Optimization method for positioning fetal brain three-axis position by nuclear magnetic resonance

Abstract

The present application relates to the optimization method of nuclear magnetic resonance positioning fetal brain three-axis position. The specific implementation steps include: initialization three-axis positioning: through the standard T2 weighted sequence, the three-dimensional image of the fetal brain is obtained initially; after detecting the initial three-axis position of the fetal brain, the optical and magnetic resonance dual-mode motion tracking system is started, and the real-time monitoring of the fetal movement is started; the motion capture algorithm is used to collect the position information of the fetal brain movement in real time, and the image processing chip FPGA is used for processing; the real-time motion information of the fetus is fed back to the MRI system, and the three-axis position of the current scanning is adjusted; based on the three-dimensional motion compensation algorithm, the motion of the fetus is calculated by interpolation, and the image displacement error is compensated in real time; the real-time scanning image is processed through the fast Fourier transform FFT algorithm, the correct position of the fetal brain is reset in the image reconstruction process, and the scanning sequence is continuously optimized, including the scanning angle, the time interval.

Description

Technical Field

[0001] This invention relates to an optimized method for locating the fetal brain in three axial positions, specifically an optimized method for locating the fetal brain in three axial positions using magnetic resonance imaging (MRI). Background Technology

[0002] The method described in Chinese invention patent 2011100787719, while solving the problem of fetal brain triaxial localization to some extent, still has many shortcomings, mainly in terms of operational complexity, low imaging efficiency, poor adaptability to fetal movement, and excessive human intervention. Firstly, the localization step in this patented method relies on traditional sequential imaging methods, such as HASTE sequences and heavily T2-weighted imaging. These methods are prone to artifacts and imaging distortion when encountering fetal movement, thus affecting the accuracy of localization. Since fetal movement in the uterus is uncontrollable and frequent, especially in the mid-to-late stages of pregnancy, traditional static image localization methods cannot track and dynamically compensate for fetal movement in real time, requiring multiple repeated scans to obtain clear images. This not only increases imaging time but also increases the time the fetus is exposed to magnetic fields, affecting imaging safety.

[0003] Secondly, this method relies on gradual adjustments and human intervention to achieve standard triaxial positioning. This manual adjustment process is not only time-consuming but also requires operators with high levels of experience and skill; otherwise, operational errors and duplicate scans are easily caused. Especially when positioning different triaxial positions, using a fixed sequence of "first positioning one standard axis, then positioning other axes" can easily lead to positioning deviations when the fetal position changes significantly, thus affecting the accuracy and consistency of subsequent imaging. Furthermore, while the "360° rotation positioning" mentioned in the method can improve the identification of the brain midline to some extent, this rotational scanning is prone to a time lag effect when the fetus is moving. That is, by the time the brain midline is actually scanned, the fetal posture may have changed, leading to inaccurate positioning. More seriously, this method lacks effective dynamic response strategies for complex fetal movements such as head rotation and up-and-down movements. Especially during spontaneous movements such as fetal breathing, swallowing, and kicking, traditional fixed scanning and sequence adjustments cannot quickly respond and adjust scanning parameters, making it difficult to guarantee the efficiency and quality of scanning imaging. Because it cannot achieve real-time dynamic tracking and adaptive adjustment, this method often requires multiple repositioning and repeated scanning in practical applications. This not only wastes a lot of time but also increases discomfort and risks for both mother and baby. In addition, this method relies on the gradual correction of static images, and it cannot monitor and adjust the alignment and consistency between the three axes in real time. This can easily cause inconsistencies in the three-axis contrast of the images, thereby affecting the doctor's accurate assessment and diagnosis of fetal brain development.

[0004] Furthermore, the manual correction and prioritization of coronal, sagittal, and axial planes mentioned in the disclosed technology affect the continuity of imaging. This method, involving significant human intervention, increases operational complexity, and the adjustment and correction process is extremely time-consuming and labor-intensive when positioning errors occur, further reducing imaging efficiency. Simultaneously, due to the lack of intelligent and automated processing methods, this method is easily influenced by the operator's subjective judgment in practical applications, leading to instability and poor repeatability in positioning. Different operators may produce different imaging results, posing a challenge to the standardization and normalization of clinical diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide an optimized method for MRI localization of the fetal brain in three axial planes, thereby addressing some of the shortcomings mentioned in the background art.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems includes the following steps:

[0007] S1. Real-time dynamic tracking and feedback:

[0008] S1.1 Initialization of triaxial localization: Initially obtain three-dimensional images of the fetal brain using standard T2-weighted sequences;

[0009] S1.2 Once the initial triaxial position of the fetal brain is detected, the optical and magnetic resonance dual-modal motion tracking system is activated to begin real-time monitoring of fetal movement;

[0010] S1.3. Real-time acquisition of fetal brain movement position information using motion capture algorithms and processing via an image processing chip FPGA; feeding back the real-time fetal movement information to the MRI system to adjust the current three-axis position of the scan.

[0011] S2. Motion Compensation and Image Reconstruction:

[0012] S2.1 Based on the three-dimensional motion compensation algorithm, the motion of the fetus is interpolated and calculated to compensate for image displacement error in real time;

[0013] S2.2. The real-time scanned image is processed by the Fast Fourier Transform (FFT) algorithm to control the image reconstruction process and reset it to the correct position of the fetal brain.

[0014] S2.3 Continuously perform adaptive optimization of the scanning sequence, including scanning angle and time interval; when a large-scale movement of the fetus, including rotation or displacement, exceeding the threshold is detected, pause the current scan and reposition the three axes;

[0015] S3, AI-driven adaptive slicing optimization:

[0016] S3.1 Using a pre-trained three-dimensional anatomical model, analyze MRI images of the fetal brain to initially identify important anatomical regions;

[0017] S3.2 Determine the priority of slice scanning based on feedback from the AI ​​model;

[0018] S3.3 During the scanning process, the AI ​​model analyzes the acquired images, optimizes the slice thickness and spacing in real time, and dynamically adjusts the slice angle to obtain a qualified imaging effect.

[0019] Furthermore, the method for resetting the image to its correct position in the fetal brain during image reconstruction includes: performing a Fast Fourier Transform (FFT) on the acquired spatial domain image I(x, y) to convert it to the frequency domain F(u, v); in the frequency domain, the motion information of the image is manifested as specific phase changes and frequency shifts, significantly revealing the details of the motion trajectory; wherein the frequency domain transformation and phase representation are:

[0020]

[0021] Where F(u, v) represents the image data in the frequency domain, which includes motion artifacts and phase information; I(x, y) represents the original image data in the spatial domain, where x and y are the horizontal and vertical coordinates of the image, respectively; u and v are the horizontal and vertical frequency coordinates in the frequency domain, representing the image frequency information; j is the imaginary unit, representing the imaginary part in complex number operations; 2π(ux+vy) is the principal phase factor in the domain, characterizing the spatial frequency features of the image; θ(x, y) is the phase shift function of the image, describing the phase change caused by fetal movement.

[0022] Furthermore, the method for resetting the correct position to the fetal brain during the control image reconstruction process includes: after acquiring the frequency domain image F(u, v), a nonlinear phase correction algorithm is used for compensation; based on the real-time motion data M(t) of the fetus and the frequency domain features, a complex compensation function is designed; the compensation function integrates frequency coordinates and time factors to perform an inverse transformation on the phase shift caused by fetal movement.

[0023] Φ(u,v,t)=a·sin(b·f(u)·cos(g(v)+h(t)))+c·e -d·t

[0024] Where Φ(u,v,t) represents the phase compensation function, used to adjust the phase error in the frequency domain; a, b, c, d represent parameters, controlling the amplitude of phase compensation, frequency response, nonlinear amplitude adjustment, and time attenuation factor, respectively; f(u) represents a custom adjustment function for the horizontal coordinate u, used to correct the horizontal component of the phase shift; g(v) represents a custom adjustment function for the vertical coordinate v, used to correct the vertical component of the phase shift; h(t) is the motion prediction function of the inter-correlation function, describing the real-time motion characteristics of the fetus; sin and cos are functions used to implement nonlinear phase correction; e -d·t This represents the attenuation term, which controls the compensation to gradually decrease over time, adapting to dynamic motion.

[0025] Furthermore, the method for resetting the image to the correct position in the fetal brain during image reconstruction includes: after compensating for the phase shift, using inverse fast Fourier transform (iFFT) to convert the adjusted frequency domain data F′(u, v) back to the spatial domain, and reconstructing the compensated image; simultaneously, during image reconstruction, resetting adjustment is performed by comparing the current frame with adjacent frames; FT reconstruction and resetting correction are employed.

[0026]

[0027] Where I′(x, y) is the reconstructed image after phase compensation, restored to the correct spatial position; F′(u, v) is the frequency domain data after phase compensation;

[0028] Where, Δφ(t)=α·(t-t0) 2 +β·sin(γ·t) represents the dynamic adjustment function for phase reset; α, β, and γ are adjustment coefficients that control the effect of time changes on reset; (t-t0) represents the difference between the current time t and the initial time t0, describing the trend of phase change over time; sin(γ·t) describes the periodic phase change, adapting to the natural movement of the fetal brain.

[0029] Furthermore, the method for obtaining satisfactory imaging results includes: introducing a real-time analysis scanning image mode, performing quality scoring on triaxial slices, and dynamically adjusting parameters during the scanning process; the triaxial image quality scoring function is as follows:

[0030]

[0031] Among them, Q xyz A comprehensive quality score for triaxial images, used to evaluate the overall imaging effect of coronal, sagittal, and transverse images; A xyzThe integral region represents the anatomical region of interest in triaxial imaging, including key structures of the cerebral cortex and corpus callosum; R(x, y, z) is the three-dimensional resolution function, representing the image resolution on each axis (x, y, z); N(x, y, z) is the noise distribution function, reflecting the noise level on each axis; C(x, y, z) represents the contrast function, measuring the contrast intensity between different anatomical structures.

[0032] The AI ​​model receives triaxial MRI image data in real time and calculates the image quality using the scoring function. The scoring result is used to determine whether the current scanning parameters meet the imaging requirements and is fed back to the scanning control system to adjust the parameters.

[0033] Furthermore, the method for obtaining satisfactory imaging results includes: after obtaining an image quality score, adjusting the slice thickness and spacing based on the analysis results of the AI ​​model; requiring fine slice thickness for key regions of the cerebral cortex during fetal brain development; and increasing the thickness in regions with less information to reduce scanning time; the slice thickness and spacing optimization function is as follows:

[0034]

[0035] Among them, T xyz (h, z) represents the dynamically optimized slice thickness in the three-axis view; h is the sharpness score of the current three-axis image, evaluated by the AI ​​model, reflecting the overall detail quality of the image; z is the importance index of anatomical structures, measuring the need for high-resolution imaging in different regions; α xyz It is a constant coefficient that controls the range of thickness adjustment in the three-axis position; γ represents the derivative of image sharpness with respect to slice thickness, used to dynamically evaluate the response of image quality to thickness adjustments; xyz To adjust parameters and control the degree to which anatomical importance affects slice thickness; D xyz (z) is the anatomical complexity function, describing the structural complexity of a specific anatomical region; log(γ) xyz ·D xyz (z) represents the process of performing a logarithmic transformation on the complexity to smooth it out.

[0036] Furthermore, the method for obtaining qualified imaging results includes: dynamically adjusting the slice angles in the three-axis view to match the natural anatomical orientation of the fetal brain; the AI ​​model calculates the required angle deviation by analyzing the current angles in the three-axis view in real time and comparing them with the ideal anatomical structure orientation; the slice angle adjustment function uses:

[0037]

[0038] Where, Θ xyz(t) represents the correction value of the slice angle in the three-axis view at the current time t, used to dynamically adjust the scanning direction to match the natural position of the fetal brain; E xyz (x, y, z, τ) represents the error function between the current triaxial slice and the ideal anatomical orientation; The time rate of change of error reflects how quickly the error changes over time and is used to predict future error trends; A xyz (τ) is the angle change function, reflecting the dynamic change of the current slice angle over time; ρ xyz It is an adjustment coefficient that controls the degree to which the error affects the angle correction; κ xyz ω is the stability coefficient, controlling the smoothness of the angle adjustment process; xyz It adjusts the frequency coefficient to control the frequency of angle correction; sin(ω) xyz ·A xyz (τ) is the periodic angle adjustment term; t0 and t represent the time range of the adjustment process, and the integral is used to calculate the cumulative effect of the angle correction.

[0039] The optimized method for MRI localization of the fetal brain in three-axis views, as described in this invention, significantly improves the quality and accuracy of fetal brain MRI imaging by combining deep learning, real-time tracking, dynamic adjustment, and three-axis slice optimization techniques. It also solves the image artifacts and distortion problems caused by fetal movement in traditional imaging processes. The specific beneficial effects of this invention are as follows:

[0040] AI models and deep learning networks are introduced to track and dynamically adjust the fetal brain in real time. This allows for rapid identification of changes in the fetal brain's position and automatic adjustment of scanning parameters, such as slice thickness, spacing, and angle. Through a slice angle adjustment function, the system can adapt to the fetal movement, ensuring that each frame of the image closely matches the fetus's natural anatomical position, effectively reducing motion artifacts and improving image alignment and clarity.

[0041] By evaluating image quality through a real-time scoring function, this invention dynamically adjusts scanning parameters based on the scoring results. For example, an optimization function for slice thickness ensures high-resolution images in key anatomical regions (such as the cerebral cortex), while appropriately increasing thickness in areas with less information to save scanning time. This method can guarantee the continuity and high quality of triaxial imaging even with frequent fetal movements, providing doctors with clearer and more coherent imaging data.

[0042] By introducing a nonlinear phase correction algorithm and inverse fast Fourier transform (iFFT), the phase shift caused by fetal movement is effectively compensated, ensuring that the reconstructed image can be accurately repositioned to the correct location of the fetal brain. This motion compensation and image reconstruction method significantly reduces image artifacts and distortion, enabling the acquisition of high-quality MRI images even when the fetus is moving or rotating significantly, thus providing more reliable data support for the early diagnosis of fetal brain developmental abnormalities. Attached Figure Description

[0043] Figure 1 This is a flowchart of the optimized method for MRI localization of the fetal brain in three axial positions according to the present invention.

[0044] Figure 2 This is a flowchart of the method for controlling the repositioning of the image to the correct position in the fetal brain during the image reconstruction process of the present invention.

[0045] Figure 3 The flowchart shows the method used to obtain satisfactory imaging results in this invention.

[0046] Figure 4 Images of five fetal subjects (gestational ages 25 / 18.6 / 15.7 / 26 / 21) obtained at five echo times (10 / 52 / 98 / 142.5 / 240 ms) according to the present invention, including a network-predicted brain mask (red) and a corresponding hand-drawn mask (yellow).

[0047] Figure 5 The images of fetus 5 of the present invention at TE1 / TE2 / TE3 / TE4 / TE5 = 13.8 / 70.4 / 127 / 183.6 / 240.2 ms are shown, with real brain mask (yellow) and predicted brain mask (red) respectively.

[0048] Figure 6 This is a histogram of Dice similarity coefficients obtained from brain masks automatically extracted from images at five echo times of 16 fetal subjects in this invention.

[0049] Figure 7 The density map of this invention represents the distribution of Dice coefficient scores over five echo times.

[0050] Figure 8 For this invention, Dice coefficient values ​​were calculated and plotted for each fetus at each echo time.

[0051] Figure 9Images of five fetal subjects (gestational ages 26.5 / 20.7 / 21 / 27 / 36) obtained at 0.55T and 5 echo times (46 / 120 / 194 / 268 / 342 ms) according to this invention are shown, including the brain mask predicted by the network (red) and the corresponding hand-drawn mask (yellow).

[0052] Figure 10 The coronal, sagittal, and axial views of the brain obtained from all ten prospectively acquired fetal datasets are shown in red, with the automatically implemented segmentation highlighted. Detailed Implementation

[0053] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0054] Combined with appendix Figure 1 In this process, real-time dynamic tracking and feedback are used to improve imaging accuracy. First, initial three-axis positioning is performed. Specifically, a standard T2-weighted sequence is used to initially acquire a three-dimensional image of the fetal brain. This step provides the foundational data for subsequent precise positioning. Once the initial three-axis position of the fetal brain is detected, the system activates a dual-modal motion tracking system using optical and magnetic resonance imaging to begin real-time monitoring of fetal movement. This tracking system combines optical capture with the magnetic field data inherent in MRI, thereby achieving high-precision capture of fetal movement.

[0055] Next, motion capture algorithms are used to acquire real-time positional information of fetal brain movement. This data is processed at high speed by an image processing chip (FPGA). The parallel processing capability of the FPGA allows for the computation of large amounts of data in a very short time, ensuring the real-time performance of the system. The processed motion information is then fed back to the MRI system to dynamically adjust the triaxial position of the current scan. In this way, the MRI system can adjust the angle and position of the triaxial scan based on the real-time movement of the fetus, ensuring that the triaxial scan maintains precise positioning of the fetal brain throughout fetal movement. This significantly reduces motion artifacts and improves image quality and accuracy. This real-time tracking and feedback mechanism is a significant innovation in optimizing triaxial positioning of the fetal brain, enabling the acquisition of high-quality MRI images even with continuous fetal movement, providing crucial support for clinical diagnosis and research.

[0056] Imaging quality is improved through precise algorithms and optimization strategies in motion compensation and image reconstruction.

[0057] First, based on a 3D motion compensation algorithm, interpolation calculations are performed on the fetal movement to compensate for image displacement errors in real time. This algorithm utilizes the trajectory information of fetal movement to interpolate and estimate the displacement at each time point, thereby correcting image shifts caused by motion in real time. Next, the real-time scanned images are processed using a Fast Fourier Transform (FFT) algorithm. FFT transforms the image data from the spatial domain to the frequency domain, allowing for clear identification and correction of phase shifts caused by motion. During reconstruction, the FFT-processed data is inversely transformed back to the spatial domain, precisely controlling the image reconstruction to be repositioned to the correct location of the fetal brain, ensuring image alignment and continuity. Simultaneously, the system continuously adaptively optimizes the scanning sequence, including adjusting the scanning angle and time interval, to adapt to the optimal imaging conditions under different motion states. The optimization process is dynamic, responding in real time to the subtle movements of the fetus.

[0058] When the system detects large-scale fetal movements, including rotation or displacement exceeding a threshold, it immediately pauses the current scan to avoid image distortion caused by violent movement. After pause, the system repositions itself along the three axes, reconfirms the spatial orientation of the fetal brain, and then continues scanning. This ensures that the image data from each scan is accurately repositioned to the correct location. Through this series of motion compensation and reconstruction optimization measures, the system can continuously acquire high-quality three-axis images of the fetal brain in dynamic motion, providing reliable image data support for accurate clinical diagnosis and research.

[0059] Artificial intelligence technology is used to adjust scanning parameters in real time to improve the accuracy and clarity of images.

[0060] First, the system analyzes fetal brain MRI images using a pre-trained 3D anatomical model to initially identify important anatomical regions of the brain, such as the cerebral cortex, corpus callosum, and thalamus. This 3D anatomical model, trained on a large amount of fetal MRI data, can accurately identify various parts of the fetal brain and continuously optimizes itself during the scanning process. Based on feedback from the AI ​​model, the system automatically determines the priority of different slice scans. This means that important anatomical regions are scanned at high resolution first, while less important regions have a reduced scan density to optimize scanning efficiency. During the scanning process, the AI ​​model analyzes each frame of the acquired image in real time, dynamically adjusting the slice thickness and spacing through a comprehensive evaluation of multiple parameters such as image quality, signal-to-noise ratio, and contrast. For important structures, the AI ​​instructs the system to reduce the slice thickness to obtain higher resolution images, while for regions with simple structures or less information, the slice thickness can be appropriately increased to reduce scanning time.

[0061] Optimizing the slice spacing ensures image continuity, preventing skipping or loss of anatomical structures. Simultaneously, the AI ​​model dynamically adjusts the slice angle based on real-time analysis to better match the natural anatomical orientation of the fetal brain, ensuring each slice covers key areas and maintains optimal imaging quality. Overall, this AI-driven adaptive slice optimization technology allows the scanning process to continuously adjust according to actual anatomical needs and fetal movement, ensuring the integrity, clarity, and accuracy of triaxial images, providing an intelligent solution for fetal brain MRI imaging.

[0062] Example 1:

[0063] Combined with appendix Figure 2 In a clinical case, an MRI of the fetal brain is being performed on a pregnant woman.

[0064] During the scanning process, the fetus's continuous movements, such as slight head rotations or minute displacements caused by breathing, can lead to blurred images or artifacts, severely affecting the doctor's assessment of the fetal brain development. To overcome these challenges, the system utilizes real-time motion compensation and reconstruction optimization techniques to process the acquired image data. First, the acquired spatial domain image I(x, y) is subjected to a Fast Fourier Transform (FFT) to transform it into the frequency domain F(u, v) to better reveal motion information in the image.

[0065] The formula is:

[0066]

[0067] Where F(u, v) represents the image data in the frequency domain, which includes motion artifacts and phase information; I(x, y) is the original spatial domain image, where x and y are the horizontal and vertical coordinates; u and v are the horizontal and vertical frequency coordinates in the frequency domain, representing the frequency information of the image; j is the imaginary unit, representing the imaginary part of the complex number; 2π(ux+vy) is the principal phase factor, characterizing the spatial frequency features of the image; θ(x, y) is the phase shift function, describing the phase change caused by fetal movement.

[0068] In this example, it is assumed that the fetal head experiences slight left-right swaying during the scan, and the resulting motion information can be described by the phase shift function θ(x, y). Specific data is set as follows:

[0069] I(x, y) is the original spatial domain image intensity of 0.8 × sin(2πx)·cos(2πy), simulating images of common complex brain tissue. The phase shift θ(x, y) caused by motion is fitted from fetal movement data detected by a motion sensor and is defined as θ(x, y) = 0.05x·cos(0.1y) (in radians), reflecting minute head movements.

[0070] The actual scan frequency coordinate range: the frequency domain range of u and v is [-5, 5] Hz, which covers the common spatial frequencies in fetal brain imaging.

[0071] Substitute into the formula to calculate:

[0072] 1. First, perform frequency domain transformation:

[0073]

[0074] 2. During the calculation, the phase shift term in j(2π(ux+vy)+0.05x·cos(0.1y)) significantly reflects the frequency domain changes caused by motion. Through this transformation, the frequency distribution of the image shows the trajectory characteristics of the fetal head's motion. Due to the introduction of θ(x,y), the phase change is concentrated in the low-frequency part, corresponding to larger spatial motion.

[0075] 3. By adjusting the frequency domain data, the system can identify specific phase shift patterns, match them with fetal movement data, and correct motion artifacts through inverse phase compensation.

[0076] Next, the corrected frequency domain data F′(u, v) is transformed back to the spatial domain using inverse fast Fourier transform (iFFT) to reconstruct the compensated image I′(x, y). At this point, the scanning time interval and angle are adaptively adjusted, and the anatomical structures of the current frame are compared with those of previous frames during image reconstruction to ensure that the image is repositioned to the correct location of the fetal brain.

[0077] During this process, if the detected angle of fetal head rotation reaches 5° (exceeding the system's set threshold of 3°), the system will automatically pause the current scan, reposition and adjust parameters, and then continue scanning to ensure image quality. This process is continuously optimized through adaptive algorithms, ensuring high-quality triaxial images can still be acquired even when fetal movements are frequent. The above example demonstrates the feasibility and specific operational procedures of the MRI-based triaxial positioning optimization method for fetal brain. Based on the actual measured parameter range, such as the phase offset θ(x, y) typically ranging from 0.01 to 0.1 radians, and the scanning frequencies u and v adjusted between [-10, 10] Hz, precise compensation for fetal movement and image reconstruction can be achieved, ensuring imaging accuracy under complex fetal movement conditions.

[0078] To address the issue of repositioning the image to the correct location of the fetal brain during image reconstruction, a nonlinear phase correction algorithm is employed to compensate for phase shifts caused by fetal movement, thereby ensuring the accuracy and continuity of the imaging.

[0079] After obtaining the frequency domain image F(u, v), a phase compensation function Φ(u, v, t) was designed by analyzing the fetal real-time motion data M(t) and frequency domain features to correct the phase error in the frequency domain and reconstruct an accurate image.

[0080] In this clinical scenario, the fetus exhibits significant vertical and horizontal movement during the scan. These movements are recorded in real-time by a motion sensor and converted into motion data M(t). The system needs to use this data to dynamically compensate for phase shifts in the scanned images. The compensation function is defined as follows:

[0081] Φ(u,v,t)=a·sin(b·f(u)·cos(g(v)+h(t)))+c·e -d·t

[0082] Where a, b, c, and d are parameters controlling phase compensation, with their values ​​set to the following ranges: a = 0.5–1.0, b = 0.1–0.5, c = 0.2–0.8, and d = 0.05–0.1. These parameters control the amplitude, frequency response, nonlinear amplitude adjustment, and time decay of the phase compensation, respectively, ensuring that the compensation changes with time and frequency accurately match the actual fetal movements. f(u) = u 2 +0.3u represents a custom adjustment function for the horizontal frequency coordinate u, used to correct the horizontal component of the phase shift; specifically, when u = -5, f(u) = 22.75, reflecting the correction force for horizontal motion. g(v) = v 3 -0.2v represents a custom adjustment function for the vertical frequency coordinate v, used to correct the vertical component of the phase shift; if v = 3, then g(v) = 26.4, adjusting the effect of vertical motion on the phase.

[0083] h(t) = 0.1·M(t) is a time-dependent motion prediction function that directly reflects the real-time motion characteristics of the fetus. M(t) is set to indicate that the fetal motion exhibits periodic changes during the scanning process. For example, if M(t) = 2sin(0.2t), then h(t) = 0.2sin(0.2t) is used to predict instantaneous motion changes.

[0084] Substitute the actual data for compensation calculation. Set the scanning time t = 4 seconds, the fetal motion parameter M(t) = 1.5, the corresponding time correlation function h(t) = 0.15, and set the current frequency domain point u = 2Hz, v = -3Hz.

[0085] Calculating f(u) yields f(2) = 4.6, which reflects the adjustment of the horizontal component.

[0086] Calculating g(v), we get g(-3) = -26.7, which characterizes the effect of motion in the vertical direction.

[0087] Substitute into the compensation function:

[0088] Φ(2,-3,4)=0.8·sin(0.3·4.6·cos(-26.7+0.15))+0.5·e -0.084

[0089] The calculation yielded:

[0090] Φ(2,-3,4)=0.8·sin(1.38·cos(-26.55))+0.5·e -0.32

[0091] ≈0.8·sin(-0.94)+0.5·0.726

[0092] ≈-0.68+0.363

[0093] ≈-0.317

[0094] This compensation value is used to adjust the phase shift of the image in the frequency domain, ensuring that the reconstructed image remains accurately positioned even when the fetus undergoes significant movement. By repeatedly applying the phase compensation function, the system can continuously correct motion artifacts in the frequency domain and gradually reduce phase errors. Meanwhile, the time decay term e... -d·t Ensure that the compensation intensity decreases over time to adapt to dynamic motion characteristics and prevent overcompensation from causing image artifacts or distortion.

[0095] In practical applications, this nonlinear phase correction function allows the MRI system to make highly customized compensation adjustments based on the fetal movement characteristics. This method effectively avoids image blurring caused by frequent fetal movements, greatly improving the accuracy of triaxial image reconstruction. Especially in more complex motion modes, such as large head rotations or displacements, the system can pause the scan, reposition the three axes, and then apply the compensation function to ensure image continuity and accuracy.

[0096] In the example of the above-mentioned optimized method for MRI localization of the fetal brain in three axes, considering the accurate reconstruction and repositioning correction of the image after compensating for phase shift, the inverse fast Fourier transform (iFFT) is used to convert the adjusted frequency domain data F′(u, v) back to the spatial domain to obtain the compensated image I′(x, y) repositioned to the correct position.

[0097] In actual clinical applications, fetuses often exhibit slight or significant movements during scanning, such as rotation and displacement. These movements can lead to image misalignment and blurring. To ensure the accuracy of image reconstruction, the system reconstructs the image using iFFT after compensating for phase shift. Simultaneously, a reset adjustment strategy is employed to further refine the image position by comparing the positional information of the current frame with that of adjacent frames.

[0098] In a practical application, a fetal brain MRI is performed on a pregnant woman at 24 weeks gestation. Due to the continuous small head rotations (approximately 3°–5°) during the scan, combined with real-time motion data M(t), the system first performs phase compensation to correct phase errors in the frequency domain. The compensated frequency domain data F′(u, v) contains the nonlinearly corrected phase information, which then needs to be restored to the spatial domain through inverse transformation and reset for further correction.

[0099] The formula for reconstructing the image is:

[0100]

[0101] Where I′(x, y) is the reconstructed image, with phase error compensated and reset to the correct spatial position; F′(u, v) is the compensated frequency domain data. The key component in this reconstruction formula is the dynamic adjustment function Δφ(t), which is defined as:

[0102] Δφ(t)=α·(t-t0) 2 +β·sin(γ·t)

[0103] Here, Δφ(t) is the dynamic adjustment function for phase reset, used to control the reset correction of the image and ensure the continuity of each scan frame. α, β, and γ are adjustment coefficients, controlling the response amplitude of the reset process to changes in time. The set range is α = 0.01–0.05, β = 0.1–0.3, and γ = 0.5–1.5. These coefficients were determined experimentally to accommodate different fetal movements. (t-t0) represents the difference between the current time t and the initial time t0, reflecting the time trend of phase reset; a larger value indicates a more significant phase shift correction. sin(γ·t) is the periodic phase change, reflecting the natural periodic movements of the fetal head, such as minute rotations and subtle tremors caused by breathing.

[0104] In this example, we set a time t = 5 seconds and an initial scan time t0 = 1 second. At this time, the fetal brain's movement causes errors in the image. Substitute the parameters:

[0105] The adjustment coefficients are set as follows: α = 0.03, β = 0.2, γ = 1.0.

[0106] Calculate the phase reset function:

[0107] Δφ(5)=0.03·(5-1) 2 +0.2·sin(1.0·5)

[0108] Calculation process:

[0109] Δφ(5)=0.03·16+0.2·sin(5)

[0110] ≈0.48+0.2·(-0.96)

[0111] =0.48-0.192

[0112] =0.288

[0113] This phase correction value Δφ(5) = 0.288 radians indicates that at t = 5 seconds, the corresponding image needs a phase correction of 0.288 radians to ensure that the reconstructed image can be accurately repositioned to the correct location of the fetal brain. During the iFFT reconstruction process, this correction value is applied to all frequency domain data points for overall phase adjustment. The adjusted image is compared with the positions of the preceding and following frames in real time to ensure the continuity and accuracy of the imaging.

[0114] Example 2:

[0115] Combined with appendix Figure 3 In a real-world clinical application, doctors were performing a triaxial MRI scan of the fetal brain on a pregnant woman at 28 weeks gestation. During the scan, slight head movements and breathing caused changes in the brain's position, resulting in significant fluctuations in image quality. The system incorporated an AI model to analyze each frame of the triaxial image in real time, using a quality scoring function to assess image quality and ensure that each frame met diagnostic requirements.

[0116] The three-axis image quality scoring function is:

[0117]

[0118] Among them: Q xyz It is a comprehensive quality score for three-axis images, evaluating the overall imaging effect of coronal, sagittal, and transverse images. A xyz The integral region represents the anatomical region of interest in three-axis imaging, such as the cerebral cortex and corpus callosum. R(x, y, z) is the three-dimensional resolution function, representing the image resolution along each axis (x, y, z). A higher value indicates clearer image details. The set range is R(x, y, z) = 1.0–5.0. N(x, y, z) is the noise distribution function, reflecting the noise level along each axis. Lower values ​​indicate less noise interference. The set range is N(x, y, z) = 0.1–1.0. C(x, y, z) is the contrast function, measuring the contrast intensity between different anatomical structures. A higher C value indicates better structural discrimination ability. The set range is C(x, y, z) = 1.0–3.0.

[0119] In practical applications, when setting the current time to perform a triaxial image scan, the system receives the following data:

[0120] For region of interest Axyz The system selected the cerebral cortex and corpus callosum as the main anatomical regions for scoring.

[0121] The actual measured resolution function R(x, y, z) = 4.2 indicates that the image has high detail resolution in these areas.

[0122] The noise level N(x, y, z) = 0.5, indicating a moderate level of noise interference during the imaging process.

[0123] The contrast ratio C(x, y, z) = 2.5, indicating good anatomical structure resolution.

[0124] Substitute into the quality scoring formula:

[0125]

[0126] Set the volume of the integration region Vol(A) xyz ) = 10mm 3 ,but:

[0127] Q xyz =21×10=210

[0128] Rating result Q xyz =210 reflects that the current scanned image quality is high and meets the imaging requirements. However, if the noise level suddenly increases at some point, for example, N(x,y,z) = 0.9, then the result is recalculated as follows:

[0129]

[0130] Q at this time xyz A score drop to 116.75 indicates a significant decline in image quality, falling below the set imaging standard (threshold value is 150). The AI ​​model will use this score to provide real-time feedback to the scanning control system, optimizing imaging conditions by adjusting scanning parameters such as slice thickness, spacing, and angle. For example, the system might reduce slice thickness, increase the number of scans, or adjust the angle to reduce noise, while simultaneously optimizing contrast to ensure the imaging quality returns to the standard.

[0131] In this case, doctors were performing a brain MRI on a fetus at 28 weeks of gestation, with the cerebral cortex being a key area of ​​focus as it is crucial for monitoring fetal brain development. However, during the imaging process, the system detected fetal movement, causing fluctuations in image quality. An AI model evaluated the image quality score in real time and dynamically adjusted the slice thickness and spacing based on the score to ensure imaging quality in critical areas such as the cerebral cortex, while reducing scanning time for less important areas to optimize overall efficiency.

[0132] The core formula for optimizing slice thickness and spacing is:

[0133]

[0134] in:

[0135] T xyz (h, z) represents the dynamically optimized slice thickness in the three-axis view, adjusted in real-time by the AI ​​model to balance resolution and scanning efficiency. h is the sharpness score of the current three-axis image, evaluated in real-time by the AI ​​model, reflecting the overall detail quality of the image; its value range is h = 1.0–5.0, with higher scores indicating sharper images. z is the importance index of anatomical structures, measuring the need for high-resolution imaging in different regions; its value range is z = 1.0–3.0, with higher values ​​indicating more critical regions. α xyz It is a constant coefficient that controls the range of thickness adjustment in the three axes; the set range is α. xyz =0.5~1.5, used to adjust the thickness adjustment sensitivity in different areas.

[0136] γ represents the derivative of image sharpness with respect to slice thickness, used to assess the impact of thickness on image quality. xyz This involves adjusting parameters to control the degree to which anatomical importance affects slice thickness, with a set range of γ. xyz =0.8~1.2, ensuring a smooth adjustment range. D xyz (z) is the anatomical complexity function, which describes the structural complexity of a specific anatomical region, and its value ranges from D. xyz (z) = 1.0–4.0, reflecting the complexity of different brain regions. log(γ) xyz ·D xyz (z) is used to perform a logarithmic transformation on the complexity to smooth the adjustment process and avoid over-adjustment caused by excessive complexity.

[0137] In practical applications, the system receives the following data:

[0138] For the cerebral cortex region, the AI ​​model scored h=4.8, reflecting high-quality image detail; the importance index of anatomical structures was z=3.0, indicating a very high demand for high resolution.

[0139] The constant coefficient is set as α xyz =1.2, adjust parameter to γ xyz =1.0, while the complexity function D of the cerebral cortex xyz (z) = 3.5, indicating that the anatomical structure of the cerebral cortex is very complex and requires high image resolution. (Setting) This indicates that image sharpness is quite sensitive to slice thickness; even a small adjustment in thickness can significantly improve image quality.

[0140] Substitute into the formula to calculate the slice thickness:

[0141] T xyz (4.8, 3.0)=1.2·(0.7·log(1.0·3.5))=1.2·(0.7·log(3.5))≈1.2·(0.7·1.25)

[0142] =1.2·0.875≈1.05

[0143] Calculation result T xyz (4.8, 3.0) = 1.05 mm indicates that the system dynamically adjusted the slice thickness through AI model analysis, optimizing it to 1.05 mm to ensure high resolution in cerebral cortex imaging and to clearly display the details of fetal brain development. In other areas with less information, such as the ventricular system, z = 1.5 and image sharpness score h = 3.0 were set, using the same parameters:

[0144] T xyz (3.0, 1.5)=1.2·(0.7·log(1.0·2.0))=1.2·(0.7·0.69)≈1.2·0.483≈0.58

[0145] In this scenario, the system adjusts the slice thickness to 0.58 mm, slightly thicker than a scan of the cerebral cortex, to reduce scanning time and improve imaging efficiency. This adaptive adjustment mechanism for slice thickness effectively improves image quality in key anatomical regions through real-time analysis and parameter optimization, while reducing unnecessary scanning burden in secondary areas, resulting in more efficient and accurate overall imaging.

[0146] In this example, a doctor is performing a triaxial MRI scan on a fetus at 30 weeks of gestation. Due to frequent fetal movements, especially slight head rotations and vertical movements, positional shifts occur in the images. To ensure imaging accuracy, the system uses an AI model to analyze the current triaxial slice angles in real time and compares them with the ideal anatomical orientation. It calculates the angle deviation that needs adjustment and dynamically adjusts the scan direction to match the actual position of the fetal brain. The core formula for the slice angle adjustment function is:

[0147]

[0148] in:

[0149] Θ xyz (t) represents the correction value of the slice angle in the three-axis position at the current time t, which is used to dynamically adjust the scanning direction to match the natural position of the fetal brain;

[0150] E xyz (x, y, z, τ) is the error function between the current triaxial slice and the ideal anatomical direction, reflecting the deviation between the current scanning angle and the ideal angle. The larger the error, the more the current angle deviates from the ideal direction.

[0151] The time rate of change of error reflects how quickly the error changes over time and is used to predict future error trends and adjust the angle to correct the speed.

[0152] A xyz (τ) is an angle change function that describes the dynamic change of the slice angle over time and can track the rhythm of fetal movement;

[0153] ρ xyz This is an adjustment coefficient that controls the degree of influence of the error on the angle correction; its value range is ρ. xyz =0.1~0.5, to ensure that angle adjustments can smoothly reflect the current error;

[0154] κ xyz The stability coefficient controls the smoothness of the angle adjustment process, and its set range is κ. xyz =0.05~0.2, to prevent the angle adjustment process from being too abrupt;

[0155] ω xyz It adjusts the frequency coefficient to control the frequency of angle correction, with a value range of ω. xyz =0.5~1.0, reflecting the response speed to periodic motion;

[0156] sin(ω xyz ·A xyz (τ) is a periodic angle adjustment term used to handle the periodic deviation of the fetal brain caused by breathing or small movements;

[0157] t0 and t represent the time range of the adjustment process, and the cumulative effect of the angle correction is calculated by integration.

[0158] In practical applications, with the current time t = 6 seconds and the initial time t0 = 0 seconds, the system detects a deviation in the fetal brain. Comparing this to the ideal anatomical orientation, the system obtains the current error function E. xyz (x, y, z, τ) = 0.3·τ, representing the angle deviation gradually increasing over time, indicating that the fetal head is slowly turning. The rate of change over time is... This indicates that the deviation increases steadily over time.

[0159] The parameters are set as follows:

[0160] Adjustment coefficient ρ xyz =0.4;

[0161] Stability coefficient κ xyz =0.1;

[0162] Adjusting the frequency ω xyz =0.8;

[0163] Angle change function A xyz (τ)=0.5·T.

[0164] Substitute into the formula to calculate the angle correction value:

[0165]

[0166] First calculate Recalculate The integral term of sin(0.4τ) can be roughly estimated as follows: From τ = 0 to τ = 6, we get:

[0167]

[0168] Therefore, the angle correction value is:

[0169] Θ xyz (6) = 0.72 + 0.07 = 0.79 radians.

[0170] This 0.79 radian angle correction is used to dynamically adjust the slice angle of the current scan, making it more closely match the natural anatomical position of the fetal brain. Through real-time feedback from the AI ​​model and this calculation method, the system can continuously fine-tune the scan angle, ensuring that the MRI images maintain a high degree of alignment and continuity even when the fetal head frequently rotates slightly. This fine-grained dynamic adjustment of the slice angle greatly improves the accuracy of triaxial imaging, providing doctors with clearer, more coherent, and accurate images of the fetal brain, supporting the scientific diagnosis of fetal development.

[0171] Example 3:

[0172] To achieve high-quality three-dimensional image reconstruction of the liver, gallbladder, and pancreas, this invention draws on real-time fetal MRI tracking technology, combining deep learning models, dynamic tracking, and triaxial optimized scanning to ensure high-quality images even in complex motion environments. The following is a summary of... Figures 4 to 10 The analysis is carried out one by one, and the specific implementation details of the present invention in the three-dimensional image reconstruction of the liver, gallbladder and pancreas are incorporated.

[0173] Figure 4This paper presents fetal MRI images at five different gestational weeks, acquired at five different echo times (TE = 10 ms, 52 ms, 98 ms, 142.5 ms, and 240 ms). The figures show brain masks predicted by the 3DUNet network (red) and manually labeled masks (yellow), with the accuracy of automatic and manual segmentation compared using the Dice similarity coefficient (DSC). DSC values ​​ranged from 79.2% to 86.0%, indicating differences in performance of automatic segmentation at different echo times. This technique directly inspired the reconstruction of three-dimensional images of the liver, gallbladder, and pancreas, as the morphology and boundaries of these organs constantly change during respiration, gastrointestinal motility, and other movements.

[0174] In this invention, a similar deep learning model is used to segment and track the hepatobiliary and pancreatic regions in real time. For example, during a hepatobiliary scan, when the system detects displacement of the liver boundary due to respiration, the system first uses the 3DUNet model to segment the liver boundary and then scores the quality based on the segmentation results.

[0175]

[0176] At some point, the system detected score Q. xyz =180, slightly below the ideal threshold of 200. At this point, the system will automatically reduce the slice thickness and fine-tune the scanning angle to further improve image clarity and resolution, avoiding misalignment and blurring. For critical anatomical regions, such as the hepatic hilum, the system will increase the scanning priority and adjust the echo time to optimize imaging results.

[0177] Figure 5 MRI images of fetus 5 at five different echo times (TE1 = 13.8 ms, TE2 = 70.4 ms, TE3 = 127 ms, TE4 = 183.6 ms, TE5 = 240.2 ms) are shown, illustrating the overlap between the real mask (yellow) and the predicted mask (red). As the TE time increases, the DSC decreases from 87.9% to 56.9%, indicating a significant deterioration in image quality over time. Similar phenomena occur during hepatobiliary and pancreatic imaging in this invention, under different functional states and motion amplitudes. For example, when the DSC is detected to be below a set 60% threshold, the system initiates the following adjustment process:

[0178] 1. Adjust the slice thickness T xyz (h, z) to ensure high resolution, especially in important anatomical regions such as the gallbladder and head of the pancreas:

[0179] T xyz (h, z)=1.2·(0.7·log(1.0·3.5))≈1.05mm

[0180] 2. Dynamically adjust the slice angle to match the natural position during movement, using the angle correction formula:

[0181] Θ xyz (t)=∫0 6 (0.4·0.3+0.1·sin(0.8·0.5·τ))dτ=0.79 radians

[0182] This adjustment process ensures that even at high echo times, images of the hepatobiliary region maintain good spatial alignment and clarity.

[0183] Figure 6 The DSC histograms are shown for different echo times. Figure 7 The density distribution of DSC is shown. Figure 8 The figure shows the DSC values ​​for each fetus at different TE times. The DSC values ​​in the figure reflect the impact of different echo times on segmentation accuracy. In the hepatobiliary and pancreatic reconstruction process of this invention, DSC is used to assess image quality. Through real-time scoring, the system can quickly identify image quality problems and take corrective measures. For example, in the scanning of the hepatobiliary region, if the DSC score of a certain scan drops below 0.6, the system immediately feeds back to the control module to adjust the scanning parameters, including increasing the number of scans or changing the slice spacing, to ensure that the imaging process adapts to the current anatomical changes and motion state.

[0184] Furthermore, the system adjusts the priority of the scanning sequence based on the scoring results. For example, when the system detects an unclear boundary between the liver and gallbladder, it immediately adjusts the TE time to the optimal range (e.g., 70ms to 100ms) and compares the current frame with adjacent frames in real time to ensure continuity and accuracy. This technology guarantees that the data from each scan meets the expected diagnostic criteria, greatly improving the stability and continuity of image reconstruction.

[0185] Figure 9 Five fetal MRI images under a 0.55T magnetic field are presented, with echo times of 46 ms, 120 ms, 194 ms, 268 ms, and 342 ms, respectively. The predicted brain mask and the manually drawn mask show DSC ranges from 79.9% to 90.1%, demonstrating segmentation performance under low magnetic field strength. This method also has important implications for hepatobiliary and pancreatic reconstruction because noise levels are higher and motion artifacts are more pronounced at low field strengths.

[0186] To address these challenges, this invention applies deep learning segmentation and dynamic adjustment techniques under 0.55T conditions, reducing artifacts in low-field imaging through slice optimization and angle correction. Specifically, the system first uses AI to assess the current image quality and calculates the error function E. xyz Then, based on the real-time error change rate Make fine adjustments:

[0187] Θ xyz (t)=∫0 6 (0.3·0.2+0.15·sin(0.7·0.5·τ))dτ≈0.6 radians

[0188] This adjustment ensures that the scanning angle can adapt to the current low field strength environment, significantly improving the recognition accuracy of liver, gallbladder and pancreas tissues under low field imaging, and providing reliable image support for subsequent three-dimensional reconstruction.

[0189] Figure 10 The paper presents the results of automatic brain segmentation from different views (coronal, sagittal, and axial) across 10 fetal datasets, demonstrating the adaptability of the 3DUNet network under various conditions. For hepatobiliary and pancreatic reconstruction, a similar segmentation technique is used to identify and adjust the location of hepatobiliary and pancreatic tissues in real time. In practical applications, the system segments and assesses the quality of the current scan results every second, adjusting slice position and scanning angle by comparing the boundary consistency between preceding and following frames.

[0190] For example, in the junction region between the liver and pancreas, due to the complexity of the tissue, when the AI ​​system detects discontinuities in the segmentation boundary, it will correct them through the following adjustment mechanism: first, it detects the dynamic changes in the boundary and calculates the necessary slice spacing adjustment value T. xyz Simultaneously, the scanning angle is adjusted in real time to the optimal state to ensure seamless integration of all slices during 3D reconstruction. These real-time adjustments ensure that 3D imaging of the liver, gallbladder, and pancreas maintains high quality and consistency even under complex organ movements.

[0191] This invention achieves high-quality three-dimensional image reconstruction of the liver, gallbladder, and pancreas by integrating deep learning segmentation, dynamic tracking, and three-axis slice optimization techniques. In its implementation, it draws upon... Figures 1 to 5 The deep learning network and real-time tracking technology can effectively overcome the motion artifact problem in traditional MRI, and improve imaging accuracy and diagnostic reliability.

Claims

1. An optimized method for MRI localization of the fetal brain in three axial planes, characterized in that... Includes the following steps: S1. Real-time dynamic tracking and feedback: S1.1 Initialization of triaxial localization: Initially obtain three-dimensional images of the fetal brain using standard T2-weighted sequences; S1.2 Once the initial triaxial position of the fetal brain is detected, the optical and magnetic resonance dual-modal motion tracking system is activated to begin real-time monitoring of fetal movement; S1.

3. Real-time acquisition of fetal brain movement position information using motion capture algorithms and processing via an image processing chip FPGA; feeding back the real-time fetal movement information to the MRI system to adjust the current three-axis position of the scan. S2. Motion Compensation and Image Reconstruction: S2.1 Based on the three-dimensional motion compensation algorithm, the motion of the fetus is interpolated and calculated to compensate for image displacement error in real time; S2.

2. The real-time scanned image is processed by the Fast Fourier Transform (FFT) algorithm to control the image reconstruction process and reset it to the correct position of the fetal brain. S2.3 Continuously perform adaptive optimization of the scanning sequence, including scanning angle and time interval; when a large-scale movement of the fetus, including rotation or displacement, exceeding the threshold is detected, pause the current scan and reposition the three axes; S3, AI-driven adaptive slicing optimization: S3.1 Using a pre-trained three-dimensional anatomical model, analyze MRI images of the fetal brain to initially identify important anatomical regions; S3.2 Determine the priority of slice scanning based on feedback from the AI ​​model; S3.3 During the scanning process, the AI ​​model analyzes the acquired images, optimizes the slice thickness and spacing in real time, and dynamically adjusts the slice angle to obtain a qualified imaging effect.

2. The optimized method for MRI localization of the fetal brain in three axial planes according to claim 1, characterized in that... The method for resetting the image to its correct position in the fetal brain during image reconstruction includes: performing a Fast Fourier Transform (FFT) on the acquired spatial domain image I(x,y) to convert it to the frequency domain F(u,v); in the frequency domain, the motion information of the image is represented by specific phase changes and frequency shifts, significantly revealing the details of the motion trajectory; wherein the frequency domain transformation and phase representation are as follows: Where F(u,v) represents the image data in the frequency domain, which includes motion artifacts and phase information; I(x,y) represents the original image data in the spatial domain, where x and y are the horizontal and vertical coordinates of the image, respectively; u and v are the horizontal and vertical frequency coordinates in the frequency domain, representing the image frequency information; j is the imaginary unit, representing the imaginary part in complex number operations; 2π(ux+uy) is the principal phase factor in the domain, characterizing the spatial frequency features of the image; θ(x,y) is the phase shift function of the image, describing the phase change caused by fetal movement.

3. The optimized method for MRI localization of the fetal brain in three axial planes according to claim 2, characterized in that... The method for resetting the image to the correct position in the fetal brain during image reconstruction includes: after acquiring the frequency domain image, a nonlinear phase correction algorithm is used for compensation; a complex compensation function is designed based on the fetal real-time motion data and frequency domain characteristics; the compensation function integrates frequency coordinates and time factors to perform inverse transformation on the phase shift caused by fetal movement.

4. The optimized method for MRI localization of the fetal brain in three axial planes according to claim 3, characterized in that... The method for resetting the image to its correct position in the fetal brain during image reconstruction includes: after compensating for phase shift, using inverse fast Fourier transform (iFFT) to reconstruct the adjusted frequency domain data F. ′ (u,v) is transformed back to the spatial domain to reconstruct the compensated image; simultaneously, during image reconstruction, a reset adjustment is performed by comparing the current frame with adjacent frames; FT reconstruction and reset correction are employed. Among them, I ′ (x,y) represents the reconstructed image after compensation, restored to the correct spatial position; F ′ (u,v) represents the frequency domain data after phase compensation; Where, Δφ(t)=α·(t-t0) 2 +β·sin(γ·t) represents the dynamic adjustment function for phase reset; α, β, and γ are adjustment coefficients that control the effect of time changes on reset; (t-t0) represents the difference between the current time t and the initial time t0, describing the trend of phase change over time; sin(γ·t) describes the periodic phase change, adapting to the natural movement of the fetal brain.

5. The optimized method for MRI localization of the fetal brain in three axial planes according to claim 1, characterized in that... The method for obtaining qualified imaging results includes: introducing a real-time analysis scanning image mode, scoring the quality of triaxial slices, and dynamically adjusting parameters during the scanning process; the AI ​​model receives triaxial MRI image data in real time and calculates the image quality through the scoring function; the scoring result is used to determine whether the current scanning parameters meet the imaging requirements and is fed back to the scanning control system for parameter adjustment.

6. The optimized method for MRI localization of the fetal brain in three-axis planes according to claim 5, characterized in that... The method for achieving satisfactory imaging results includes: after obtaining an image quality score, adjusting the slice thickness and spacing based on the analysis results of the AI ​​model; requiring fine slice thickness for key areas of the cerebral cortex during fetal brain development; and increasing the thickness in areas with less information to reduce scanning time; the slice thickness and spacing optimization function is as follows: Among them, T xyz (h, z) represents the dynamically optimized slice thickness in the three-axis view; h is the sharpness score of the current three-axis image, evaluated by the AI ​​model, reflecting the overall detail quality of the image; z is the importance index of anatomical structures, measuring the need for high-resolution imaging in different regions; α xyz It is a constant coefficient that controls the range of thickness adjustment in the three-axis position; γ represents the derivative of image sharpness with respect to slice thickness, used to dynamically evaluate the response of image quality to thickness adjustments; xyz To adjust parameters and control the degree to which anatomical importance affects slice thickness; D xyz (z) is the anatomical complexity function, describing the structural complexity of a specific anatomical region; log(γ) xyz ·D xyz (z) represents the process of performing a logarithmic transformation on the complexity to smooth it out.

7. The optimized method for MRI localization of the fetal brain in three axial planes according to claim 6, characterized in that... The method used to obtain qualified imaging results includes: fetal movement, dynamic adjustment of slice angles in the three-axis view to match the natural anatomical orientation of the fetal brain; and AI model calculating the angle deviation that needs to be adjusted by analyzing the current angles of the three-axis view in real time and comparing them with the ideal anatomical structure orientation.

Images