Intraoperative low-field magnetic resonance intelligent navigation method based on preoperative high-field data guidance

Abstract

The invention relates to the technical field of magnetic resonance imaging, in particular to an intra-operative low-field magnetic resonance intelligent navigation method based on preoperative high-field data guidance, which comprises the following steps: acquiring a high-field MRI image of a patient before an operation and extracting a three-dimensional segmentation mask and clinical features of a focus; adjusting according to the clinical features to obtain an intraoperative scanning protocol; and performing low-field magnetic resonance examination during the operation to generate an intra-operation navigation image, and marking focus information. Aiming at the problem that the accuracy of a focus image provided by an ultra-low field system in the operation process in the prior art is not enough, a high-field MRI image collected before the operation is introduced as a reference, scanning sequence parameters of the ultra-low field system used in the operation are dynamically adjusted based on clinical features embodied in the high-field MRI image, a better imaging effect is achieved, and meanwhile, the accuracy of the focus image provided by the ultra-low field system in the operation process is improved. Focus information is marked on the intraoperative navigation image based on the three-dimensional segmentation mask as prior information, so that more refined segmentation of focuses on the intraoperative navigation image is realized, and the surgical requirements are met.

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging technology, and specifically to an intraoperative low-field magnetic resonance intelligent navigation method based on preoperative high-field data guidance. Background Technology

[0002] In recent years, low-field (LF) and ultra-low-field (ULF) magnetic resonance imaging (MRI) technologies have attracted widespread attention both domestically and internationally. Advances in related technologies have enabled MRI to offer advantages such as good openness, ultra-quiet operation, no need for electromagnetic shielding, miniaturization, and portability to the patient's bedside. Under low-field or ultra-low-field conditions, it can achieve imaging with common clinical contrasts, including T1-weighted, T2-weighted, fluid attenuated inversion recovery (FLAIR), and diffusion-weighted imaging (DWI), providing clinically valuable information for the diagnosis of stroke and tumors. Based on these characteristics, existing clinical research is gradually exploring the use of ultra-low-field MRI systems intraoperatively to provide navigation fields for neurosurgery. The use of ultra-low-field MRI systems can reduce reliance on electromagnetically shielded operating rooms.

[0003] For example, patent application CN202510725630.3 discloses a portable ultra-low field magnetic resonance imaging (MRI) device suitable for use in operating rooms, comprising: a scanning device including an ultra-low field permanent magnet, a gradient coil, and a radio frequency coil, wherein the magnetic field strength generated by the ultra-low field permanent magnet is no higher than 100mT, and the radio frequency coil is a transceiver coil; a motion actuator including a drive motor and wheels for carrying the scanning device to a target position; a sensor device including at least one of a camera, lidar, and ultrasonic distance sensor, configured to sense the surrounding environment of the portable ultra-low field MRI device and its position information within that environment; and a motion control unit configured to send control commands to the motion actuator based on the path planning obtained from the sensing information of the sensor device. By automatically adjusting the movement path of the MRI device, automated and intelligent control of the MRI device is achieved, reducing the workload of operators and improving the convenience and safety of the equipment.

[0004] However, in actual implementation, the inventors found that the ultra-low / low field strength resulted in a low signal-to-noise ratio and poor tissue contrast, making it difficult to identify small residual lesions and failing to meet the accuracy requirements for intraoperative navigation. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, a method for intelligent intraoperative low-field magnetic resonance imaging navigation based on preoperative high-field data guidance is provided.

[0006] The specific technical solution is as follows: A method for intelligent intraoperative low-field magnetic resonance imaging navigation based on preoperative high-field data guidance includes: Step S1: Before the operation, high-field MRI images of the patient are acquired and segmented to obtain a three-dimensional segmentation mask corresponding to the lesion, and feature extraction is performed to obtain clinical features; Step S2: Adjust the intraoperative scanning sequence parameters according to the clinical characteristics to obtain the intraoperative scanning protocol; Step S3: During the operation, a low-field magnetic resonance imaging examination is performed on the patient based on the intraoperative scanning protocol to generate an intraoperative navigation image, and lesion information is marked in the intraoperative navigation image based on the three-dimensional segmentation mask.

[0007] On the other hand, the method for adjusting the intraoperative scanning sequence parameters in step S2 includes: When the lesion is adjacent to a brain functional area, T2WI weighting is increased in the intraoperative scanning protocol to shorten the TR / TE. When the lesion contains cystic changes, a FLAIR sequence is added to the intraoperative scanning protocol; When the volume of the lesion is smaller than the preset volume, the intraoperative scanning protocol is adjusted to prioritize high-resolution acquisition.

[0008] On the other hand, step S1 includes: Step S11: Acquire the high-field MRI image of the patient, and identify and segment the lesion in the high-field MRI image to obtain a three-dimensional pre-segmentation mask; Step S12: Perform morphological processing on the three-dimensional pre-segmentation mask to obtain the three-dimensional segmentation mask; Step S13: Extract the clinical features from the three-dimensional segmentation mask.

[0009] On the other hand, in step S11, the high-field MRI image includes a three-dimensional MRI image formed by multiple scanning sequences; Step S11 uses a multimodal 3D convolutional neural network model to segment the lesion into regions, including: normal brain tissue, solid tumor portion, cystic / necrotic region, and peritumoral edema region.

[0010] On the other hand, step S12 includes: Step S121: Perform morphological closing operations on the three-dimensional pre-segmented mask in different mask types to obtain multiple connected components; Step S122: Perform pixel threshold filtering on the connected components to remove isolated noise; Step S123: For the remaining connected components, retain the largest connected region as the output of the three-dimensional segmentation mask of the corresponding mask type.

[0011] On the other hand, the lesion features extracted in step S13 include anatomical location features; The first extraction process for extracting the anatomical location features includes: Step A131: Register the three-dimensional segmentation mask and map it onto a standard brain atlas; Step A132: Calculate the Euclidean distance between the tumor boundary and the functional area, and use a distance threshold to filter and obtain the neighboring functional areas; Step A133: Based on the anatomical location characteristics described in the adjacent functional areas.

[0012] On the other hand, the lesion features extracted in step S13 include internal structural features; The second extraction process for extracting the internal structural features includes: Step B131: Detect the signal intensity in the high-field MRI image; Step B132: Determine the specific internal structure according to the signal strength matching detection rule; The specific internal structures include cystic degeneration, necrosis, and hemorrhage; Step B133: Mark the specific internal structure as the internal structure feature in the three-dimensional segmentation mask.

[0013] On the other hand, the lesion features extracted in step S13 include volumetric and morphological features; The third extraction process for extracting the volumetric morphological features includes: Step C131: Perform voxel quantification on the three-dimensional segmentation mask to obtain the total volume; Step C132: Calculate the length of the major axis of the circumscribed ellipsoid of the three-dimensional segmentation mask as the maximum diameter of the lesion region; Step C133: Calculate the fractal dimension of the three-dimensional segmentation mask; The volumetric morphological features include the total volume, the maximum diameter of the lesion region, and the fractal dimension.

[0014] On the other hand, step S3 includes: Step S31: Acquire the intraoperative navigation image for the patient, register the intraoperative navigation image with the high-field MRI image, and then transfer the three-dimensional segmentation mask onto the intraoperative navigation image to form a mapping mask; Step S32: Based on the mapping mask as the initial condition, perform fine segmentation on the intraoperative navigation image to obtain an intraoperative segmentation map; Step S33: Quantitatively analyze the intraoperative segmentation map to obtain the lesion information.

[0015] On the other hand, step S32 includes: The mapping mask is used as a spatial prior channel and is input into a dual-branch code along with the intraoperative segmentation map for fusion to achieve segmentation.

[0016] The above technical solution has the following advantages or beneficial effects: To address the issue of insufficient precision in lesion images provided by existing ultra-low field systems during surgery, this solution introduces preoperative high-field MRI images as a reference. Based on the clinical features revealed in the high-field MRI images, the scanning sequence parameters of the ultra-low field system used during surgery are dynamically adjusted, achieving better imaging results. Simultaneously, based on a three-dimensional segmentation mask as prior information, lesion information is annotated on the intraoperative navigation image to achieve more refined segmentation of the lesion on the intraoperative navigation image, meeting surgical requirements. Attached Figure Description

[0017] Embodiments of the invention will be described more fully with reference to the accompanying drawings. However, the drawings are for illustration and explanation only and do not constitute a limitation on the scope of the invention.

[0018] Figure 1 This is an overall schematic diagram of an embodiment of the present invention; Figure 2 This is a schematic diagram of step S1 in an embodiment of the present invention; Figure 3 This is a schematic diagram of step S12 in an embodiment of the present invention; Figure 4 This is a schematic diagram of the first extraction process in an embodiment of the present invention; Figure 5 This is a schematic diagram of the second extraction process in an embodiment of the present invention; Figure 6 This is a schematic diagram of the third extraction process in an embodiment of the present invention; Figure 7 This is a schematic diagram of step S3 in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0022] This invention includes: A method for intelligent intraoperative low-field magnetic resonance imaging navigation based on preoperative high-field data, such as... Figure 1 As shown, it includes: Step S1: Before the operation, high-field MRI images of the patient are acquired and segmented to obtain a three-dimensional segmentation mask corresponding to the lesion, and feature extraction is performed to obtain clinical features; Step S2: Adjust the intraoperative scanning sequence parameters according to clinical characteristics to obtain the intraoperative scanning protocol; Step S3: During the operation, perform a low-field magnetic resonance imaging (MRI) scan on the patient based on the intraoperative scanning protocol to generate intraoperative navigation images, and annotate lesion information in the intraoperative navigation images based on a three-dimensional segmentation mask.

[0023] Specifically, addressing the issue of insufficient precision in lesion images provided by existing ultra-low field systems during surgery, this embodiment introduces preoperatively acquired high-field MRI images as a reference. Based on the clinical features reflected in the high-field MRI images, the scanning sequence parameters of the ultra-low field system used during surgery are dynamically adjusted, achieving better imaging results. Simultaneously, based on a three-dimensional segmentation mask as prior information, lesion information is annotated on the intraoperative navigation image to achieve more refined segmentation of the lesion on the intraoperative navigation image, meeting surgical requirements.

[0024] Specifically, the aforementioned intraoperative low-field magnetic resonance intelligent navigation method is mainly configured as a software implementation in computer equipment, such as the workstation of a magnetic resonance system, to achieve control of the examination process, imaging, lesion segmentation, and annotation.

[0025] The aforementioned intraoperative low-field magnetic resonance intelligent navigation method mainly includes three steps: Before the surgery begins, high-field MRI images of the patient need to be acquired and segmented to obtain a three-dimensional segmentation mask corresponding to the lesion, and feature extraction is performed to obtain clinical features. High-field MRI images are acquired and reconstructed using existing high-field MRI systems, typically 3-4T MRI systems. To facilitate subsequent acquisition of additional image details from low-field / ultra-low-field MRI systems, the high-field MRI images generated during this examination essentially need to include image data from multiple sequences, such as T1W, T2W, and FLAIR, to achieve clear imaging of any type of lesion in brain tissue, such as the solid portion of a tumor, cystic / necrotic areas, and peritumoral edema.

[0026] The three-dimensional segmentation mask is used to provide additional image information during the operation to compensate for the low resolution of images acquired by the ultra-low field magnetic resonance system during the operation. Clinical features are used to indicate to physicians the possible symptoms a patient may have, such as cystic changes and size, and to provide a basis for the automated adjustment of parameters in intraoperative scanning protocols.

[0027] After completing the above preoperative examination process, the intraoperative scanning sequence parameters are automatically adjusted according to clinical characteristics to obtain the intraoperative scanning protocol. When the surgery actually begins, the low-field / ultra-low-field MRI equipment in the operating room will perform low-field MRI examinations on the patient based on the intraoperative scanning protocol. Since the intraoperative scanning protocol has been optimized based on clinical characteristics, it can achieve better imaging results for the features that need attention during the operation.

[0028] Furthermore, after performing a low-field MRI scan, intraoperative navigation images can be generated. Since the field strength of low-field / ultra-low-field MRI equipment is usually below 0.2T, the image resolution is significantly degraded compared to high-field MRI images. This degraded image resolution means that doctors need to spend more time identifying residual tumors and surrounding tissues during the intraoperative process. In addition, even if traditional artificial intelligence models are introduced to assist in image interpretation, problems such as decreased model annotation accuracy and incomplete annotation will occur.

[0029] To address this issue, this embodiment uses a 3D segmentation mask obtained during preoperative examination as a reference, registers it with images obtained during intraoperative examination, and allows the model to annotate intraoperative navigation images in low field by referring to the image features of the 3D segmentation mask acquired in high field, forming lesion information. This achieves better prompting for doctors and provides surgeons with objective, quantitative, and risk-aware secondary resection decision support.

[0030] In one embodiment, such as Figure 2 As shown, step S1 includes: Step S11: Acquire high-field MRI images of the patient, identify and segment the lesions in the high-field MRI images to obtain a three-dimensional pre-segmentation mask; Step S12: Perform morphological processing on the 3D pre-segmentation mask to obtain the 3D segmentation mask; Step S13: Extract clinical features from the 3D segmentation mask.

[0031] Specifically, in order to achieve better extraction of the lesion area and acquisition of clinical features, in this embodiment, high-field MRI images are first acquired from the patient. The high-field MRI images are multi-sequence MRI images, including T1W, T2W, FLAIR and other sequences.

[0032] For high-field MRI images, a corresponding deep learning model, such as the 3D U-Net network, is pre-trained to automatically perform fine segmentation of the tumor region and generate a three-dimensional pre-segmentation mask. For multiple sequences, multi-channel input can be performed during the input process, or multi-channel encoding can be performed first, followed by fusion and decoding of the multi-channel sequences to achieve segmentation of multiple sequences.

[0033] Based on this segmentation process, the final output 3D pre-segmentation mask has the following annotations: Pixel value identifier: 0: Normal brain tissue; 1: The solid portion of the tumor; 2: Cystic / necrotic areas; 3: Peritumoral edema area.

[0034] Considering the potential noise during model recognition, and the fact that normal lesions are usually continuously distributed, after obtaining a three-dimensional pre-segmentation mask through preliminary recognition, morphological operations are performed on it to obtain the actual regional distribution.

[0035] Finally, another model is used to identify the 3D pre-segmented mask and extract clinical features related to the doctor's surgery.

[0036] In one embodiment, in step S11, the high-field MRI image includes a three-dimensional MRI image formed by multiple scanning sequences; Step S11 uses a multimodal 3D convolutional neural network model to segment the lesion into regions, including: normal brain tissue, solid tumor parts, cystic / necrotic areas, and peritumoral edema areas.

[0037] Specifically, to achieve better recognition results, a multimodal 3D convolutional neural network model is pre-trained for brain tumors. Its base is implemented using nnU-Net, 3D U-Net++, or Transformer-based architecture. This model has been pre-trained on a labeled dataset containing ≥1000 brain tumors (glioma, metastatic tumor, meningioma, etc.).

[0038] For multiple sequences in the acquired high-field MRI images, the registered multi-sequence MRI images (T1WI, T2WI, FLAIR, etc.) are used as multi-channel tensors, enabling the model to identify multiple regions separately and form a three-dimensional pre-segmentation mask.

[0039] In one embodiment, such as Figure 3 As shown, step S12 includes: Step S121: Perform morphological closing operations on the 3D pre-segmented mask in different mask types to obtain multiple connected components; Step S122: Perform pixel thresholding on connected components to remove isolated noise; Step S123: For the remaining connected components, retain the largest connected region as the output of the 3D segmentation mask of the corresponding mask type.

[0040] Specifically, in order to achieve better noise reduction, in this embodiment, multiple mask types of channels are separated on the basis of the three-dimensional pre-segmentation mask, including the solid part of the tumor, the cystic / necrotic area and the peritumoral edema area.

[0041] Morphological closing operations are performed on the 3D pre-segmentation mask in different mask types to obtain multiple connected components. Then, isolated noise points with too few pixels or that do not meet the pixel threshold are removed. Finally, the largest connected region is retained from the remaining connected components as the 3D segmentation mask output for the corresponding mask type, thereby achieving a better denoising effect.

[0042] Based on this, clinical features can be extracted, including: Anatomical location (whether it is adjacent to the motor / language functional area); Internal structure (whether there is cystic degeneration, necrosis, or hemorrhage); Size and morphological complexity.

[0043] In one embodiment, such as Figure 4 As shown, the lesion features extracted in step S13 include anatomical location features. The first extraction process for extracting anatomical location features includes: Step A131: Register the 3D segmentation mask and map it onto the standard brain atlas; Step A132: Calculate the Euclidean distance between the tumor boundary and the functional area, and use a distance threshold to filter and obtain the neighboring functional areas; Step A133: Generate anatomical location features based on adjacent functional areas.

[0044] Specifically, in order to describe the anatomical location of the lesion, the preoperative T1WI images are first nonlinearly registered with standard brain atlases (such as MNI152 or JHU ICBM-DTI-81) to obtain a registration mapping matrix.

[0045] Subsequently, the 3D segmentation mask was registered and mapped onto a standard brain atlas, which is marked with multiple functional areas, such as Brodmann 4 area of ​​the motor cortex and Broca's area of ​​the language area. Then, the Euclidean distance between the tumor boundary and the functional area was calculated, and a distance threshold was used to filter out neighboring functional areas. If the minimum distance was ≤ 10 mm, it was determined to be a "neighboring functional area", and the specific name of the functional area was recorded to form an anatomical location feature.

[0046] In one embodiment, such as Figure 5 As shown, the lesion features extracted in step S13 include internal structural features; The second extraction process for extracting internal structural features includes: Step B131: Detect the signal intensity in the high-field MRI image; Step B132: Determine the specific internal structure according to the signal strength matching detection rules; Specific internal structures include cystic degeneration, necrosis, and hemorrhage; Step B133: Mark specific internal structures as internal structural features in the 3D segmentation mask.

[0047] Specifically, to achieve a better representation of the internal structure of the lesion, this embodiment first detects the signal intensity in the high-field MRI image, marks the signal intensity on the pixels of the three-dimensional segmentation mask, and then determines the specific internal structure according to the signal intensity matching detection rule. Specifically, this includes: Cystic degeneration detection: In T2WI, if the signal intensity of a certain area within the mask is greater than 90% of the mean value of cerebrospinal fluid, and it shows a low signal in FLAIR, it is marked as "cystic degeneration"; Necrosis detection: If the signal is low in T1WI, high in T2WI, and without enhancement (if enhanced images are available), it is marked as "necrosis"; Hemorrhage detection: Focal high signal on T1WI (short T1 effect) and magnetic susceptibility artifacts on T2* or SWI sequences (if present) are marked as "hemorrhage".

[0048] Finally, specific internal structures are labeled in the 3D segmentation mask as internal structural features.

[0049] In one embodiment, such as Figure 6 As shown, the lesion features extracted in step S13 include volumetric and morphological features. The third extraction process for extracting volumetric morphological features includes: Step C131: Perform voxel quantification on the 3D segmentation mask to obtain the total volume; Step C132: Calculate the length of the major axis of the circumscribed ellipsoid of the three-dimensional segmentation mask as the maximum diameter of the lesion region; Step C133: Calculate the fractal dimension of the 3D segmentation mask; Volumetric morphological characteristics include total volume, maximum diameter of the lesion area, and fractal dimension.

[0050] Specifically, in order to achieve better quantitative results, in this embodiment, the number of non-zero voxels in the three-dimensional segmentation mask is first counted by multiplying the voxel volume (mm³) and then converted to cm³ to obtain the total volume.

[0051] Then, the length of the major axis of the circumscribed ellipsoid of the three-dimensional segmentation mask is calculated as the maximum diameter of the lesion region.

[0052] Finally, quantification is achieved by calculating the fractal dimension or boundary gradient entropy of the mask surface—the higher the value, the more irregular the boundary.

[0053] All the above steps are completed automatically by the AI ​​model without human intervention, ensuring a highly efficient and standardized process. Furthermore, the AI ​​model supports continuous learning: after each surgery, the segmentation results confirmed by the doctor can be fed back into the training set for model fine-tuning; the system has uncertainty quantification capabilities: if the model's segmentation confidence for a certain region is <0.7, it is marked as a "low-confidence region" in the report, prompting intraoperative attention; and it supports multi-center data standardization: through domain adaptation technology, it adapts to MRI data from different manufacturers and with different scanning protocols.

[0054] In one embodiment, step S2, the method for adjusting the intraoperative scanning sequence parameters includes: When the lesion is adjacent to a brain functional area, increase the T2WI weight in the intraoperative scanning protocol and shorten the TR / TE. When the lesion contains cystic changes, the FLAIR sequence is added to the intraoperative scanning protocol; When the lesion volume is smaller than the preset volume, the intraoperative scanning protocol is adjusted to prioritize high-resolution acquisition.

[0055] After extracting the aforementioned clinical features, the intraoperative scanning sequence parameters can be effectively adjusted through rule matching. The system adjusts the scanning sequence parameters based on the tumor features extracted by the AI ​​model and dynamically generates an intraoperative low-field MRI scanning protocol for imaging.

[0056] Generally, for brain surgery, after anesthesia, the patient's head is fixed using a standard neurosurgical head frame (such as a standard Mayfield or similar three-screw neurosurgical head frame). Using a readily available, modular radiofrequency coil (the upper part of which is removable), the surgeon performs initial tumor resection under open visualization. The resection process can be combined with neurophysiological monitoring (such as motor evoked potentials (MEP) and somatosensory evoked potentials (SSEP)) to protect functional areas.

[0057] During image reconstruction, the electromagnetic interference cancellation module built into the existing ultra-low field MRI system (such as electromagnetic interference cancellation technology based on sensing coils and deep learning) is invoked to suppress environmental electromagnetic noise and generate clinically usable images.

[0058] In one embodiment, such as Figure 7 As shown, step S3 includes: Step S31: Acquire intraoperative navigation images for the patient, register the intraoperative navigation images with high-field MRI images, and then transfer the three-dimensional segmentation mask onto the intraoperative navigation images to form a mapping mask; Step S32: Based on the mapping mask as the initial condition, perform fine segmentation on the intraoperative navigation image to obtain the intraoperative segmentation map; Step S32 includes: The mapping mask is used as a spatial prior channel and is input into the bi-branch code along with the intraoperative segmentation map for fusion to achieve segmentation.

[0059] Step S33: Quantify the intraoperative segmentation map to obtain lesion information.

[0060] Specifically, to achieve better intraoperative imaging results, in this embodiment, intraoperative navigation images are first acquired from the patient, and then the intraoperative navigation images are registered with high-field MRI images. The image registration adopts a multimodal non-rigid registration algorithm (such as Advanced Normalization Tools, ANTs), with the preoperative high-field T1WI as the fixed image and the intraoperative low-field T1WI / T2WI as the floating image. The registration objective function combines mutual information and deformation field smoothing constraints to avoid excessive distortion.

[0061] Based on the above process, a registration mapping matrix is ​​obtained, and then the 3D segmentation mask is transferred onto the intraoperative navigation image to form a mapping mask.

[0062] Subsequently, based on the mapping mask as the initial condition, and combined with the local texture and contrast of the intraoperative images, the residual lesions were finely segmented to obtain the intraoperative segmentation map. Specifically, a conditional segmentation network is constructed, whose inputs include intraoperative multi-contrast images (T1WI, T2WI, FLAIR) and mapped preoperative segmentation masks (as spatial prior channels). The network structure can adopt a dual-encoder U-Net: one encoder processes the intraoperative images, and the other encoder processes the prior mask, with features fused at the bottleneck layer. The output is a refined probability map of intraoperative residual lesions, which is binarized by a threshold (e.g., 0.5) to generate the final segmentation result.

[0063] Finally, the system outputs a highlighted image, the volume of residual lesions, and combines preoperative features to generate risk warnings and annotate lesion information.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent intraoperative low-field magnetic resonance imaging navigation based on preoperative high-field data guidance, characterized in that, include: Step S1: Before the operation, high-field MRI images of the patient are acquired and segmented to obtain a three-dimensional segmentation mask corresponding to the lesion, and feature extraction is performed to obtain clinical features; Step S2: Adjust the intraoperative scanning sequence parameters according to the clinical characteristics to obtain the intraoperative scanning protocol; Step S3: During the operation, a low-field magnetic resonance imaging examination is performed on the patient based on the intraoperative scanning protocol to generate an intraoperative navigation image, and lesion information is marked in the intraoperative navigation image based on the three-dimensional segmentation mask.

2. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 1, characterized in that, In step S2, the method for adjusting the intraoperative scanning sequence parameters includes: When the lesion is adjacent to a brain functional area, T2WI weighting is increased in the intraoperative scanning protocol to shorten the TR / TE. When the lesion contains cystic changes, a FLAIR sequence is added to the intraoperative scanning protocol; When the volume of the lesion is smaller than the preset volume, the intraoperative scanning protocol is adjusted to prioritize high-resolution acquisition.

3. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 1, characterized in that, Step S1 includes: Step S11: Acquire the high-field MRI image of the patient, and identify and segment the lesion in the high-field MRI image to obtain a three-dimensional pre-segmentation mask; Step S12: Perform morphological processing on the three-dimensional pre-segmentation mask to obtain the three-dimensional segmentation mask; Step S13: Extract the clinical features from the three-dimensional segmentation mask.

4. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 3, characterized in that, In step S11, the high-field MRI image includes a three-dimensional MRI image formed by multiple scanning sequences; Step S11 uses a multimodal 3D convolutional neural network model to segment the lesion into regions, including: normal brain tissue, solid tumor portion, cystic / necrotic region, and peritumoral edema region.

5. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 3, characterized in that, Step S12 includes: Step S121: Perform morphological closing operations on the three-dimensional pre-segmented mask in different mask types to obtain multiple connected components; Step S122: Perform pixel threshold filtering on the connected components to remove isolated noise; Step S123: For the remaining connected components, retain the largest connected region as the output of the three-dimensional segmentation mask of the corresponding mask type.

6. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 4, characterized in that, The lesion features extracted in step S13 include anatomical location features; The first extraction process for extracting the anatomical location features includes: Step A131: Register the three-dimensional segmentation mask and map it onto a standard brain atlas; Step A132: Calculate the Euclidean distance between the tumor boundary and the functional area, and use a distance threshold to filter and obtain the neighboring functional areas; Step A133: Generate the anatomical location features according to the adjacent functional areas.

7. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 4, characterized in that, The lesion features extracted in step S13 include internal structural features; The second extraction process for extracting the internal structural features includes: Step B131: Detect the signal intensity in the high-field MRI image; Step B132: Determine the specific internal structure according to the signal strength matching detection rule; The specific internal structures include cystic degeneration, necrosis, and hemorrhage; Step B133: Mark the specific internal structure as the internal structure feature in the three-dimensional segmentation mask.

8. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 4, characterized in that, The lesion features extracted in step S13 include volumetric and morphological features; The third extraction process for extracting the volumetric morphological features includes: Step C131: Perform voxel quantification on the three-dimensional segmentation mask to obtain the total volume; Step C132: Calculate the length of the major axis of the circumscribed ellipsoid of the three-dimensional segmentation mask as the maximum diameter of the lesion region; Step C133: Calculate the fractal dimension of the three-dimensional segmentation mask; The volumetric morphological features include the total volume, the maximum diameter of the lesion region, and the fractal dimension.

9. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 1, characterized in that, Step S3 includes: Step S31: Acquire the intraoperative navigation image for the patient, register the intraoperative navigation image with the high-field MRI image, and then transfer the three-dimensional segmentation mask onto the intraoperative navigation image to form a mapping mask; Step S32: Based on the mapping mask as the initial condition, perform fine segmentation on the intraoperative navigation image to obtain an intraoperative segmentation map; Step S33: Quantitatively analyze the intraoperative segmentation map to obtain the lesion information.

10. The intraoperative low-field magnetic resonance intelligent navigation method according to claim 9, characterized in that, Step S32 includes: The mapping mask is used as a spatial prior channel and is input into a dual-branch code along with the intraoperative segmentation map for fusion to achieve segmentation.

Images