Artificial intelligence assisted intraoperative imaging method and system and storage medium

Abstract

The invention relates to the technical field of medical image processing, in particular to an artificial intelligence assisted intraoperative imaging method, which comprises the following steps: S1, preprocessing a multi-modal medical image, segmenting and recognizing an anatomical structure by a deep learning model according to the preprocessed image, measuring anatomical parameters based on a segmentation and recognition result, and generating an operation planning path by artificial intelligence according to the anatomical parameters; s2, collecting a C-shaped arm perspective image stream in real time, dynamically tracking space coordinates of a surgical instrument, comparing the position of the instrument with a surgical planned path, calculating offset, and when the offset is greater than an offset threshold, outputting correction guidance through an AR superposition layer; and S3, monitoring an image quality index in real time, dynamically adjusting exposure parameters through a reinforcement learning model, and when a metal implant is detected, switching a dual-energy-spectrum mode and executing an artifact suppression algorithm. According to the method, preoperative precise planning and intraoperative assistance are realized through artificial intelligence, the problems of poor image quality and high radiation risk are solved through technical optimization, and the method has important clinical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to an intraoperative imaging method and system assisted by artificial intelligence and a storage medium. BACKGROUND

[0002] As the core equipment for real-time navigation in orthopedic, cardiovascular and other surgeries, the mobile C-arm X-ray machine has provided important support for precise surgery since it was put into clinical use.

[0003] The common C-arm intraoperative imaging method in the market still has some obvious defects in actual application, which affects the surgery efficiency, precision and safety to some extent, as follows: Lack of AI assistance, relying on manual judgment The existing equipment mainly relies on manual judgment of the operator in image analysis, and cannot automatically identify and label the patient's anatomical structure, which means that the operator needs to extract key information from the image based on his own experience, not only increasing the operator's work burden, but also affecting the accuracy of judgment. At the same time, the device has no AI assisted analysis function, and cannot give surgery path suggestions based on the image, thereby reducing the operation efficiency, especially in complex surgery, the operator may need to adjust the perspective angle repeatedly, analyze the image multiple times, and prolong the surgery time; Single image guidance function, no personalized support The image guidance function of the traditional C-arm is relatively fixed and cannot be dynamically adapted to the individual differences of patients or the operation habits of operators. For example, the bone morphology and blood vessel distribution of different patients are different, and the operation styles of different operators (such as perspective angle preference and key structure focus) are also different, but the device cannot provide targeted feedback by learning these information. Therefore, the operator is difficult to obtain personalized image prompts that fit his own habits or the specific situation of the patient, and the practicality of the guidance is limited; Metal artifacts interfere with image quality In orthopedic surgery, metal implants such as steel plates and screws are often implanted, and the existing device has weak suppression ability for metal artifacts. Metal can produce strong reflection or absorption of X-rays, resulting in uneven light and dark artifacts, blurred edges or structure overlap in the perspective image, which may cover key information such as bone healing, implant and surrounding tissue fitment, etc. This will increase the difficulty of the operator to observe the key structure, and may even misjudge the position of the implant due to the blurred image, affecting the surgery effect; Radiation control is lagging behind, with potential safety risks Existing equipment often uses fixed radiation parameters (such as exposure dose and tube voltage) during intraoperative exposure, which cannot be adjusted according to real-time imaging needs. For example, when observing fine structures (such as small screws), fixed parameters may result in insufficient radiation dose and inadequate image clarity; while when observing large structures, excessive dose may cause unnecessary radiation. This control method not only affects the stability of image quality but also exposes patients and medical staff to potential radiation risks. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an artificial intelligence-assisted intraoperative imaging method, comprising the following steps: S1: Preprocessing multimodal medical images to obtain preprocessed images; deep learning model segmentation and identification of anatomical structures based on the preprocessed images; measurement of anatomical parameters including pedicle diameter, angle, and spatial distance based on the segmentation and identification results; and artificial intelligence generation of surgical planning path based on the anatomical parameters. S2: Real-time acquisition of C-arm fluoroscopic image stream, dynamic tracking of surgical instrument spatial coordinates, comparison of instrument position with the surgical planning path, calculation of offset, and when the offset is greater than the offset threshold, output of correction guidance through AR overlay layer; S3: Monitor the image quality index in real time, and dynamically adjust the exposure parameters through a reinforcement learning model to make the image quality index greater than or equal to the quality threshold and minimize the radiation dose. When a metal implant is detected, switch to dual-spectrum mode and execute an artifact suppression algorithm to achieve dynamic optimization.

[0005] Preferably, in step S1, preprocessing the multimodal medical images to obtain preprocessed images, and the deep learning model segmenting and recognizing anatomical structures based on the preprocessed images, further includes: Multimodal medical images, including CT image sequences, T1-weighted MRI images, and X-ray anteroposterior and lateral views, are acquired. The multimodal medical images are registered to the same three-dimensional coordinate system through affine transformation to obtain a registered image. The registered image is then subjected to isotropic resampling and window width and window level standardization to obtain the preprocessed image. The deep learning model extracts multi-scale feature maps through the encoder, and the decoder performs feature fusion based on the multi-scale feature maps to obtain a feature fusion map. It also dynamically adjusts spatial and channel spatial attention through attention gating to highlight key point features. The output layer outputs the anatomical structures, including bones, blood vessels, nerves, and lesions, based on the feature fusion map.

[0006] Preferably, in step S1, the anatomical parameters, including geometric dimensions, spatial relationships, and density / signal features, are measured based on the segmentation and recognition results, further including: Based on the segmentation and recognition results, geometric dimensions including length, angle, area, and volume are calculated. Among them, the closed contour of the inner boundary of the bone cortex is extracted based on the segmentation and recognition results, and the maximum inscribed circle of the closed contour of the inner boundary of the bone cortex is calculated. The pedicle diameter is the diameter of the safety passage. Based on the segmentation and recognition results, spatial relationships including distance, angle, and relative position are calculated. A three-dimensional coordinate system is established with the center of the vertebra as the origin, the sagittal plane as the XZ plane, and the transverse plane as the XY plane. The pedicle axis vector is calculated, and the pedicle angle, including the sagittal plane tilt angle and the transverse plane inclination angle, is calculated based on the pedicle axis vector. The straight line for the pedicle screw planning path is obtained. The neural structure contour point set is extracted based on the segmentation and recognition results, and the minimum Euclidean distance from each neural structure contour point to the screw path straight line is calculated. The average grayscale value / CT value of a specific region is calculated based on the segmentation and recognition results.

[0007] Preferably, in step S1, generating a surgical planning path based on the anatomical parameters further includes: Load a guide library including anatomical constraints, instrument rules, and path templates, wherein the anatomical constraints include the minimum diameter of the bony channel and the threshold for the nerve safety distance, the instrument rules include the implant size calculation formula, and the path templates include the standard pin insertion point coordinate range; The optimal implant size is calculated based on the implant size calculation formula and the anatomical parameters. The guide library and the anatomical parameters are then verified according to the verification conditions to obtain the verification results. Artificial intelligence obtains the surgical planning path based on the anatomical structure, the anatomical parameters, the guide library, and the optimal implant size; The verification conditions include: If the pedicle diameter is greater than or equal to the minimum diameter of the bony canal, the screw diameter is large enough to pass through the bony canal. If the pedicle diameter is less than the minimum diameter of the bony canal, the doctor is prompted to adjust the screw diameter or replan the path to obtain a safe pedicle diameter.

[0008] Preferably, in step S2, the C-arm fluoroscopic image stream is acquired in real time, the spatial coordinates of the surgical instruments are dynamically tracked, the instrument positions are compared with the surgical planning path, and the offset is calculated. This further includes: Real-time acquisition of C-arm fluoroscopic image stream, and acquisition of the three-dimensional coordinates of the surgical instrument tip, i.e. the current position of the instrument, through an optical tracking system; The three-dimensional coordinates are mapped to the C-arm image coordinate system to generate projected coordinates; Calculate the offset of the projected coordinates from the surgical planning path.

[0009] Preferably, in step S2, when the offset is greater than the offset threshold, a correction guide is output through the AR overlay layer, further including: AR guidance including directional arrows, distance scales, and safety passages is superimposed on real-time C-arm fluoroscopic images. The directional arrows are drawn in a conical shape along the offset direction, starting from the current position of the instrument. The arrow length is obtained according to the offset. The distance scale is displayed according to the offset. A semi-transparent green strip area is rendered along the surgical planning path. The width of the safety passage is set according to the offset threshold. When the offset exceeds the offset threshold, the degree of deviation is indicated by changing the arrow, and a prompt is given through vibration and voice.

[0010] Preferably, in step S3, monitoring the image quality index and dynamically adjusting the exposure parameters through a reinforcement learning model to ensure that the image quality index is greater than or equal to a quality threshold and the radiation dose is minimized, further includes: Real-time calculation of image quality parameter IQI on C-arm fluoroscopic images; The current state, including the image quality parameters, cumulative dose, tissue thickness, and operation stage, is input into the reinforcement learning model. The reinforcement learning model outputs an action to adjust the exposure parameters, and is trained and optimized by maximizing the reward function. When a sensitive organ is detected, a penalty term is added to the maximum reward function. In the fetal region, a strict mode is enabled to limit the exposure parameters. The diaphragm movement speed is calculated by optical flow, and a motion blur penalty term is added to the reward function.

[0011] Preferably, in step S3, when a metal implant is detected, switching to the dual-spectral mode and executing an artifact suppression algorithm further includes: Real-time analysis of C-arm fluoroscopic images to calculate metal probability maps for identifying metal regions: When the area ratio of the metal region is greater than the metal region threshold, the dual-energy spectrum mode is triggered; After triggering, the X-ray source is controlled to alternately emit low-energy and high-energy spectrum data over a period of time, while simultaneously acquiring dual-energy projection data. The dual-energy projection data is decomposed into a base material density map. Metal trajectory regions are extracted from the base material density map, and contamination projection data is repaired using neighborhood-weighted interpolation to obtain a metal projection edge map. Input contamination projection data and the metal projection edge map into the U-Net++ network, and output an artifact-suppressed image; The final image is obtained based on the base material density map, the artifact suppression image, and the low-frequency and high-frequency components.

[0012] Based on the same concept, the present invention also provides an artificial intelligence-assisted intraoperative imaging system, comprising: The preoperative planning module preprocesses multimodal medical images to obtain preprocessed images. A deep learning model segments and identifies anatomical structures based on the preprocessed images. Based on the segmentation and identification results, it measures anatomical parameters including pedicle diameter, angle, and spatial distance. Based on the anatomical parameters, it generates a surgical planning path. The intraoperative navigation module acquires C-arm fluoroscopic images in real time, dynamically tracks the spatial coordinates of surgical instruments, compares the instrument positions with the planned surgical path, calculates the offset, and outputs correction guidance through AR overlay when the offset exceeds the offset threshold. The imaging optimization module monitors the image quality index and dynamically adjusts the exposure parameters through a reinforcement learning model to ensure that the image quality index is greater than or equal to the quality threshold and the radiation dose is minimized. When a metal implant is detected, the dual-spectrum mode is switched and an artifact suppression algorithm is executed.

[0013] Based on the same concept, the present invention also provides a computer-readable storage medium storing computer code, which, when executed, performs the steps of the single-sensor-based adaptive control method for a water pump as described in any one of the embodiments.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes a deep learning model to segment and identify anatomical structures from preprocessed images. Based on the segmentation and identification results, it measures anatomical parameters including pedicle diameter, angle, and spatial distance. This enables rapid identification of key anatomical structures in preoperative imaging data such as CT and X-rays without relying on manual analysis by the surgeon, and automatically completes the measurement and annotation. Furthermore, artificial intelligence generates surgical planning paths based on these anatomical parameters, creating personalized surgical guidance tailored to the individual anatomical characteristics of each patient. This function directly addresses the problems of low efficiency and experience-dependent accuracy in manual judgment, providing surgeons with objective and accurate preoperative references and reducing decision-making biases caused by individual experience differences. This invention acquires C-arm fluoroscopic image streams in real time, dynamically tracks the spatial coordinates of surgical instruments, compares the instrument positions with the planned surgical path, calculates the offset, and outputs correction guidance through an AR overlay layer when the offset exceeds an offset threshold, thereby improving the practicality of the guidance and operational efficiency.

[0015] This invention monitors the image quality index in real time and dynamically adjusts exposure parameters using a reinforcement learning model to ensure the image quality index is greater than or equal to a quality threshold while minimizing radiation dose. When a metal implant is detected, a dual-energy spectral mode is switched and an artifact suppression algorithm is executed. By integrating dual-energy spectral imaging technology with an intelligent metal artifact removal scheme, this invention effectively solves the problem of image blurring caused by metal implants in orthopedic surgery, ensuring that surgeons can clearly observe key information such as implant position and fit, reducing the risk of misjudgment due to insufficient image quality. By dynamically adjusting exposure parameters through a reinforcement learning model, the radiation exposure risk for patients and medical staff is significantly reduced while ensuring image quality, thus improving surgical safety. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0017] Figure 1 This is a flowchart of an artificial intelligence-assisted intraoperative imaging method according to the present invention; Figure 2 This is a structural diagram of an artificial intelligence-assisted intraoperative imaging device according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. Obviously, the described embodiments are only some, not all, of the embodiments described in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application.

[0019] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a” and “an” used herein, and “the”, may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0020] First Embodiment Please see Figure 1 As shown in the figure, this embodiment provides an artificial intelligence-assisted intraoperative imaging method, which includes the following steps: S1: Preprocessing multimodal medical images yields preprocessed images. A deep learning model segments and identifies anatomical structures based on the preprocessed images. Based on the segmentation and identification results, anatomical parameters, including pedicle diameter, angle, and spatial distance, are measured. Artificial intelligence generates surgical planning paths based on the anatomical parameters. Specifically, in this embodiment, the deep learning model employs advanced convolutional neural networks, particularly U-Net or its variants (such as nnU-Net and 3DU-Net). These models excel in the field of medical image segmentation. The model is trained and validated using a massive amount of labeled medical image datasets (containing precise outlines of various anatomical structures, lesions, and implants). During the identification process, the model analyzes the image layer by layer (for 3DCT / MRI) or the entire image (for 2DX light), automatically outputting pixel-level / voxel-level labels to distinguish different anatomical structures (such as "L4 vertebral body", "femoral artery", and "tumor tissue"), while also identifying key points (landmarks).

[0021] Preferably, in step S1, preprocessing the multimodal medical images to obtain preprocessed images, and the deep learning model segmenting and identifying anatomical structures based on the preprocessed images, further includes: Multimodal medical images, including CT image sequences, T1-weighted MRI images, and X-ray anteroposterior and lateral views, are acquired. The multimodal medical images are registered to the same three-dimensional coordinate system through affine transformation to obtain registered images. The registered images are then subjected to isotropic resampling and window width and level standardization to obtain preprocessed images. Specifically, in this embodiment, CT is used as the core input, and the CT images are preprocessed including denoising, standardization (intensity normalization), spatial alignment (registering images of different sequences / modalities), and possible segmentation (preliminarily separating the target region from the background). The deep learning model extracts multi-scale feature maps through the encoder, and the decoder performs feature fusion based on the multi-scale feature maps to obtain a feature fusion map. It also dynamically adjusts the spatial and channel spatial attention through attention gating to highlight key point features. Specifically, in this embodiment, an attention module is introduced into the encoder to enhance the ability to sense multi-scale features. The decoder restores the resolution of the multi-scale image through upsampling. At each decoding level, the corresponding feature map from the encoder is fused with the current decoder feature map. Before fusion, the spatial and channel weights between different feature maps are automatically learned through the attention gating module, and the fusion weight is calculated. This can significantly improve the model's attention to the target region, especially when dealing with complex structures (such as bones, nerves, blood vessels, and lesions). At each layer of the decoder, the feature map after attention gating is concatenated with the upsampled feature map to form a richer feature representation. The output layer outputs anatomical structures, including bones, blood vessels, nerves, and lesions, based on the feature fusion map. Specifically, in this embodiment, the skeletal structures include: vertebral bodies, pedicles, intervertebral discs, articular surfaces (such as the acetabulum and femoral head), long bones (such as the femur and tibia) and their key anatomical landmarks (such as the greater trochanter and anterior superior iliac spine), and fracture lines; vascular networks: major arteries and veins and their branches (especially in interventional procedures or operations near vascular areas); neural structures: major nerve bundles (identified when image resolution allows and is critical to surgical safety, such as nerve roots in spinal surgery); organ contours: relevant organ boundaries in specific surgeries (such as the kidneys and ureters in urinary tract stone surgery); implants / devices (if present): existing internal fixation devices (screws, plates, artificial joints), catheters, guidewires, etc.; lesion / target area: Tumors, cysts, stones, vascular malformations, and fracture fragments requiring reduction or fixation are identified by outlining the boundaries of bones, blood vessels, nerves, lesions, etc. with different colors / line types; important anatomical landmarks (such as pedicle entry points and joint rotation centers) are highlighted; measurement results (such as "length: 35.2mm", "angle: 28°") are displayed directly at the corresponding positions in the image; and structures or highly variable areas adjacent to important neurovascular structures are marked conspicuously (such as flashing red outlines or exclamation mark icons).

[0022] Preferably, in step S1, the anatomical parameters, including geometric dimensions, spatial relationships, and density / signal features, are measured based on the segmentation and recognition results, further including: Based on the segmentation and recognition results, geometric dimensions including length, angle, area, and volume are calculated. Specifically, the closed contour of the inner boundary of the bone cortex is extracted based on the segmentation and recognition results, and the maximum inscribed circle of the closed contour is calculated. The pedicle diameter is the safety passage diameter. In this embodiment, computer vision and geometric algorithms are applied to calculate length (e.g., pedicle diameter, distance between fracture ends), angle (e.g., Cobb angle, neck-shaft angle, acetabular anteversion / abduction angle), area (e.g., intervertebral disc degeneration area), and volume (e.g., tumor volume). Based on the segmentation and recognition results, spatial relationships including distance, angle, and relative position are calculated. A three-dimensional coordinate system is established with the center of the vertebral body as the origin, the sagittal plane as the XZ plane, and the transverse plane as the XY plane. The pedicle axis vector is calculated, and the pedicle angle, including the sagittal plane inclination angle and the transverse plane inclination angle, is calculated based on the pedicle axis vector. The straight line for the planned pedicle screw path is obtained. A set of neural structure contour points is extracted based on the segmentation and recognition results, and the minimum Euclidean distance from each neural structure contour point to the screw path straight line is calculated. Specifically, in this embodiment, the relative positions between different structures (distance from a point to a line, angle between surfaces) are calculated. Distances include the distance from key nerves and blood vessels to the predetermined screw implantation path, displacement between fracture fragments, etc.; angles include the scoliosis angle, relative angles of articular surfaces, etc.; and relative positions include the coordinates of the lesion center relative to bony landmarks, etc. Based on the segmentation and recognition results, the average gray value / CT value of a specific region is calculated. Specifically, in this embodiment, the average density / signal intensity within a specific region of interest (ROI) is calculated. The average gray value / CT value can be used to assist in determining bone density and lesion nature.

[0023] Preferably, in step S1, generating the surgical planning path based on anatomical parameters further includes: The system loads a guideline library that includes anatomical constraints, instrument rules, and path templates. Anatomical constraints include the minimum diameter of bony channels and the threshold for nerve safety distance. Instrument rules include implant size calculation formulas, and path templates include the coordinate range of standard screw insertion points. Specifically, in this embodiment, digital standard surgical guidelines (such as pedicle screw placement specifications and joint replacement positioning parameters in orthopedics) are invoked. These guidelines include safety thresholds (such as nerve distance > 2mm), standard operating procedures (such as the screw insertion angle range), and implant selection rules. In addition to the rule-layer data, which includes anatomical constraints, instrument rules, and path templates, the guideline library also includes case-layer data, including the surgeon's past surgical images, instrument movement trajectories, operation time nodes, and intraoperative decision records. The case-layer data is associated with desensitized time-series data using the surgeon's ID. Hidden Markov Models (HMMs) or LSTM neural networks are used to obtain the surgeon's operating habits (instrument type / angle preference / action speed), risk propensity (safety distance threshold / radiation sensitivity), efficiency mode (stage time), etc., to generate a surgeon's behavioral profile. Based on the anatomical structure and anatomical parameters, the current scenario is compared with similar historical cases, and the corresponding behavioral profile is loaded. The optimal implant size is calculated based on the implant size calculation formula and anatomical parameters. The guide library and anatomical parameters are then verified according to the verification conditions to obtain the verification results. Specifically, in this embodiment, the implant size is screw length / diameter, artificial joint model, etc. Artificial intelligence obtains surgical planning paths based on the anatomical structures, anatomical parameters, the guide library, optimization targets, and the optimal implant size. Specifically, in this embodiment, anatomical parameters and structures are overlaid onto the original or reconstructed images according to preset rules and visualization schemes. Risk labeling is triggered based on the spatial relationships between anatomical structures. Based on the anatomical structures, anatomical parameters, and safe operating paths based on the standard guide library (such as the ideal entry point and direction for pedicle screws), and the target resection area, recommended implant specifications are displayed at appropriate locations to generate a preliminary plan. The preliminary plan includes calculating the optimal implant size (outputting the matching implant type, size, and quantity (e.g., "L4 pedicle screw: diameter 6.5mm × length 45mm")), dynamically displaying the guiding path on the 3D reconstructed image (e.g., green dashed arrows), and annotating the coordinates of key nodes (XYZ coordinates of the entry point and the target point) and angle parameters (sagittal tilt angle 15°±3°, transverse inclination angle 10°±2°) to generate safe operating path suggestions (e.g., screw channels, puncture paths). The system automatically generates alternative routes and contingency plans (e.g., "Alternative route B: increase the lateral inclination angle by 5°") by defining target areas (such as tumor resection range, vertebral artery malformation, etc.) and presenting fully annotated images, detailed measurement reports, and preliminary surgical plans on the preoperative planning workstation or intraoperative touch display interface. Surgeons are allowed to review, modify, confirm, or reject the AI's annotation, measurement, and planning suggestions (with user-friendly interactive tools for fine-tuning). For highly variable areas (such as vertebral artery malformation), alternative routes and contingency plans (e.g., "Alternative route B: increase the lateral inclination angle by 5°") are automatically generated. Confirmed plans are saved and input into the "intraoperative assistance module" for real-time guidance. Based on the annotated neurovascular locations, a three-dimensional safe / no-go zone model is constructed. Potential field algorithms or artificial intelligence path search algorithms are used to generate the shortest operating path (e.g., screw channel, puncture trajectory) that avoids no-go zones, while simultaneously satisfying: minimum invasiveness (shortest path length), maximum safety (maximum distance from risk structures), and optimal mechanical stability (e.g., screw axis coincides with the area with the highest bone density). The verification conditions include: If the pedicle diameter is greater than or equal to the minimum diameter of the bony canal, the screw diameter is large enough to pass through the bony canal. If the pedicle diameter is less than the minimum diameter of the bony canal, the physician is prompted to adjust the screw diameter or replan the path to obtain a safe pedicle diameter. Specifically, in this city's implementation, the patient's measurement data (e.g., pedicle diameter 8.5mm) is matched with the safety conditions in the guidelines (screw diameter ≤ 80% of pedicle diameter), and compliant implant specifications (e.g., recommended screw diameter ≤ 6.8mm) are automatically selected. The path avoidance strategy is adjusted according to anatomical variations (e.g., vascular tortuosity), and the standard path template in the guidelines is dynamically corrected.

[0024] S2: Real-time acquisition of C-arm fluoroscopic image stream, dynamic tracking of surgical instrument spatial coordinates, comparison of instrument position with surgical planning path, calculation of offset, and when the offset is greater than the offset threshold, output of correction guidance through AR overlay layer.

[0025] Preferably, in step S2, the C-arm fluoroscopic image stream is acquired in real time, the spatial coordinates of the surgical instruments are dynamically tracked, the instrument positions are compared with the planned surgical path, and the offset is calculated. This further includes: Real-time acquisition of C-arm fluoroscopic image stream, and acquisition of the three-dimensional coordinates of the surgical instrument tip, i.e. the current position of the instrument, through an optical tracking system; The three-dimensional coordinates are mapped to the C-arm imaging coordinate system to generate projected coordinates. Specifically, in this embodiment, the spatial coordinate system of the planned path is registered with the real-time intraoperative C-arm imaging coordinate system (based on bony landmark matching), and the planned path (semi-transparent colored guide line) and real-time instrument position (such as red dot marking at the tip of the drill) are superimposed on the intraoperative fluoroscopic image. Calculate the offset of the projected coordinates from the surgical planning path.

[0026] Preferably, in step S2, when the offset is greater than the offset threshold, a correction guide is output through the AR overlay layer, further including: AR guidance including directional arrows, distance scales, and safety passages is superimposed on real-time C-arm fluoroscopic images. The directional arrows are drawn as a cone-shaped arrow starting from the current position of the instrument and along the offset direction. The arrow length is obtained according to the offset. The distance scale is displayed according to the offset. A semi-transparent green strip area is rendered along the surgical planning path. The width of the safety passage is set according to the offset threshold. When the offset is greater than the offset threshold, the degree of deviation is indicated by changing the arrow, and a prompt is given through vibration and voice. Specifically, in this embodiment, if the device deviates from the path by more than 2mm, an audible alarm is triggered and the deviation area is marked in red on the interface.

[0027] S3: Real-time monitoring of image quality index, dynamic adjustment of exposure parameters through reinforcement learning model to ensure that image quality index is greater than or equal to quality threshold and radiation dose is minimized. When a metal implant is detected, the dual-spectrum mode is switched and an artifact suppression algorithm is executed to achieve dynamic optimization. Specifically, in this embodiment, based on real-time image feedback and intraoperative dynamic monitoring, X-ray emission parameters (such as voltage, current, and exposure time) are automatically adjusted to minimize radiation dose exposure for patients and operators while ensuring image quality.

[0028] Preferably, in step S3, monitoring the image quality index and dynamically adjusting exposure parameters through a reinforcement learning model to ensure the image quality index is greater than or equal to a quality threshold and the radiation dose is minimized, further includes: The image quality parameter IQI on C-arm fluoroscopic images is calculated in real time. The formula for calculating the image quality parameter is as follows: in, As the first parameter, For the second parameter, As the third parameter, For signal-to-noise ratio, To compare resolutions, Weights for artifact scoring; The current state, including image quality parameters, cumulative dose, tissue thickness, and operation stage, is input into the reinforcement learning model. Specifically, in this embodiment, the current state includes image feedback, patient physiology, surgical scenario, and radiation dose. Image feedback includes signal-to-noise ratio (SNR), contrast resolution, and artifact score; patient physiology includes changes in tissue thickness, bone density distribution, and bleeding volume; surgical scenario includes operation stage, instrument position, and image quality requirements; and radiation dose includes cumulative dose, real-time dose rate, and organ exposure warning. The reinforcement learning model outputs actions to adjust exposure parameters, and is trained and optimized by maximizing the reward function. When a sensitive organ is detected, a penalty term is added to the reward function. A strict mode is enabled in the fetal region to limit exposure parameters. The diaphragm movement velocity is calculated using optical flow, and a motion blur penalty term is added to the reward function. Specifically, in this embodiment, the output action constraints are: 50kV≤kV≤120kV, 0.5mA≤mA≤5mA, t≤200ms. The reward function consists of three parts: quality reward, dose penalty, and stability penalty. , For quality awards, As a dose penalty, This is a stability penalty.

[0029] Preferably, in step S3, when a metal implant is detected, switching to the dual-spectral mode and executing an artifact suppression algorithm further includes: Real-time analysis of C-arm fluoroscopic images to calculate metal probability maps for identifying metal regions: When the area of ​​the metal region is greater than the metal region threshold, the dual-energy spectral mode is triggered. Specifically, in this embodiment, when the area of ​​the metal region is greater than 5%, the dual-energy spectral imaging mode is triggered. After triggering, the X-ray source is controlled to alternately emit low-energy spectrum (80kVp, 3.5mm Al filter) and high-energy spectrum data (140kVp, 0.2mm Cu filter) for a certain period of time, and dual-energy projection data is collected simultaneously for subsequent material decomposition and artifact elimination processing. Dual-energy projection data is decomposed into a base material density map. Metal trajectory regions are extracted based on the base material density map, and contaminated projection data is repaired using neighborhood weighted interpolation to obtain a metal projection edge map. The contaminated projection data and the metal projection edge map are input into the U-Net++ network, and the artifact suppression image is output. Specifically, in this embodiment, the L1+SSIM joint loss function is used for optimization. The final image is obtained based on the base material density map, artifact suppression image, and low-frequency and high-frequency components.

[0030] Please see Figure 2 As shown, this embodiment employs a mobile C-arm fluoroscopy device for real-time intraoperative applications to implement the above method, including: The C-arm includes a load C-ring 1, which has an equipment opening 4 with a maximum opening width of 99cm. An arc-shaped track is provided along the circumference of the C-ring with a radius of curvature R=650mm±10mm, which can be flexibly adjusted to meet the needs of multi-angle and multi-position imaging during surgery and provide the surgeon with ample operating space. The image acquisition device 2, located at the first end of the load C ring 1, includes a high-definition detector with a maximum imaging field of view of 43cm×43cm, ensuring a wide coverage of intraoperative images and clear detail. The X-ray emitting device 3, located at the second end of the load C ring 1, is coaxially aligned with the image acquisition device. One side of the load C-ring 1 and the compensation C-ring 6 are connected by the sliding arm 5; On the other side of the compensation ring C6, there is an equipment rack 7. The equipment rack 7 includes a movable base and an operating interface. The whole machine is equipped with an electric movable base, which can realize precise translation and rotation positioning, and is equipped with a touch display interface, which supports the surgeon to quickly switch imaging modes, image processing parameters and surgical guidance options, improving the convenience and efficiency of intraoperative operation.

[0031] Second Embodiment Based on the same concept, this embodiment provides an artificial intelligence-assisted intraoperative imaging system, including: The preoperative planning module preprocesses multimodal medical images to obtain preprocessed images. A deep learning model segments and identifies anatomical structures based on the preprocessed images. Based on the segmentation and identification results, it measures anatomical parameters including pedicle diameter, angle, and spatial distance. Based on the anatomical parameters, it generates a surgical planning path. The intraoperative navigation module acquires C-arm fluoroscopic images in real time, dynamically tracks the spatial coordinates of surgical instruments, compares the instrument positions with the planned surgical path, calculates the offset, and outputs correction guidance through AR overlay when the offset exceeds the offset threshold. The imaging optimization module monitors the image quality index and dynamically adjusts the exposure parameters through a reinforcement learning model to ensure that the image quality index is greater than or equal to the quality threshold and the radiation dose is minimized. When a metal implant is detected, the dual-spectrum mode is switched and an artifact suppression algorithm is executed.

[0032] Third Embodiment In this embodiment, a computer device is provided, including a memory and one or more processors. The memory stores computer code, and when the computer code is executed by one or more processors, it causes the one or more processors to perform the steps of the artificial intelligence-assisted intraoperative imaging method in the first embodiment.

[0033] In some embodiments of this application, a computer-readable storage medium is also provided, wherein the computer-readable instructions, when executed by one or more processors, cause one or more processors to perform the steps of the single-sensor-based adaptive control method for a water pump as described in any of the first embodiments.

[0034] It is understood that, for the aforementioned AI-assisted intraoperative imaging methods, if they are all implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer server or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0035] Computer-readable storage media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0036] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-assisted intraoperative imaging method, characterized in that, Includes the following steps: S1: Preprocessing multimodal medical images to obtain preprocessed images; deep learning model segmentation and identification of anatomical structures based on the preprocessed images; measurement of anatomical parameters including pedicle diameter, angle, and spatial distance based on the segmentation and identification results; and artificial intelligence generation of surgical planning path based on the anatomical parameters. S2: Real-time acquisition of C-arm fluoroscopic image stream, dynamic tracking of surgical instrument spatial coordinates, comparison of instrument position with the surgical planning path, calculation of offset, and when the offset is greater than the offset threshold, output of correction guidance through AR overlay layer; S3: Monitor the image quality index in real time, and dynamically adjust the exposure parameters through a reinforcement learning model to make the image quality index greater than or equal to the quality threshold and minimize the radiation dose. When a metal implant is detected, switch to dual-spectrum mode and execute an artifact suppression algorithm to achieve dynamic optimization.

2. The artificial intelligence-assisted intraoperative imaging method according to claim 1, characterized in that, In step S1, preprocessing the multimodal medical images yields preprocessed images, and the deep learning model segments and identifies anatomical structures based on the preprocessed images, further including: Multimodal medical images, including CT image sequences, T1-weighted MRI images, and X-ray anteroposterior and lateral views, are acquired. The multimodal medical images are registered to the same three-dimensional coordinate system through affine transformation to obtain a registered image. The registered image is then subjected to isotropic resampling and window width and window level standardization to obtain the preprocessed image. The deep learning model extracts multi-scale feature maps through the encoder, and the decoder performs feature fusion based on the multi-scale feature maps to obtain a feature fusion map. It also dynamically adjusts spatial and channel spatial attention through attention gating to highlight key point features. The output layer outputs the anatomical structures, including bones, blood vessels, nerves, and lesions, based on the feature fusion map.

3. The artificial intelligence-assisted intraoperative imaging method according to claim 2, characterized in that, In step S1, anatomical parameters, including geometric dimensions, spatial relationships, and density / signal features, are measured based on the segmentation and recognition results, further including: Based on the segmentation and recognition results, geometric dimensions including length, angle, area, and volume are calculated. Among them, the closed contour of the inner boundary of the bone cortex is extracted based on the segmentation and recognition results, and the maximum inscribed circle of the closed contour of the inner boundary of the bone cortex is calculated. The pedicle diameter is the diameter of the safety passage. Based on the segmentation and recognition results, spatial relationships including distance, angle, and relative position are calculated. A three-dimensional coordinate system is established with the center of the vertebra as the origin, the sagittal plane as the XZ plane, and the transverse plane as the XY plane. The pedicle axis vector is calculated, and the pedicle angle, including the sagittal plane tilt angle and the transverse plane inclination angle, is calculated based on the pedicle axis vector. The straight line for the pedicle screw planning path is obtained. The neural structure contour point set is extracted based on the segmentation and recognition results, and the minimum Euclidean distance from each neural structure contour point to the screw path straight line is calculated. The average grayscale value / CT value of a specific region is calculated based on the segmentation and recognition results.

4. The artificial intelligence-assisted intraoperative imaging method according to claim 3, characterized in that, In step S1, generating a surgical planning path based on the anatomical parameters further includes: Load a guide library including anatomical constraints, instrument rules, and path templates, wherein the anatomical constraints include the minimum diameter of the bony channel and the threshold for the nerve safety distance, the instrument rules include the implant size calculation formula, and the path templates include the standard pin insertion point coordinate range; The optimal implant size is calculated based on the implant size calculation formula and the anatomical parameters. The guide library and the anatomical parameters are then verified according to the verification conditions to obtain the verification results. Artificial intelligence obtains the surgical planning path based on the anatomical structure, the anatomical parameters, the guide library, and the optimal implant size; The verification conditions include: If the pedicle diameter is greater than or equal to the minimum diameter of the bony canal, the screw diameter is large enough to pass through the bony canal. If the pedicle diameter is less than the minimum diameter of the bony canal, the doctor is prompted to adjust the screw diameter or replan the path to obtain a safe pedicle diameter.

5. The artificial intelligence-assisted intraoperative imaging method according to claim 1, characterized in that, In step S2, the C-arm fluoroscopic image stream is acquired in real time, the spatial coordinates of the surgical instruments are dynamically tracked, the instrument positions are compared with the surgical planning path, and the offset is calculated. This further includes: Real-time acquisition of C-arm fluoroscopic image stream, and acquisition of the three-dimensional coordinates of the surgical instrument tip, i.e. the current position of the instrument, through an optical tracking system; The three-dimensional coordinates are mapped to the C-arm image coordinate system to generate projected coordinates; Calculate the offset of the projected coordinates from the surgical planning path.

6. The artificial intelligence-assisted intraoperative imaging method according to claim 5, characterized in that, In step S2, when the offset is greater than the offset threshold, a correction guide is output through the AR overlay layer, further including: AR guidance including directional arrows, distance scales, and safety passages is superimposed on real-time C-arm fluoroscopic images. The directional arrows are drawn as cone-shaped arrows with the current position of the instrument as the starting point and along the offset direction. The arrow length is obtained according to the offset. The distance scale is displayed according to the offset. A semi-transparent green strip area is rendered along the surgical planning path. The width of the safety passage is set according to the offset threshold. When the offset exceeds the offset threshold, the degree of deviation is indicated by changing the arrow, and a prompt is given through vibration and voice.

7. The artificial intelligence-assisted intraoperative imaging method according to claim 1, characterized in that, In step S3, the image quality index is monitored, and exposure parameters are dynamically adjusted using a reinforcement learning model to ensure that the image quality index is greater than or equal to a quality threshold and the radiation dose is minimized. This further includes: Real-time calculation of image quality parameter IQI on C-arm fluoroscopic images; The current state, including the image quality parameters, cumulative dose, tissue thickness, and operation stage, is input into the reinforcement learning model. The reinforcement learning model outputs an action to adjust the exposure parameters, and is trained and optimized by maximizing the reward function. When a sensitive organ is detected, a penalty term is added to the maximum reward function. In the fetal region, a strict mode is enabled to limit the exposure parameters. The diaphragm movement speed is calculated by optical flow, and a motion blur penalty term is added to the reward function.

8. The artificial intelligence-assisted intraoperative imaging method according to claim 7, characterized in that, In step S3, when a metal implant is detected, the dual-spectral mode is switched and an artifact suppression algorithm is executed, further including: Real-time analysis of C-arm fluoroscopic images to calculate metal probability maps for identifying metal regions: When the area ratio of the metal region is greater than the metal region threshold, the dual-energy spectrum mode is triggered; After triggering, the X-ray source is controlled to alternately emit low-energy and high-energy spectrum data over a period of time, while simultaneously acquiring dual-energy projection data. The dual-energy projection data is decomposed into a base material density map. Metal trajectory regions are extracted from the base material density map, and contamination projection data is repaired using neighborhood-weighted interpolation to obtain a metal projection edge map. Input contamination projection data and the metal projection edge map into the U-Net++ network, and output an artifact-suppressed image; The final image is obtained based on the base material density map, the artifact suppression image, and the low-frequency and high-frequency components.

9. An artificial intelligence-assisted intraoperative imaging system, characterized in that, include: The preoperative planning module preprocesses multimodal medical images to obtain preprocessed images. A deep learning model segments and identifies anatomical structures based on the preprocessed images. Based on the segmentation and identification results, it measures anatomical parameters including pedicle diameter, angle, and spatial distance. Based on the anatomical parameters, it generates a surgical planning path. The intraoperative navigation module acquires C-arm fluoroscopic images in real time, dynamically tracks the spatial coordinates of surgical instruments, compares the instrument positions with the planned surgical path, calculates the offset, and outputs correction guidance through AR overlay when the offset exceeds the offset threshold. The imaging optimization module monitors the image quality index and dynamically adjusts the exposure parameters through a reinforcement learning model to ensure that the image quality index is greater than or equal to the quality threshold and the radiation dose is minimized. When a metal implant is detected, the dual-spectrum mode is switched and an artifact suppression algorithm is executed.

10. A computer-readable storage medium storing computer code that, when executed, performs the steps of the artificial intelligence-assisted intraoperative imaging method as described in any one of claims 1-8.

Images