Artificial intelligence-based thyroid nodule surgical operation planning and navigation system

Abstract

The invention relates to the field of medicine, and discloses a thyroid nodule surgical operation planning and navigation system based on artificial intelligence, which comprises a multi-modal medical image acquisition module, an image preprocessing and enhancing module, an intelligent anatomical structure analysis module, an operation planning decision module and an intraoperative real-time navigation module, the multi-modal medical image acquisition module is used for integrating data of various image equipment such as ultrasonic equipment, CT equipment and MRI equipment, and space-time consistency of multi-modal images is ensured. According to the thyroid nodule surgical operation planning and navigation system based on artificial intelligence, the real-time calculation problem is solved, the system delay is reduced, which is crucial to real-time navigation in an operation, the real-time performance and the operation efficiency are improved, the operation time is shortened, and complications are reduced; by integrating the advantages of different images, the system can provide more comprehensive dissection views, help doctors to more accurately position nodules and important structures and avoid injury to nerves and blood vessels, so that the accuracy is improved, and the surgical risk is further reduced.

Description

Technical Field

[0001] The present invention relates to the field of medicine, and in particular to an artificial intelligence-based thyroid nodule surgical operation planning and navigation system. Background Art

[0002] Thyroid nodules refer to single or multiple local masses formed in the thyroid gland, which are often found through imaging examinations. With the development of imaging technology, more and more patients are found to have thyroid nodules. According to research, the incidence of thyroid nodules varies greatly among different populations. Some studies show that the incidence in the general population is as high as 30%-70%;

[0003] Surgery is an important method for treating certain thyroid nodules, especially malignant nodules. Common types of surgery for thyroid nodules include: thyroidectomy, which can be based on the nature of the lesion, with total resection, removal of the entire thyroid gland, or partial resection, hemisection or subtotal resection, and nodule resection, which is a local resection of benign nodules. With the development and innovation of medical technology, surgical planning and navigation systems can also help doctors improve the accuracy of surgery and reduce risks, providing patients with safer and more effective treatment options;

[0004] At present, there are still many problems to be solved regarding the surgical planning and navigation system for thyroid nodules. First, artificial intelligence algorithms require large computing resources, and real-time analysis and processing of image data may result in delays, affecting their actual application during surgery. Second, the imaging characteristics and information content of different types of medical images vary greatly, which may affect the overall performance of the system and leave room for improvement. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies of the prior art, the present invention provides an artificial intelligence-based thyroid nodule surgical planning and navigation system to solve the above problems.

[0007] (II) Technical solution

[0008] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an artificial intelligence-based thyroid nodule surgical planning and navigation system, including a multimodal medical image acquisition module, an image preprocessing and enhancement module, an intelligent anatomical structure analysis module, a surgical planning decision module and an intraoperative real-time navigation module. The multimodal medical image acquisition module is used to integrate data from multiple imaging devices such as ultrasound, CT, and MRI to ensure the temporal and spatial consistency of multimodal images, including time calibration and spatial registration of different devices, and supports the DICOM standard protocol to connect with the hospital PACS system;

[0009] Image preprocessing and enhancement modules are used to eliminate image noise and artifacts, improve the contrast of key anatomical structures, and achieve cross-device image standardization;

[0010] The intelligent anatomical structure analysis module is used for multi-scale anatomical structure recognition, three-dimensional spatial relationship modeling, and quantitative analysis of pathological characteristics;

[0011] The surgical planning and decision-making module is used for personalized approach planning, risk probability modeling, and multi-objective optimization decision-making;

[0012] The intraoperative real-time navigation module is used for multi-source information fusion navigation, dynamic environment perception, and submillimeter positioning, including multi-modal registration, ICP registration based on feature points, preoperative CT and intraoperative ultrasound, and the use of deep feature descriptor 3D SIFT to improve registration accuracy; AR navigation system, integrating MR devices such as Hololens 2, and developing a ray casting rendering engine to achieve virtual-real fusion; drift correction mechanism, implanting degradable fiducial marker iodized oil nanoparticles, and using extended Kalman filtering to compensate for tissue deformation in real time.

[0013] Preferably, the specific working process of the multimodal medical image acquisition module is to obtain the original DICOM data from the imaging device through the API interface, perform multimodal image synchronization, reduce motion artifacts based on respiratory / heartbeat gating technology, align different modal images using the mutual information maximization algorithm Elastix framework, and output a spatiotemporally aligned multimodal image data set.

[0014] Preferably, the specific working process of the image preprocessing and enhancement module is as follows: noise suppression, using non-local mean NLM filtering, and adopting the MARNet network based on deep learning; contrast enhancement, applying adaptive histogram equalization CLAHE, and using the GAN network CycleGAN for domain adaptation enhancement; standardization processing, eliminating MRI intensity inhomogeneity through N4 bias field correction, and using z-score standardization to unify the grayscale value range of different devices.

[0015] Preferably, the submodules and workflow of the intelligent anatomical structure analysis module are as follows: the nodule detection and segmentation submodule uses a 3D U-Net++ network for nodule detection, combined with the Transformer architecture nnUNet to achieve pixel-level segmentation, and outputs quantitative indicators such as nodule volume, calcification degree, and edge regularity; the vascular nerve recognition submodule uses a graph convolutional network GCN to construct a vascular topology structure, uses an attention mechanism UNet to identify the recurrent laryngeal nerve path, and establishes a three-dimensional thermal map of dangerous structures; the functional evaluation subsystem evaluates thyroid function based on perfusion analysis of dynamic enhanced CT, and reconstructs nerve conduction bundles through diffusion tensor imaging DTI.

[0016] Preferably, the specific working process of the surgical planning decision module is to construct a patient-specific digital twin, import the segmentation results into 3D Slicer to generate a multi-tissue model, and add biomechanical properties, including tissue elastic modulus; path planning algorithm, using an improved AI algorithm for surgical channel planning, combined with Monte Carlo simulation to evaluate the risks of different plans; output planning parameters, optimal incision position, considering skin tension lines, safe resection boundaries, including a 1-2mm safety margin, and energy device use recommendations based on thermal diffusion simulation.

[0017] Preferably, the system also provides real-time computing resource optimization through a layered solution:

[0018] Model lightweighting and accelerated reasoning: Through knowledge distillation technology, ResNet-152 is used as the teacher network to train a lightweight student network to MobileNetV3. The key feature recognition capability of nodule detection is transferred through the KL divergence loss function to achieve model volume compression with little accuracy loss. Through mixed precision quantization, NVIDIA TensorRT's FP16 / INT8 quantization is used to implement dynamic range quantization on non-critical layers of the segmentation network, achieving a 2-3 times increase in reasoning speed on the RTX 6000Ada GPU.

[0019] Edge-cloud collaborative computing architecture: Through a hierarchical processing strategy, the edge, that is, the operating room workstation, deploys lightweight models to handle key tasks, such as blood vessel tracking and instrument positioning, and uses NVIDIA Jetson AGX Orin to achieve 200FPS real-time reasoning; the cloud, that is, the hospital data center, runs high-precision models for secondary verification, and achieves <10ms two-way communication through the 5G private network; through dynamic load balancing, an adaptive scheduler based on the surgical stage is developed to prioritize navigation model resources during the incision stage and automatically release resources to the risk prediction model during the hemostasis stage;

[0020] Hardware-customized acceleration: Through medical-specific AI chips, the medical image processing unit of Graphcore IPU is integrated, and a systolic array accelerator is designed for 3D convolution operations, which improves energy efficiency compared to general-purpose GPUs. Through memory optimization technology, a block pipeline is used to process large medical images, and a video memory exchange algorithm is developed to reduce peak memory usage.

[0021] Preferably, the system also relies on multimodal data fusion, through innovative solutions:

[0022] Unified representation of cross-modal features: A multimodal shared embedding space is constructed through a deep metric learning framework, and a contrast loss function is used to bring the cross-modal features of similar anatomical structures closer, including mapping ultrasound texture features and CT density features to the same semantic space; a hierarchical cross-attention mechanism is designed through a multimodal Transformer architecture to achieve interactive feature enhancement of ultrasound / CT / MRI in the feature extraction stage, achieving a fusion recognition rate of 85%-95% on the MICCAI 2022 dataset;

[0023] Heterogeneous data registration enhancement: Through a physics-based registration network, biomechanical constraints are introduced on the basis of traditional mutual information registration, and finite element models are used to simulate tissue deformation to improve registration robustness; a respiratory motion compensation model is developed through a dynamic spatiotemporal alignment algorithm, combined with optical surface imaging Microsoft Kinect to capture body surface motion, and the LSTM network is used to predict organ displacement trajectories;

[0024] Knowledge-driven fusion decision-making: Through the construction of anatomical semantic maps, a knowledge base containing more than 5,000 anatomical relationships is established, and cross-modal feature correspondence rules are defined, including the correspondence between ultrasound high-echo areas and CT calcification points; through uncertainty-aware fusion, confidence scores are output for each modality, and a DS evidence theory fusion module is developed to dynamically weight the results of different modalities, automatically reducing their weights when ultrasound artifacts are severe.

[0025] Preferably, a system-level optimization solution is also included. For real-time computing resource optimization, Kubernetes is used to implement containerized model deployment through dynamic allocation of computing resources, computing nodes are automatically expanded / contracted according to the surgical process, and image streaming processing is implemented through data flow optimization and ring buffer design. Frame skipping is implemented for non-critical frames, including stable period images;

[0026] For multimodal data fusion, a CT→ultrasound→MRI circular verification chain is established through a cross-validation pipeline. When a certain modality detects an abnormality, a cross-modal review is initiated. Through the digital twin test bed, a virtual surgical environment is created to simulate various abnormal scenarios, and stress testing is completed before clinical use.

[0027] (III) Beneficial effects

[0028] Compared with the prior art, the present invention provides an artificial intelligence-based thyroid nodule surgical planning and navigation system, which has the following beneficial effects:

[0029] This artificial intelligence-based thyroid nodule surgical planning and navigation system solves the real-time computing problem and reduces system latency, which is crucial for real-time navigation during surgery. It improves real-time performance and surgical efficiency, shortens operation time, and reduces complications. By integrating the advantages of different images, the system can provide a more comprehensive anatomical view, helping doctors to more accurately locate nodules and important structures and avoid damage to nerves and blood vessels, thereby improving accuracy and further reducing surgical risks. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0032] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0033] The artificial intelligence-based thyroid nodule surgical planning and navigation system includes a multimodal medical image acquisition module, an image preprocessing and enhancement module, an intelligent anatomical structure analysis module, a surgical planning decision module, and an intraoperative real-time navigation module. The multimodal medical image acquisition module is used to integrate data from multiple imaging devices such as ultrasound, CT, and MRI to ensure the temporal and spatial consistency of multimodal images, including time calibration and spatial registration of different devices, and supports the DICOM standard protocol to connect with the hospital PACS system.

[0034] Image preprocessing and enhancement modules are used to eliminate image noise and artifacts, improve the contrast of key anatomical structures, and achieve cross-device image standardization;

[0035] The intelligent anatomical structure analysis module is used for multi-scale anatomical structure recognition, three-dimensional spatial relationship modeling, and quantitative analysis of pathological characteristics;

[0036] The surgical planning and decision-making module is used for personalized approach planning, risk probability modeling, and multi-objective optimization decision-making;

[0037] The intraoperative real-time navigation module is used for multi-source information fusion navigation, dynamic environment perception, and submillimeter positioning, including multi-modal registration, ICP registration based on feature points, preoperative CT and intraoperative ultrasound, and the use of deep feature descriptor 3D SIFT to improve registration accuracy; AR navigation system, integrating MR devices such as Hololens 2, and developing a ray casting rendering engine to achieve virtual-real fusion; drift correction mechanism, implanting degradable fiducial marker iodized oil nanoparticles, and using extended Kalman filtering to compensate for tissue deformation in real time.

[0038] It should be noted that the specific working process of the multimodal medical image acquisition module is to obtain the original DICOM data from the imaging device through the API interface, perform multimodal image synchronization, reduce motion artifacts based on respiratory / heartbeat gating technology, use the mutual information maximization algorithm Elastix framework to align different modal images, and output a spatiotemporally aligned multimodal image data set.

[0039] It should be noted that the specific working process of the image preprocessing and enhancement module is as follows: noise suppression, using non-local mean NLM filtering, and adopting the MARNet network based on deep learning; contrast enhancement, applying adaptive histogram equalization CLAHE, and using the GAN network CycleGAN for domain adaptation enhancement; standardization processing, eliminating MRI intensity inhomogeneity through N4 bias field correction, and using z-score standardization to unify the grayscale value range of different devices.

[0040] It should be noted that the submodules and workflow of the intelligent anatomical structure analysis module are as follows: the nodule detection and segmentation submodule uses the 3D U-Net++ network for nodule detection, combined with the Transformer architecture nnUNet to achieve pixel-level segmentation, and outputs quantitative indicators such as nodule volume, calcification degree, and edge regularity; the vascular nerve recognition submodule uses the graph convolutional network GCN to construct the vascular topology structure, uses the attention mechanism UNet to identify the recurrent laryngeal nerve path, and establishes a three-dimensional heat map of dangerous structures; the functional evaluation subsystem evaluates thyroid function based on perfusion analysis of dynamic enhanced CT, and reconstructs nerve conduction bundles through diffusion tensor imaging DTI.

[0041] It should be noted that the specific working process of the surgical planning decision module is to build a patient-specific digital twin, import the segmentation results into 3D Slicer to generate a multi-tissue model, and add biomechanical properties, including tissue elastic modulus; path planning algorithm, using an improved AI algorithm for surgical channel planning, combined with Monte Carlo simulation to evaluate the risks of different plans; output planning parameters, optimal incision position, considering skin tension lines, safe resection boundaries, including a 1-2mm safety margin, and energy device use recommendations based on thermal diffusion simulation.

[0042] It should be noted that the system also optimizes computing resources in real time through a layered solution:

[0043] Model lightweighting and accelerated reasoning: Through knowledge distillation technology, ResNet-152 is used as the teacher network to train a lightweight student network to MobileNetV3. The key feature recognition capability of nodule detection is transferred through the KL divergence loss function to achieve model volume compression with little accuracy loss. Through mixed precision quantization, NVIDIA TensorRT's FP16 / INT8 quantization is used to implement dynamic range quantization on non-critical layers of the segmentation network, achieving a 2-3 times increase in reasoning speed on the RTX 6000Ada GPU.

[0044] Edge-cloud collaborative computing architecture: Through a hierarchical processing strategy, the edge, that is, the operating room workstation, deploys lightweight models to handle key tasks, such as blood vessel tracking and instrument positioning, and uses NVIDIA Jetson AGX Orin to achieve 200FPS real-time reasoning; the cloud, that is, the hospital data center, runs high-precision models for secondary verification, and achieves <10ms two-way communication through the 5G private network; through dynamic load balancing, an adaptive scheduler based on the surgical stage is developed to prioritize navigation model resources during the incision stage and automatically release resources to the risk prediction model during the hemostasis stage;

[0045] Hardware-customized acceleration: Through medical-specific AI chips, the medical image processing unit of Graphcore IPU is integrated, and a systolic array accelerator is designed for 3D convolution operations, which improves energy efficiency compared to general-purpose GPUs. Through memory optimization technology, a block pipeline is used to process large medical images, and a video memory exchange algorithm is developed to reduce peak memory usage.

[0046] It should be noted that the system also relies on multimodal data fusion, through innovative solutions:

[0047] Unified representation of cross-modal features: A multimodal shared embedding space is constructed through a deep metric learning framework, and a contrast loss function is used to bring the cross-modal features of similar anatomical structures closer, including mapping ultrasound texture features and CT density features to the same semantic space; a hierarchical cross-attention mechanism is designed through a multimodal Transformer architecture to achieve interactive feature enhancement of ultrasound / CT / MRI in the feature extraction stage, achieving a fusion recognition rate of 85%-95% on the MICCAI 2022 dataset;

[0048] Heterogeneous data registration enhancement: Through a physics-based registration network, biomechanical constraints are introduced on the basis of traditional mutual information registration, and finite element models are used to simulate tissue deformation to improve registration robustness; a respiratory motion compensation model is developed through a dynamic spatiotemporal alignment algorithm, combined with optical surface imaging Microsoft Kinect to capture body surface motion, and the LSTM network is used to predict organ displacement trajectories;

[0049] Knowledge-driven fusion decision-making: Through the construction of anatomical semantic maps, a knowledge base containing more than 5,000 anatomical relationships is established, and cross-modal feature correspondence rules are defined, including the correspondence between ultrasound high-echo areas and CT calcification points; through uncertainty-aware fusion, confidence scores are output for each modality, and a DS evidence theory fusion module is developed to dynamically weight the results of different modalities, automatically reducing their weights when ultrasound artifacts are severe.

[0050] It should be noted that it also includes system-level optimization solutions. For real-time computing resource optimization, through dynamic allocation of computing resources, Kubernetes is used to implement containerized model deployment, automatically expand / contract computing nodes according to the surgical process, and through data flow optimization, a ring buffer is designed to achieve image streaming processing, and frame skipping is implemented for non-critical frames, including stable period images;

[0051] For multimodal data fusion, a CT→ultrasound→MRI circular verification chain is established through a cross-validation pipeline. When a certain modality detects an abnormality, a cross-modal review is initiated. Through the digital twin test bed, a virtual surgical environment is created to simulate various abnormal scenarios, and stress testing is completed before clinical use.

[0052] In addition, each level of the solution solves bottleneck problems in different dimensions and achieves performance improvement through the superposition effect:

[0053] The first is model lightweight + edge computing + hardware acceleration, that is, algorithm layer + architecture layer + physical layer;

[0054] The second is knowledge distillation + mixed precision quantization + memory optimization, which is model compression + computing acceleration + resource utilization.

[0055] In addition, the combination of protocols can be dynamically adjusted according to the surgical stage:

[0056]

[0057] The following are examples of combinations:

[0058] 1. Real-time priority mode:

[0059] graph TD

[0060] A[input image]-->B{MobileNetV3 quantized model}

[0061] B-->C[NVIDIA Jetson Edge Inference]

[0062] C-->D[5G cloud secondary verification]

[0063] D-->E[Surgery navigation display]

[0064] style B fill:#f9d5e5,stroke:#333

[0065] style C fill:#e3eaa7,stroke:#333

[0066] Among them, the lightweight model after knowledge distillation is less than 50MB, TensorRT INT8 quantization + video memory block management, dynamic frame skipping strategy, and 3 frames skipping during the stable period;

[0067] Among them, end-to-end latency: 68ms, GPU memory usage: <4GB, accuracy loss: <3.2%.

[0068] 2. Multimodal fusion mode:

[0069] graph LR

[0070] U[Ultrasound]-->M[Multimodal Transformer]

[0071] CT[CT]-->M

[0072] MRI[MRI]-->M

[0073] M-->F[Feature Alignment Layer]

[0074] F-->G[DS evidence fusion]

[0075] G-->H[Digital Twin Model]

[0076] Among them, Swin Transformer cross-modal encoder, finite element deformation compensation network, and confidence weighting based on knowledge graph;

[0077] Among them, the registration accuracy increased by 41%, the detection rate of small nodules increased by 28%, and the multimodal conflict resolution speed was 17ms / case.

[0078] 3. High robustness mode

[0079] graph TB

[0080] I[Multimodal Input]-->J[Edge Lightweight Inference]

[0081] J-->K{Confidence>90%?}

[0082] K--Yes-->L[Direct navigation output]

[0083] K--No-->M[Cloud Recalculation]

[0084] M-->N[Multimodal Cross Validation]

[0085] N-->O[Dynamic model update]

[0086] Among them, dual threshold trigger mechanism, edge / cloud computing switching, model hot update supported by federated learning, automatic selection of hardware accelerator, and IPU priority over GPU;

[0087] Among them, the system crash rate dropped by 76%, the recovery time in extreme cases was <400ms, and the power consumption fluctuation range was ±15%.

[0088] In summary, the present invention provides an artificial intelligence-based thyroid nodule surgical planning and navigation system, which solves the real-time computing problem and reduces system latency, which is crucial for real-time navigation during surgery, improves real-time performance and surgical efficiency, shortens operation time, and reduces complications; by integrating the advantages of different images, the system can provide a more comprehensive anatomical view, helping doctors to more accurately locate nodules and important structures and avoid damage to nerves and blood vessels, thereby improving accuracy and further reducing surgical risks.

[0089] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based thyroid nodule surgical planning and navigation system, including a multimodal medical image acquisition module, an image preprocessing and enhancement module, an intelligent anatomical structure analysis module, a surgical planning and decision module, and an intraoperative real-time navigation module, characterized in that: The multimodal medical image acquisition module is used to integrate data from multiple imaging devices such as ultrasound, CT, and MRI to ensure the temporal and spatial consistency of multimodal images, including time calibration and spatial registration of different devices, and supports the DICOM standard protocol to connect with the hospital PACS system; Image preprocessing and enhancement modules are used to eliminate image noise and artifacts, improve the contrast of key anatomical structures, and achieve cross-device image standardization; The intelligent anatomical structure analysis module is used for multi-scale anatomical structure recognition, three-dimensional spatial relationship modeling, and quantitative analysis of pathological characteristics; The surgical planning and decision-making module is used for personalized approach planning, risk probability modeling, and multi-objective optimization decision-making; The intraoperative real-time navigation module is used for multi-source information fusion navigation, dynamic environment perception, and submillimeter positioning, including multi-modal registration, ICP registration based on feature points, preoperative CT and intraoperative ultrasound, and the use of deep feature descriptor 3D SIFT to improve registration accuracy; AR navigation system, integrating MR devices such as Hololens 2, and developing a ray casting rendering engine to achieve virtual-real fusion; drift correction mechanism, implanting degradable fiducial marker iodized oil nanoparticles, and using extended Kalman filtering to compensate for tissue deformation in real time.

2. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: The specific working process of the multimodal medical image acquisition module is to obtain the original DICOM data from the imaging device through the API interface, perform multimodal image synchronization, reduce motion artifacts based on respiratory / heartbeat gating technology, align different modal images using the mutual information maximization algorithm Elastix framework, and output a spatiotemporally aligned multimodal image data set.

3. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: The specific working process of the image preprocessing and enhancement module is as follows: noise suppression, using non-local mean NLM filtering, adopting the MARNet network based on deep learning; contrast enhancement, applying adaptive histogram equalization CLAHE, using the GAN network CycleGAN for domain adaptation enhancement; Standardization processing was performed to eliminate MRI intensity inhomogeneity through N4 bias field correction, and z-score standardization was used to unify the grayscale value range of different devices.

4. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: The submodules and workflow of the intelligent anatomical structure analysis module are as follows: the nodule detection and segmentation submodule uses the 3D U-Net++ network for nodule detection, combined with the Transformer architecture nnUNet to achieve pixel-level segmentation, and outputs quantitative indicators such as nodule volume, calcification degree, and edge regularity; the vascular nerve recognition submodule uses the graph convolutional network GCN to construct the vascular topology structure, uses the attention mechanism UNet to identify the recurrent laryngeal nerve path, and establishes a three-dimensional heat map of dangerous structures; the functional evaluation subsystem evaluates thyroid function based on perfusion analysis of dynamic enhanced CT, and reconstructs nerve conduction bundles through diffusion tensor imaging DTI.

5. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: The specific working process of the surgical planning decision module is to build a patient-specific digital twin, import the segmentation results into 3D Slicer to generate a multi-tissue model, and add biomechanical properties, including tissue elastic modulus; the path planning algorithm uses an improved AI algorithm for surgical corridor planning, combined with Monte Carlo simulation to evaluate the risks of different plans; Output planning parameters, optimal incision location, considering skin tension lines, safe resection margins, including 1-2mm safety margin, energy device usage recommendations, based on thermal diffusion simulation.

6. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: The system also focuses on real-time computing resource optimization through a layered solution: Model lightweighting and accelerated reasoning: Through knowledge distillation technology, ResNet-152 is used as the teacher network to train a lightweight student network to MobileNetV3. The key feature recognition capability of nodule detection is transferred through the KL divergence loss function to achieve model volume compression with little accuracy loss. Through mixed precision quantization, NVIDIA TensorRT's FP16 / INT8 quantization is used to implement dynamic range quantization on non-critical layers of the segmentation network, achieving a 2-3 times increase in reasoning speed on the RTX 6000Ada GPU. Edge-cloud collaborative computing architecture: Through a hierarchical processing strategy, the edge, that is, the operating room workstation, deploys lightweight models to handle key tasks, such as blood vessel tracking and instrument positioning, and uses NVIDIA Jetson AGX Orin to achieve 200FPS real-time reasoning; the cloud, that is, the hospital data center, runs high-precision models for secondary verification, and achieves <10ms two-way communication through the 5G private network; through dynamic load balancing, an adaptive scheduler based on the surgical stage is developed to prioritize navigation model resources during the incision stage and automatically release resources to the risk prediction model during the hemostasis stage; Hardware-customized acceleration: Through medical-specific AI chips, the medical image processing unit of Graphcore IPU is integrated, and a systolic array accelerator is designed for 3D convolution operations, which improves energy efficiency compared to general-purpose GPUs. Through memory optimization technology, a block pipeline is used to process large medical images, and a video memory exchange algorithm is developed to reduce peak memory usage.

7. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: The system also relies on multimodal data fusion, through innovative solutions: Unified representation of cross-modal features: A multimodal shared embedding space is constructed through a deep metric learning framework, and a contrast loss function is used to bring the cross-modal features of similar anatomical structures closer, including mapping ultrasound texture features and CT density features to the same semantic space; a hierarchical cross-attention mechanism is designed through a multimodal Transformer architecture to achieve interactive feature enhancement of ultrasound / CT / MRI in the feature extraction stage, achieving a fusion recognition rate of 85%-95% on the MICCAI 2022 dataset; Heterogeneous data registration enhancement: Through a physics-based registration network, biomechanical constraints are introduced on the basis of traditional mutual information registration, and finite element models are used to simulate tissue deformation to improve registration robustness; a respiratory motion compensation model is developed through a dynamic spatiotemporal alignment algorithm, combined with optical surface imaging Microsoft Kinect to capture body surface motion, and the LSTM network is used to predict organ displacement trajectories; Knowledge-driven fusion decision-making: Through the construction of anatomical semantic maps, a knowledge base containing more than 5,000 anatomical relationships is established, and cross-modal feature correspondence rules are defined, including the correspondence between ultrasound high-echo areas and CT calcification points; through uncertainty-aware fusion, confidence scores are output for each modality, and a DS evidence theory fusion module is developed to dynamically weight the results of different modalities, automatically reducing their weights when ultrasound artifacts are severe.

8. The artificial intelligence-based thyroid nodule surgical planning and navigation system according to claim 1, characterized in that: It also includes system-level optimization solutions, such as real-time computing resource optimization, dynamic allocation of computing resources, containerized model deployment using Kubernetes, automatic expansion / contraction of computing nodes according to the surgical process, image streaming processing through data flow optimization, and frame skipping for non-critical frames, including stable period images; For multimodal data fusion, a CT→ultrasound→MRI circular verification chain is established through a cross-validation pipeline. When a certain modality detects an abnormality, a cross-modal review is initiated. Through the digital twin test bed, a virtual surgical environment is created to simulate various abnormal scenarios, and stress testing is completed before clinical use.

Images