, each PDF indicating the distribution of pixel intensity values ​​in the corresponding cropped frame. Jensen-Shannon divergence is calculated between every two PDFs in the multiple PDFs. For example, a first Jensen-Shannon divergence is calculated for a first PDF of a first cropped frame (relative to the first PDF), a second Jensen-Shannon divergence is calculated for the first PDF and a second PDF of a second cropped frame, a third Jensen-Shannon divergence is calculated for the first PDF and a third PDF of a third cropped frame, and so on. Each Jensen-Shannon divergence is plotted as a point (e.g., a pixel) on a first similarity metric map 602, where the value of each divergence is represented by a grayscale value. Each similarity metric map can be an image. Since each PDF is compared to itself, the first similarity metric map 602 includes a line of low divergence values ​​extending diagonally downwards along the map. Furthermore, although a set of cropped frames 506 comprises 26 frames, more or fewer cropped frames can be used to generate the similarity metric map. To handle different numbers of PDF / cropped frames, the similarity metric map can be converted (e.g., via interpolation) to a normalized size (e.g., 30x30 points / pixel or 64x64 points / pixel). Figure 6 It also includes a second similarity metric map 604, which uses Hellinger distance instead of Jensen-Shannon divergence and is generated similarly to the first similarity metric map 602. Furthermore, although... Figure 6 Each similarity metric is depicted in grayscale, but it should be understood that each similarity metric can be visualized in color, as shown above regarding... Figure 4 The explanation given.

[0069] One or both of the first similarity metric map 602 and the second similarity metric map 604 are input to the TI mask predictor 606. The mask predictor 606 may be a non-limiting example of the mask predictor 213 and may include an encoder-decoder neural network configured to receive and process at least one similarity metric map. The network may selectively process a single similarity metric map or multiple maps provided on different input channels, thereby allowing flexible handling of the first similarity metric map 602 and the second similarity metric map 604 as input. In some embodiments, the input similarity metric map may be normalized to a predetermined size (e.g., 64x64, 32x32, or 128x128 points) by interpolation before processing.

[0070] The encoder-decoder neural network of the mask predictor 606 can be implemented with various architectural configurations to balance performance and computational efficiency. In some embodiments, the mask predictor 606 includes an input layer configured to accept normalized inputs of various dimensions (e.g., 32x32, 64x64, or 128x128 points) via interpolation, and these inputs can be processed using single-channel or multi-channel configurations to handle different similarity metrics. The encoding path may include multiple convolutional layers with configurable kernel sizes and channel depths, various types of pooling layers for spatial dimensionality reduction, and different activation functions between layers. Optional normalization layers and configurable stride and padding options may also be employed. The decoding path may include upsampling layers or transposed convolutional layers to recover spatial dimensions, utilizing various methods for feature map processing and refinement. In some specific implementations, skip connections may be incorporated between the corresponding encoder and decoder layers to preserve fine spatial details. The network architecture of the mask predictor 606 can be configured with different numbers of encoding and decoding stages (e.g., two to four stages), different numbers of kernels per layer (e.g., eight kernels, sixteen kernels, or other suitable numbers), and different activation functions (such as ReLU or sigmoid), with the output layer typically employing sigmoid activation to generate pixel-wise regression values. Various optimization strategies can be implemented during training, such as different loss functions (e.g., dice loss, cross-entropy loss, or combinations thereof), attention mechanisms focusing on relevant image regions, or ensemble methods combining predictions from multiple models. A specific architecture can be selected based on factors such as available computational resources, required inference speed, and desired prediction accuracy, while maintaining the network's ability to efficiently process similarity metric maps and generate accurate predictive masks for TI point determination.

[0071] Mask predictor 606 generates a prediction mask 608 indicating regions that may contain predicted TI points. Prediction mask 608 has the same resolution as the input similarity metric map (e.g., 30 x 30 points, where each point is a pixel). Each pixel in prediction mask 608 is assigned a regression value (or probability score) indicating the likelihood that the pixel belongs to a TI point. The regression values ​​(e.g., they may be in the range of 0 to 1) are depicted in grayscale (or color) values. For example, pixels with lower probabilities may be depicted in a first grayscale (or color) value range (e.g., darker gray), while pixels with higher probabilities may be depicted in a second grayscale (or color) value range (e.g., lighter gray). In some specific implementations, the mask generation process may employ various techniques, such as multi-scale processing handling different spatial resolutions, attention mechanisms focusing on relevant image regions, or ensemble methods combining predictions from multiple models trained with different loss functions (e.g., dice loss, cross-entropy loss, or combinations thereof).

[0072] Based on the characteristics of the prediction mask 608, such as the presence of multiple high-probability regions or unclear boundaries, the mask predictor 606 can determine whether to output a single predicted TI point or indicate that multiple TI points should be considered. This flexibility allows the system to handle situations where a single optimal TI point may not be clearly identified due to imaging artifacts or pathological changes. The prediction mask undergoes post-processing steps, including adaptive thresholding, connectivity component analysis, and centroid calculation, to determine the final predicted TI point location. Various post-processing methods can be employed in different implementations, such as: morphological operations for refining mask boundaries; weighted centroid calculation based on probability values; clustering algorithms for identifying multiple potential TI points; or dynamic thresholding techniques adapted to image characteristics. Post-processing may also include confidence scoring to assess prediction reliability, where high-confidence predictions are processed differently from low-confidence predictions. In some specific implementations, time consistency checks can be performed across multiple cardiac cycles to validate the TI point prediction.

[0073] As from Figure 6 It is understood that the prediction mask 608 may not cleanly identify an exact point in the similarity metric map corresponding to a PDF / cropped frame. Therefore, additional processing of the prediction mask 608 can be performed, such as thresholding (e.g., removing pixels with regression values ​​below a threshold (e.g., 0.5), identification of the maximum connected component, centroid placement, etc. The final processed prediction mask can be used to determine the predicted TI point. Figure 6In the example presented, prediction mask 608 identifies the seventh cropped frame 508 / seventh TI localization image 510 as a predicted TI point. Therefore, as previously mentioned, the number of frames can vary from TI localization to TI localization. To train mask predictor 606, the similarity metric map is interpolated to a fixed size, e.g., 64x64. In the interpolation space, the single-pixel mask (the ground truth mask) is also rescaled based on scaling factors derived from the original frame number / resampled image dimensions. Therefore, due to interpolation / rescaling, the mask predictor output can be a mask with more than one pixel. After post-processing, the centroid, or center pixel, is computed and further interpolated back to the original resolution.

[0074] Therefore, the system and method described in this paper provide a technical solution to specific technical problems in the field of magnetic resonance imaging. Current MRI systems face the technical challenge of effectively and accurately determining the optimal reversal time across varying protocols, anatomical structures, and image artifacts. The above text discusses... Figures 1 to 6 The disclosed solution described improves the functionality of MRI systems by implementing a novel two-stage deep learning approach that reduces computational overhead while improving robustness and accuracy.

[0075] Specifically, the ROI detector neural network is trained to efficiently process input images regardless of protocol variations, patient anatomy differences, or image artifacts, thus eliminating the need for protocol-specific training data or multiple processing passes. Compared to conventional deep learning methods, the shallow encoder-decoder architecture of the mask predictor network offers technical improvements by reducing computational resource requirements while maintaining accuracy. This architecture allows the system to process TI localization images and generate predictions in near real-time during clinical workflows.

[0076] This technological solution incorporates specific improvements to neural network training and deployment, addressing unique challenges in medical imaging applications. Instead of attempting to directly identify spatial features from raw localization images via ROI detection and PDF generation, the system first standardizes the input, requiring extensive training data covering all possible anatomical variations. This approach makes the system robust to variations in imaging protocols, patient anatomy, and image artifacts while reducing computational overhead. The result is a more efficient and reliable MRI system that automatically determines optimal reversal times across a wide range of clinical scenarios without requiring protocol-specific training or multiple processing passes.

[0077] Figure 7 An exemplary similarity metric graph that can be used to predict time range reversal is shown, such as... Figure 8 shown. Specifically, Figure 8 A series of TI positioning images 800 are shown, as described above regarding Figures 4 to 6 The process is performed to generate Figure 7 The similarity metric map 700 is shown. However, a single TI point cannot be identified from the similarity metric map 700. Therefore, the similarity metric map 700 is processed to identify the TI range indicated by the dashed box on the similarity metric map 700. For example, Otsu thresholding can be performed, which can identify one or more threshold pixel intensities that separate pixels in the similarity metric map 700 into two categories. Otsu thresholding can lead to the identification of symmetrical bands, such as... Figure 7 The dashed box is shown in the figure. The width of the dashed box indicates the PDF / cropped frames identified within the TI range corresponding to the second to eighth TI positioning images of a series of TI positioning images 800, which is indicated by a white box overlaid on the series of TI positioning images.

[0078] In some examples, after obtaining a first series of TI localization images and predicting a first reversal time as described above, the system can obtain a second series of TI localization images (e.g., from different patients), where each TI localization image in the second series was obtained at a different reversal time. The second series of TI localization images may be obtained from the same subject or from different subjects at different time points. Similar to the processing of the first series, at least a portion of each TI localization image in the second series is processed to generate a second similarity metric map, which depicts the degree of statistical similarity between each pair of TI localization images in the second series within the ROI. The second similarity metric map can be generated using the same method described above, by identifying and cropping cardiac ROIs using an ROI detector, generating PDFs of the cropped frames, and calculating the similarity metric between the PDFs to form a graph.

[0079] However, in some cases, when the second similarity metric map is input to the mask predictor, the mask predictor may be unable to predict a single reversal time point from the second similarity metric map. This may occur, for example, if the second series of TI positioning images exhibits significant motion artifacts or other image quality issues that make identifying the precise zero-reversal point challenging. In such cases where the second predicted reversal time cannot be determined from the second similarity metric map via the mask predictor, the system processes the second similarity metric map to identify a range of predicted reversal time points rather than a single point. This processing may include thresholding and analyzing transitions in the similarity metric to identify time windows containing possible zero-reversal points. The identified range of predicted reversal time points is then displayed on a display device to guide manual refinement of the TI selection. This provides a fallback mechanism when precise automatic prediction is impossible, while still utilizing the information content in the similarity metric map to assist the operator.

[0080] Figure 9The first exemplary series of TI positioning images 900 exhibiting motion artifacts is shown. However, even with motion artifacts, the above regarding... Figures 4 to 6 The described process can also predict TI points, which in this paper is the seventh TI positioning image (shown by the box highlighting the seventh TI positioning image). Figure 10 A second exemplary series of TI-localization images 1000 is shown, which visualize myocardial lesions. However, even with lesions (which are deviations from normal anatomy), the above regarding... Figures 4 to 6 The described process can also predict TI points, which in this paper is the eighth TI localization image (shown by the box highlighting the eighth TI localization image). Therefore, the method disclosed in this paper is effective for challenging situations and does not require representing various anomalies in the training data (e.g., motion artifacts, lesions, etc.) as in other methods.

[0081] Figure 11 A method 1100 for training a mask predictor to predict inversion time is illustrated. Method 1100 may be implemented according to instructions stored in non-transitory memory and executed by a processor. In some examples, method 1100 may be implemented using instructions stored in non-transitory memory 206, which are... Figure 2 The scanning control device 202 is executed by the processor 204. However, in other examples, method 1100 can be implemented on a remote / external computing device (such as a server, cloud, etc.). Method 1100 can be implemented to train the above-mentioned... Figures 2 to 6 The mask predictor is described.

[0082] At 1102, method 1100 includes obtaining training data pairs. Each training data pair may include a similarity metric map and a benchmark ground truth prediction mask. Each similarity metric map may be as described above regarding... Figure 4 The generated data, and therefore can represent the relative similarity between any two PDFs in a plurality of corresponding PDFs, each of which is generated from a corresponding set of cropped frames (which in turn are generated from a corresponding series of training TI localization images). The training TI localization images can be in one or more views and can exhibit normal and / or abnormal anatomical structures, metal artifacts, and / or motion artifacts. Furthermore, each series of training TI localization images can have the same or different number of images. However, the mask predictor does not require any specific number of images in each series. Instead, as described above, the similarity metric map can be interpolated to transform the graph to a standard size (e.g., 30x30) for training. Additionally, the mask predictor does not require any specific coverage of anatomical views, lesions, or artifacts. Therefore, the training TI localization images used to train the mask predictor are relatively readily available and their total number can be smaller compared to other deep learning models used to automatically determine the optimal TI points.

[0083] Each similarity metric map can utilize the same similarity metric (e.g., Hellinger distance or Jensen-Shannon divergence). However, in some examples, a mask predictor can be trained to use two similarity metrics, and thus the training data pair can include two similarity metric maps (one using a first similarity metric and one using a second similarity metric) and a baseline ground truth prediction mask. Each baseline ground truth prediction mask can be a binary mask with the same resolution as the similarity metric map, where the pixel of the best TI point is visualized with a first grayscale or color value, and the remaining pixels are visualized with a second, different grayscale or color value. For example, Figure 12 The process 1200 for training a mask predictor using a first training data pair 1201 is illustrated schematically. The first training data pair 1201 may include a similarity metric map 1202 generated from a set of training TI localization images and a baseline ground truth prediction mask 1204. The baseline ground truth prediction mask 1204 may include pixels colored differently from the remaining pixels of the baseline ground truth prediction mask 1204 to indicate the optimal TI point. The baseline ground truth prediction mask 1204 is rescaled based on a scaling factor derived from the original frame count of the set of training TI localization images, such that the baseline ground truth prediction mask 1204 has the same resolution as the similarity metric map 1202. It should be understood that the optimal TI point is indicated when comparing the PDF of the frame corresponding to the optimal TI point (e.g., the seventh cropped frame) to itself.

[0084] In some examples, each baseline ground truth prediction mask can be generated based on user input. Specifically, one or more experts can review each series of training TI localization images to manually identify the optimal TI point, and a baseline ground truth prediction mask reflecting the optimal TI point can be generated. In other examples, one or more baseline ground truth prediction masks can be generated automatically. For example, one or more training TI localization images from a series of training TI localization images can be acquired as part of an MDE imaging examination, and the baseline ground truth / optimal TI point can be set as the TI for, for example, subsequent MDE scans.

[0085] At 1104, method 1100 includes: for a first training data pair, inputting a similarity metric map of the first training data pair as input to an untrained mask predictor. The untrained mask predictor may be a UNet (e.g., two encoding and decoding layers, eight kernels, and one sigmoid output layer). If the mask predictor is to be trained based on two similarity metric maps, each of the two similarity metric maps of the first training data pair is input as input (e.g., on different channels). At 1106, receiving the predicted mask as output from the untrained mask predictor. At 1108, calculating the loss between the predicted mask output by the mask predictor and the baseline ground truth predicted mask of the first training data pair. The loss may be a dice loss or another suitable loss. At 1110, updating the mask predictor based on the loss. For example, the weights and / or biases of the mask predictor may be updated based on the loss. See again. Figure 12 The similarity metric map 1202 is input to the untrained mask predictor 1206. The mask predictor 1206 outputs a predicted mask 1208. The predicted mask 1208 is compared with the ground truth predicted mask 1204, and the loss between the predicted mask 1208 and the ground truth predicted mask 1204 is calculated. The loss is used to update the mask predictor 1206.

[0086] At 1112, this process is repeated with the remaining training data pairs. For example, the training data may include 500 training data pairs, and the following process may be performed for each training data pair: inputting a similarity metric map (for a given training data pair) as input to the mask predictor, receiving a predicted mask from the mask predictor, calculating the loss between the predicted mask and the baseline ground truth predicted mask (for a given training data pair), and updating the mask predictor based on the loss to minimize the loss. In some examples, the same process described above for generating similarity metric maps during inference is used to generate each training similarity metric map from the corresponding training series of TI localization images. Different training series of TI localization images may have different numbers of TI localization images; for example, a first training series may include 15 TI localization images, while a second training series may include 26 TI localization images, reflecting variations in TI localization schemes across different imaging facilities. To accommodate these differences, the training similarity metric maps generated from training series with different numbers of TI localization images are interpolated to a standard size. For example, a first training similarity metric map generated from a first training series (with 15 TI location images) and a second training similarity metric map generated from a second training series (with 26 TI location images) can each be interpolated to a standard 30x30 dot size. This normalization ensures that all training similarity metric maps input to the mask predictor during training have a consistent dimension, regardless of the number of TI location images in the original training series from which the TI location images were generated. Normalization also allows the trained mask predictor to process similarity metric maps generated from any number of TI location images during inference, thus providing flexibility in handling varying TI reconnaissance schemes.

[0087] Once this process has been performed on each training data pair, or until the loss is sufficiently minimized, the mask predictor can be considered trained. In some examples, the mask predictor can be trained for 50 to 300 epochs using the Adam optimizer (in some examples, the learning rate = 1e-4) and dice loss. Method 1100 ends.

[0088] Therefore, the implementation disclosed herein allows for the automatic analysis of a series of TI localization images to predict the optimal TI point using a statistically based approach that minimizes the intensive processing of the images themselves, where selected tissues (such as myocardium) are zeroed out. The method involves generating cropped frames from a series of TI localization images using a deep learning model to focus the analysis solely on the region of interest (e.g., cardiac ROI where myocardium is visualized). The cropped frames are converted into multiple PDFs, which are compared via one or more similarity metrics (e.g., statistical distance) to form one or more similarity metric maps. The optimal TI point can be identified from the one or more similarity metric maps using a trained mask predictor, which can be a deep learning model trained to identify specific transition points (e.g., myocardium zeroing out) in the similarity metric maps.

[0089] This method offers several benefits. For example, locating / cropping to form frames representing only the cardiac region leads to standardization, resulting in consistent performance regardless of changes in field of view, patient size, or view. Predicting the optimal TI point in a single-shot manner, without resorting to multiple passes or tissue segmentation (as demonstrated by previous automated solutions), reduces computation time and robustness to disease and patient variations. The method described in this paper provides reliable and consistent MDE imaging regardless of variations in the TI-Scout scheme, cardiac view, or imaging artifacts. Even when the optimal TI point cannot be determined due to disease, artifacts, etc., the similarity metric map conveys information that can be used to guide the user from the similarity metric map to the approximate temporal location of the optimal TI point, allowing for manual refinement without scrolling through all TI localization images.

[0090] The proposed solution provides single-shot TI prediction capability by simplifying temporal imaging data into a 2D statistical measurement map (e.g., a similarity metric map) and using deep learning (e.g., the mask predictor described above) to derive the optimal TI point. This method requires only a coarse ROI around the myocardium labeled by a deep learning model (e.g., the ROI detector described above), without requiring fine segmentation of cardiac anatomy. It is independent of the number of frames or temporal scaling, and therefore adaptable to the subjectivity of cross-site TI localization schemes. Furthermore, the solution can detect whether myocardial zeroing has not occurred in a localization sequence (e.g., a series of TI localization images) and can provide appropriate user feedback. Additionally, the solution supports multiple views, including axial, long-axis (4 chambers), and short-axis views, as well as varying patient conditions. The method disclosed herein simplifies temporal imaging data into a 2D statistical measurement map of frame-to-frame similarity and uses this map to calculate the TI location. Therefore, it overcomes the need, as required by existing known solutions, to train deep learning models (e.g., mask predictors) for various orientations and patient conditions, and can handle variable-dimensional temporal data based on clinical protocols.

[0091] Although this method has been described in this paper as a prediction of the optimal TI point in a series of TI localization images for MDE scans, it should be understood that the method disclosed herein can be applied to other imaging contexts where frame-to-frame similarity and / or transition point determination is desired, such as dynamic perfusion imaging with injected contrast agents in MRI or computed tomography.

[0092] The system and method disclosed in this paper offer several technical improvements to automated MRI inversion time prediction. By implementing a process of acquiring a series of TI localization images and processing them to generate a similarity metric map depicting the degree of statistical similarity between image pairs within a region of interest (ROI), the system achieves more efficient and robust TI prediction while reducing computational overhead compared to conventional methods. Specifically, the ROI detector's identification and cropping of each TI localization image to remove information outside the ROI provides a technical advantage by normalizing the input data regardless of changes in field of view, patient size, or cardiac view. This normalization eliminates the need for large amounts of training data covering all possible anatomical variations and imaging conditions. Furthermore, the generation of the probability density function (PDF) for the cropped frames and the subsequent generation of the similarity metric map transform the temporal imaging data into an interpretable 2D statistical measurement space. This transformation makes the analysis more robust to imaging artifacts and noise, while highlighting key tissue zeroing points more clearly than the original image data.

[0093] Furthermore, the mask predictor trained to output predicted TI points based on a similarity metric map offers additional technical benefits by achieving single-pass TI prediction without requiring multiple sliding window operations or detailed tissue segmentation. This architectural choice significantly reduces computational complexity compared to deep learning models that must process full image spatial features. The system can handle a variable number of input frames and different cardiac views without requiring scheme-specific training or model modifications.

[0094] The method disclosed in this paper also provides improved artifact handling capabilities. By manipulating image features directly in the statistical measurement space rather than by generating and analyzing similarity metric graphs, the system maintains reliable performance even in the presence of motion artifacts, metal artifacts, and pathological changes—conditions that typically degrade the performance of conventional image-based deep learning methods. This robustness is achieved without requiring explicit training examples that cover all possible artifact conditions, thus significantly reducing training data requirements while improving generalization.

[0095] Furthermore, the system's ability to output predicted reversal time or predicted reversal time range based on similarity metric graphs represents a technical improvement in reliability and clinical workflow. The visualization of the similarity metric provides interpretable feedback on the prediction process, enabling informed manual refinement when needed—a capability typically lacking in black-box deep learning solutions.

[0096] These technological improvements have resulted in more efficient, robust, and generalizable solutions for automated TI prediction in MRI scans, while reducing computational resources, lowering training data requirements, and decreasing sensitivity to imaging artifacts and protocol variations. The method disclosed in this paper fundamentally transforms the technical nature of automated TI prediction from an image feature analysis problem into a more tractable statistical analysis task.

[0097] The technical advantage of predicting reversal time or reversal time range based on similarity metric map output is that it transforms temporal MRI localization image data into a standardized statistical measurement space. This enables robust and computationally efficient prediction of optimal reversal time while reducing sensitivity to imaging artifacts, protocol variations, and anatomical differences. By processing TI localization images to generate a similarity metric map depicting the degree of statistical similarity between each pair of TI localization images in the region of interest, the system provides a more tractable solution. Compared to conventional image-based analysis methods, this solution reduces computational overhead and improves reliability by operating in the statistical domain rather than directly on image features, thus providing a solution with broad training data coverage that is less dependent on anatomical variations and imaging conditions.

[0098] This disclosure also provides support for a system comprising: one or more processors, and a memory storing instructions executable by the one or more processors to: acquire a series of TI localization images, each of the series of TI localization images being acquired at a different inversion time; process at least a portion of each of the series of TI localization images to generate a similarity metric map, the similarity metric map depicting the degree of statistical similarity between each pair of TI localization images in a region of interest (ROI); predict an inversion time based on the similarity metric map; and output the predicted inversion time to be included in a pulse sequence design for a diagnostic scan. In a first example of the system, the system further comprises: a magnetic resonance imaging (MRI) apparatus, and wherein the instructions are further executable to command the MRI apparatus to perform the diagnostic scan using the pulse sequence design including the predicted inversion time. In a second example of the system, optionally including a first example, processing at least a portion of each TI positioning image in the series of TI positioning images to generate the similarity metric map includes: identifying the ROI in each TI positioning image using a ROI detector configured to crop each TI positioning image to remove image information outside the ROI, thereby forming a set of cropped frames; calculating a probability density function (PDF) for each cropped frame; and generating the similarity metric map from each PDF, wherein the similarity metric map visualizes the similarity metric calculated between every two PDFs. In a third example of the system, optionally including one or both of the first and second examples, the similarity metric is Hellinger distance or Jensen-Shannon divergence. In a fourth example of the system, optionally including one or more or each of the first to third examples, predicting the reversal time based on the similarity metric map includes: inputting the similarity metric map as input to a mask predictor trained to output a predicted mask that identifies the predicted reversal time.In a fifth example of the system, optionally including one or more or each of the first to fourth examples, the series of TI positioning images is a first series of TI positioning images, the similarity metric map is a first similarity metric map, and the predicted reversal time is a first predicted reversal time, and wherein the instructions are further executable by the one or more processors to: obtain a second series of TI positioning images, each of the second series of TI positioning images being obtained at a different reversal time; process at least a portion of each of the second series of TI positioning images to generate a second similarity metric map, the second similarity metric map depicting the degree of statistical similarity between each pair of TI positioning images of the second series of TI positioning images in the ROI; and determine that a second predicted reversal time cannot be predicted from the second similarity metric map via the mask predictor, and in response, process the second similarity metric map to identify a range of predicted reversal times and display the range of predicted reversal times on a display device. In a sixth example of the system, optionally including one or more or each of the first to fifth examples, processing the second similarity metric map to identify the predicted inversion time range includes performing Otsu thresholding to identify one or more bands on the similarity metric map, the one or more bands corresponding to the predicted inversion time range.

[0099] This disclosure also provides support for a method comprising: obtaining a series of TI positioning images, each of the series of TI positioning images being obtained using a different inversion time; processing at least a portion of each of the series of TI positioning images to generate a similarity metric map, the similarity metric map depicting the degree of statistical similarity between each pair of TI positioning images in a region of interest (ROI); predicting an inversion time or a range of predicted inversion times based on the similarity metric map; and outputting the predicted inversion time or range of predicted inversion times to be included in a pulse sequence design for a diagnostic scan and / or for display on a display device. In a first example of the method, the method further comprises: commanding a magnetic resonance imaging (MRI) apparatus to perform the diagnostic scan using a pulse sequence design including the predicted inversion times. In a second example of the method, optionally including the first example, the method further comprises: commanding a magnetic resonance imaging (MRI) apparatus to perform the diagnostic scan using the pulse sequence design including a selected inversion time selected from the predicted inversion time range. In a third example of the method, optionally including one or both of the first and second examples, processing at least a portion of each TI positioning image in the series of TI positioning images to generate the similarity metric map includes: identifying the ROI in each TI positioning image using a ROI detector configured to crop each TI positioning image to remove image information outside the ROI, thereby forming a set of cropped frames; calculating the probability density function (PDF) for each cropped frame; and generating the similarity metric map from each PDF, wherein the similarity metric map visualizes the similarity metric calculated between every two PDFs. In a fourth example of the method, optionally including one or more or each of the first to third examples, the similarity metric is a Hellinger distance or a Jensen-Shannon distance. In a fifth example of the method, optionally including one or more or each of the first to fourth examples, predicting the reversal time or the predicted reversal time range based on the similarity metric map includes: inputting the similarity metric map as input to a mask predictor, the mask predictor being trained to output a predicted mask that identifies the predicted reversal time; if the predicted reversal time can be identified from the predicted mask, then outputting the predicted reversal time to be included in the pulse sequence design for the diagnostic scan; and if the predicted reversal time cannot be identified from the predicted mask, then processing the similarity metric map to identify the predicted reversal time range and outputting the predicted reversal time range for display on the display device.

[0100] This disclosure also provides support for a system comprising: one or more processors, and a memory storing instructions executable by the one or more processors to: obtain a series of TI localization images, each of which is obtained using a different inversion time; process each of the TI localization images in the series of TI localization images using a region of interest (ROI) detector to generate a set of cropped frames, each cropped frame visualizing only the ROI in the corresponding TI localization image; generate a similarity metric map depicting the degree of statistical similarity between each pair of cropped frames in the set of cropped frames; process the similarity metric map using a mask predictor trained to output predicted TI points based on the similarity metric map; and output a pulse sequence design for a diagnostic scan including the predicted TI points. In a first example of the system, the mask predictor is trained to output a predictive mask that identifies the predicted TI points, and wherein the mask predictor is trained using multiple training data pairs, each training data pair including a training similarity metric map and a baseline ground truth predicted mask. In a second example of the system, a first example is optionally included, wherein each training similarity metric map is generated from a corresponding training series of TI-localized images, wherein the first training series of TI-localized images has a different number of TI-localized images than the second training series of TI-localized images, and wherein the first training similarity metric map generated from the first training series of TI-localized images and / or the second training similarity metric map generated from the second training series of TI-localized images are interpolated to a standard size such that the first training similarity metric map and the second training similarity metric map are each at the standard size. In a third example of the system, one or both of the first and second examples are optionally included, wherein generating the similarity metric map includes computing a probability density function (PDF) for each cropped frame, and generating the similarity metric map from each PDF, wherein the similarity metric map visualizes the similarity metric computed between every two PDFs. In a fourth example of the system, one or more or each of the first to third examples are optionally included, wherein the similarity metric is Hellinger distance or Jensen-Shannon divergence. In a fifth example of the system, one or more or each of the first to fourth examples may be optionally included, wherein the ROI is a cardiac ROI, wherein the diagnostic scan is a delayed-enhanced myocardial scan, and wherein the pulse sequence design includes a time delay between the inversion recovery pulse and the data acquisition sequence, the time delay being equal to the predicted TI point.In a sixth example of the system, optionally including one or more or each of the first to fifth examples, the mask predictor includes: an encoder-decoder neural network configured to: receive at least one similarity metric map as input, process the at least one similarity metric map to generate a predictive mask indicating a region that may contain the predicted TI point, wherein the predictive mask is processed to determine the location of the predicted TI point, and wherein the system is configured to selectively output an indication of a single predicted TI point or that multiple TI points should be considered based on characteristics of the predictive mask.

[0101] As used herein, unless otherwise specified, the term “approximately” is understood to mean ±5% of the range.

[0102] As used herein, elements or steps listed in the singular and beginning with the word "a" or "an" should be understood to not exclude a plurality of said elements or steps unless such exclusion is explicitly stated. Furthermore, references to "one embodiment" of the invention are not intended to be construed as excluding the existence of additional embodiments that also include the referenced features. Moreover, unless explicitly stated to the contrary, embodiments that "comprise," "include," or "have" elements or multiple elements having a particular characteristic may include additional such elements that do not have that characteristic. The terms "comprise" and "in" are used as concise linguistic equivalents to the corresponding terms "comprising" and "wherein". Furthermore, the terms "first," "second," and "third," etc., are used merely as notations and are not intended to impose numerical requirements or a particular order of position on their objects.

[0103] This written description uses examples to disclose the invention, including the best mode, and also enables those skilled in the art to practice the invention, including making and using any device or system and performing any included methods. The scope of patentability of the invention is defined by the claims, but may include other examples that would occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that are not indistinguishable from the literal language of the claims, or if they include equivalent structural elements that have minor differences from the literal language of the claims.

[0104] The following claims specifically point to certain combinations and sub-combinations that are considered novel and non-obvious. These claims may refer to a "one" element or a "first" element or its equivalent. Such claims should be understood to include combinations of one or more such elements, neither requiring nor excluding two or more such elements. Other combinations and sub-combinations of the disclosed features, functions, elements, and / or properties may be claimed by amending these claims or by presenting new claims in this application or related applications. Such claims, whether broader or narrower in scope than the original claims, identical or different, are considered to be included within the subject matter of this disclosure.

Claims

1. A system comprising: One or more processors (204); and Memory (206), the memory storing instructions that can be executed by the one or more processors to: A series of TI positioning images is obtained (304), each of which is obtained using a different inversion time; At least a portion of each TI positioning image in the series of TI positioning images is processed (408) to generate a similarity metric map, which depicts the degree of statistical similarity between each pair of TI positioning images in the series of TI positioning images in the region of interest (ROI); The reversal time (418) is predicted based on the similarity metric graph; as well as The output (312) predicts the inversion time to be included in the pulse sequence design for diagnostic scans.

2. The system of claim 1, further comprising a magnetic resonance imaging (MRI) device (10), wherein the instructions are further executable to command (312) the MRI device to perform the diagnostic scan using the pulse sequence design including the predicted inversion time.

3. The system of claim 1, wherein processing at least a portion of each TI positioning image in the series of TI positioning images to generate the similarity metric map comprises: An ROI detector is used to identify (404) the ROI in each TI positioning image, the ROI detector being configured to crop each TI positioning image to remove image information outside the ROI, thereby forming a set of cropped frames; Calculate the probability density function (PDF) for each cropped frame (406); as well as (408) The similarity metric graph is generated from each PDF, wherein the similarity metric graph visualizes the similarity metric calculated between every two PDFs.

4. The system of claim 3, wherein the similarity metric is Hellinger distance or Jensen-Shannon divergence.

5. The system of claim 1, wherein predicting the reversal time based on the similarity metric graph comprises: The similarity metric map is input (410) to a mask predictor, which is trained to output a predicted mask that identifies the predicted reversal time.

6. The system of claim 5, wherein the series of TI positioning images is a first series of TI positioning images, the similarity metric map is a first similarity metric map, and the prediction reversal time is a first prediction reversal time, and wherein the instructions are further executable by the one or more processors to: Obtain (304) a second series of TI positioning images, each of which is obtained using a different inversion time; At least a portion of each TI positioning image in the second series of TI positioning images is processed (408) to generate a second similarity metric map, which depicts the degree of statistical similarity between each pair of TI positioning images in the second series of TI positioning images in the ROI; as well as It is determined that the second predicted reversal time cannot be predicted from the second similarity metric map via the mask predictor, and in response, the second similarity metric map is processed (420) to identify (422) the predicted reversal time range and the predicted reversal time range is displayed (308) on the display device.

7. The system of claim 6, wherein processing the second similarity metric map to identify the predicted inversion time range includes performing Otsu thresholding to identify one or more bands on the similarity metric map, the one or more bands corresponding to the predicted inversion time range.

8. A method, the method comprising: A series of TI positioning images is obtained (304), each of which is obtained using a different inversion time; At least a portion of each TI positioning image in the series of TI positioning images is processed (408) to generate a similarity metric map, which depicts the degree of statistical similarity between each pair of TI positioning images in the series of TI positioning images in the region of interest (ROI); Predict the reversal time (418, 422) or the range of reversal times based on the similarity metric graph; and The outputs (308, 312) predict the inversion time or the range of the predicted inversion time to be included in the pulse sequence design for diagnostic scanning and / or for display on a display device.

9. The method of claim 8, further comprising commanding (312) a magnetic resonance imaging (MRI) device to perform the diagnostic scan using a pulse sequence design including the predicted inversion time.

10. The method of claim 8, further comprising commanding (312) a magnetic resonance imaging (MRI) apparatus to perform the diagnostic scan using the pulse sequence design, the pulse sequence design including a selected inversion time selected from the predicted inversion time range.

11. The method of claim 8, wherein processing at least a portion of each TI positioning image in the series of TI positioning images to generate the similarity metric map comprises: An ROI detector is used to identify (404) the ROI in each TI positioning image, the ROI detector being configured to crop each TI positioning image to remove image information outside the ROI, thereby forming a set of cropped frames; Calculate the probability density function (PDF) for each cropped frame (406); as well as (408) The similarity metric graph is generated from each PDF, wherein the similarity metric graph visualizes the similarity metric calculated between every two PDFs.

12. The method of claim 11, wherein the similarity metric is Hellinger distance or Jensen-Shannon distance.

13. The method of claim 8, wherein predicting the reversal time or the predicted reversal time range based on the similarity metric graph comprises: The similarity metric map is input (410) to a mask predictor, which is trained to output a predicted mask that identifies the predicted reversal time. If the predicted reversal time can be identified from the prediction mask, then the predicted reversal time is output (418) to be included in the pulse sequence design for the diagnostic scan; and If the predicted reversal time cannot be identified from the prediction mask, the similarity metric map is processed (420) to identify (422) the predicted reversal time range and output (308) the predicted reversal time range for display on the display device.

14. A system comprising: One or more processors (204); and Memory (206), the memory storing instructions that can be executed by the one or more processors to: A series of TI positioning images is obtained (304), each of which is obtained using a different inversion time; Each TI localization image in the series of TI localization images is processed (402, 404) using a region of interest (ROI) detector to generate a set of cropped frames, each cropped frame only making the ROI in the corresponding TI localization image visible; Generate a (408) similarity metric map, which depicts the degree of statistical similarity between each pair of cropped frames in the set of cropped frames; The similarity metric map (410, 412) is processed using a mask predictor, which is trained to output predicted TI points based on the similarity metric map. as well as Outputs (418, 312) are used for designing pulse sequences for diagnostic scans that include the predicted TI points.

15. The system of claim 14, wherein the mask predictor is trained to output a predicted mask that identifies the predicted TI point, and wherein the mask predictor is trained using a plurality of training data pairs, each training data pair including a training similarity metric map and a baseline ground truth predicted mask.

16. The system of claim 15, wherein each training similarity metric map is generated from a corresponding training series of TI-localized images, wherein the first training series of TI-localized images has a different number of TI-localized images than the second training series of TI-localized images, and wherein the first training similarity metric map generated from the first training series of TI-localized images and / or the second training similarity metric map generated from the second training series of TI-localized images are interpolated to a standard size such that the first training similarity metric map and the second training similarity metric map are each at the standard size.

17. The system of claim 14, wherein generating the similarity metric graph comprises calculating (406) a probability density function (PDF) for each cropped frame; and (408) The similarity metric graph is generated from each PDF, wherein the similarity metric graph visualizes the similarity metric calculated between every two PDFs.

18. The system of claim 17, wherein the similarity metric is Hellinger distance or Jensen-Shannon divergence.

19. The system of claim 14, wherein the ROI is a cardiac ROI, wherein the diagnostic scan is a delayed-enhanced myocardial scan, and wherein the pulse sequence design includes a time delay between the inversion recovery pulse and the data acquisition sequence, the time delay being equal to the predicted TI point.

20. The system of claim 14, wherein the mask predictor comprises: An encoder-decoder neural network, wherein the encoder-decoder neural network is configured as follows: Receive at least one similarity metric graph as input; The at least one similarity metric map is processed to generate a prediction mask indicating regions that may contain the predicted TI point; The prediction mask is processed to determine the location of the predicted TI point; and The system is configured to selectively output a single predicted TI point or an indication that multiple TI points should be considered based on the characteristics of the prediction mask.

Images