Intelligent navigation system for cryoablation of liver cancer based on artificial intelligence

Abstract

The present invention discloses an intelligent navigation system and method for liver cancer cryoablation based on artificial intelligence, which relates to the field of medical device technology. The system includes: an image acquisition module for real-time acquisition of original ultrasound image data of liver cancer patients; an image preprocessing module for preprocessing the acquired original ultrasound images; a machine learning model for feature segmentation of the processed images based on a deep learning algorithm, and accurate prediction and positioning of the spatial position of the cryoablation needle tip in the patient's liver; a path navigation module for planning the optimal ablation needle penetration path and angle based on the needle tip positioning result, and providing real-time navigation information to ensure that the ablation needle accurately reaches the tumor center; a treatment execution module for controlling the rapid expansion of high-pressure argon gas to achieve cryoablation treatment and inactivation of tumor cells. The present invention can achieve the ablation needle accurately reaching the tumor center for automated treatment, thereby improving surgical accuracy, safety and treatment efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical devices, and specifically relates to an intelligent navigation system and method for liver cancer cryoablation based on artificial intelligence. Background Art

[0002] Liver cancer cryoablation, especially argon-helium cryoablation treatment technology, is a method of inactivating tumor cell activity by cooling and heating the local tumor of liver cancer patients with argon and helium. The ablation needle can be directly and accurately inserted into the cancer tissue under the guidance of B-ultrasound. B-ultrasound faces challenges in accurately positioning the puncture needle, mainly due to the blurry, unclear or even invisible needle body caused by acoustic interference, especially the inaccurate positioning of the needle tip. Traditional liver cryoablation treatment relies on the doctor's experience and manual operation, and there are problems such as inaccurate positioning, low treatment efficiency, and great pain for patients. In the prior art, although there is technology for tissue imaging using ultrasound detection, it lacks effective combination with machine learning algorithms to achieve high-precision needle tip positioning and automated treatment process. Summary of the invention

[0003] The present invention aims to provide an intelligent navigation system and method for liver cancer cryoablation based on artificial intelligence. The system processes images through a deep learning algorithm, alleviates acoustic interference, obtains high-quality image data, and uses a machine learning algorithm to achieve precise positioning of the needle tip, ensuring that the ablation needle accurately reaches the center of the tumor for automated treatment, thereby improving surgical accuracy, safety, and treatment efficiency.

[0004] In a first aspect, an embodiment of the present invention provides an intelligent navigation system for liver cancer cryoablation based on artificial intelligence, comprising:

[0005] An image acquisition module, used for acquiring original ultrasound image data of liver cancer patients in real time under the guidance of a B-ultrasound scanning device;

[0006] An image preprocessing module is used to perform preprocessing operations on the collected original ultrasound images to extract clear tissue boundaries;

[0007] A machine learning model for feature segmentation of processed images based on a deep learning algorithm and for accurate prediction and localization of the spatial position of the cryoablation needle tip within the patient’s liver;

[0008] The path navigation module plans the optimal ablation needle insertion path and angle based on the needle tip positioning results and the instructions input by medical staff, and provides real-time navigation information to ensure that the ablation needle accurately reaches the center of the tumor;

[0009] The treatment execution module, based on the guidance of the path planning and navigation module, controls the rapid expansion of high-pressure argon gas to achieve cryoablation treatment, thereby inactivating tumor cells by lowering the temperature of surrounding tissues to below freezing point.

[0010] Furthermore, the system also includes: a user interaction module;

[0011] The user interaction module is used to display the ultrasound image and the predicted needle tip position, the insertion path planning and navigation information in real time, and receive the operation instructions of the medical staff.

[0012] Furthermore, the system also includes: a three-dimensional model construction module, which is used to collect and organize the patient's B-ultrasound, CT and MRI image sample data; pre-process the B-ultrasound, CT and MRI image sample data respectively; use a multimodal image registration algorithm to register the B-ultrasound, CT and MRI image data under each modality to obtain registered image data; according to the registration result, based on the image fusion algorithm, fuse the information of the B-ultrasound, CT and MRI medical image modalities, and then construct a liver and tumor three-dimensional model containing rich information;

[0013] The user interaction module is also used to provide the liver and tumor three-dimensional model to medical staff through an interactive interface, and supports rotation, scaling, and cutting operations.

[0014] Furthermore, the image preprocessing module also includes filtering, denoising and edge detection algorithms.

[0015] Furthermore, the cryoablation treatment execution module is also integrated with a temperature monitoring and feedback mechanism to monitor the temperature changes in the treatment area in real time, ensure that the cryoablation process is carried out within a safe range, and adjust the treatment parameters as needed.

[0016] Furthermore, the deep learning model adopts a convolutional neural network (CNN) or its variant, and is trained through a series of ultrasound image data sets annotated with ablation needle tip position information; the machine learning model includes a feature extraction layer, a maximum pooling layer, an attention mechanism layer, a needle tip positioning layer and a fully connected layer.

[0017] Furthermore,

[0018] The feature extraction layer uses multiple convolution kernels to extract spatial local features of the needle tip and the surrounding tissue edge image, and outputs an activated feature vector;

[0019] The maximum pooling layer reduces the size of the feature vector and retains important features, and its input is the activated feature vector output by the feature extraction layer, and its output is the downsampled feature vector;

[0020] The attention mechanism layer enhances the feature representation capability by assigning higher weights to important features. Its input is the pooled feature vector and its output is the weighted feature vector.

[0021] The needle tip positioning layer generates position information by detecting and locating the position of the needle tip in real time, and its input is a weighted feature vector and its output is a predicted bounding box;

[0022] The fully connected layer outputs a prediction value by integrating features, and its input is the predicted bounding box feature vector output from the needle tip positioning layer, and its output is the spatial position coordinates of the needle tip.

[0023] Furthermore, the tip localization layer predicts bounding box parameters and class probabilities at each unit of the feature vector:

[0024]

[0025] C x,y =Softmax(S x,y )

[0026] Among them, B x,y Represents the prediction box (BoundingBoxes) at coordinates (x, y); t x ,t y ,t w ,t h represents the bounding box offset and scaling parameters predicted by the network; c x ,c y Indicates the offset of the grid unit; p w ,p h represents the width and height of the prediction box; σ represents the Sigmoid function, which ensures that the center point of the bounding box is within the grid cell; S x,y Indicates the classification score; C x,y Represents the class probability distribution.

[0027] Further, the system also includes an optimizer and a triple loss function for training the machine learning model to optimize the alignment between the predicted spatial coordinates and the real spatial coordinates;

[0028] Among them, the optimizer adopts the AdamW optimizer, which provides adaptive learning rate, momentum optimization and weight decay regularization;

[0029] Triple loss function

[0030] in, is the bounding box regression loss, is the confidence loss, is the classification loss.

[0031] In a second aspect, an embodiment of the present invention further provides an intelligent navigation method for liver cancer cryoablation based on artificial intelligence, comprising:

[0032] Under the guidance of B-ultrasound scanning device, the original ultrasound image data of liver cancer patients is collected in real time;

[0033] Perform preprocessing operations on the collected original ultrasound images to extract clear tissue boundaries;

[0034] The processed images are segmented based on deep learning algorithms, and the spatial position of the cryoablation needle tip in the patient's liver is accurately predicted and located;

[0035] Based on the needle tip positioning results and the instructions input by medical staff, the optimal ablation needle penetration path and angle are planned, and real-time navigation information is provided to ensure that the ablation needle accurately reaches the center of the tumor;

[0036] Under the guidance of the path planning and navigation module, high-pressure argon gas is controlled to expand rapidly to achieve cryoablation treatment, which inactivates tumor cells by lowering the temperature of surrounding tissues to below freezing point.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The intelligent navigation system and method for cryoablation of liver cancer based on artificial intelligence of the present invention processes images through deep learning algorithms, alleviates acoustic interference, obtains high-quality image data, and uses machine learning algorithms to achieve precise positioning of the needle tip, ensuring that the ablation needle accurately reaches the center of the tumor for automated treatment, thereby improving surgical accuracy, safety and treatment efficiency. The machine learning model improves the accuracy of needle tip positioning through convolutional neural networks and attention mechanisms. In addition, by integrating information from multiple medical image modalities such as B-ultrasound, CT, and MRI, a high-precision three-dimensional model of the liver and tumor is constructed. Compared with a two-dimensional or three-dimensional model of the liver and tumor constructed by a single modality, the complementarity and enhancement of multi-source data are achieved, providing a more intuitive and comprehensive surgical field of view and accurate navigation information for clinical liver and tumor surgery, thereby improving the success rate of surgery and patient survival rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1A schematic diagram of a module of an intelligent navigation system for liver cancer cryoablation based on artificial intelligence provided by an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of the structure of a machine learning model of an intelligent navigation system for liver cancer cryoablation based on artificial intelligence provided by an embodiment of the present invention;

[0042] Figure 3 A schematic flow chart of an intelligent navigation method for liver cancer cryoablation based on artificial intelligence provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0044] The present invention proposes an intelligent navigation system and method for liver cancer cryoablation based on artificial intelligence to process ultrasound images and predict the spatial position of the needle tip. The feature segmentation technology based on deep learning is used to alleviate acoustic interference and further enhance the visibility of the needle tip. Image processing driven by machine learning is used to accurately predict and determine the spatial position of the needle tip, ensuring ablation accuracy and ensuring that the ablation needle accurately reaches the center of the tumor.

[0045] Figure 1 FIG. 1 is a schematic diagram of a module of an intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to an embodiment of the present invention. Figure 1 As shown, an intelligent navigation system 100 for liver cancer cryoablation based on artificial intelligence comprises:

[0046] The image acquisition module 110 is used to acquire the original ultrasound image data of the liver cancer patient in real time under the guidance of the B-ultrasound scanning device; optionally, the patient lies flat on a lying board with his abdomen facing the B-ultrasound scanning device to facilitate scanning of the patient's abdomen and generating real-time ultrasound images.

[0047] The image preprocessing module 120 is used to perform preprocessing operations on the collected original ultrasound images to extract clear tissue boundaries; specifically, the image preprocessing module includes filtering, denoising and edge detection algorithms, and the above preprocessing process is used to highlight the boundary between the ablation needle tip and the surrounding tissue to reduce acoustic interference, improve image quality, and improve the accuracy of feature segmentation.

[0048] The machine learning model 130 is used to perform feature segmentation on the processed images based on a deep learning algorithm, and to accurately predict and locate the spatial position of the cryoablation needle tip in the patient's liver; it is used to automatically identify key information such as tumor boundaries and vascular distribution, provide accurate navigation paths for surgical planning, and assist doctors in formulating the optimal surgical plan.

[0049] The path navigation module 140 plans the optimal ablation needle penetration path and angle based on the needle tip positioning result and the instructions input by the medical staff, and provides real-time navigation information to ensure that the ablation needle accurately reaches the center of the tumor;

[0050] The treatment execution module 150, based on the guidance of the path planning and navigation module, can plan the penetration path and angle of the ablation needle, and control the movement of the positioning tube and the ablation needle to achieve precise penetration and positioning, and then control the rapid expansion of high-pressure argon gas to achieve cryoablation treatment, thereby reducing the temperature of surrounding tissues to below the freezing point, thereby achieving the inactivation of tumor cells.

[0051] According to the above embodiment, the system processes images through deep learning algorithms, alleviates acoustic interference, obtains high-quality image data, and uses machine learning algorithms to achieve precise positioning of the needle tip, ensuring that the ablation needle accurately reaches the center of the tumor for automated treatment, thereby improving surgical accuracy, safety and treatment efficiency.

[0052] Furthermore, the system 100 also includes: a user interaction module 160; the user interaction module 160 is used to display the ultrasound image and the predicted needle tip position, the insertion path planning and navigation information in real time, and receive the operation instructions of the medical staff.

[0053] Furthermore, the system 100 also includes: a three-dimensional model construction module 170, which is used to collect and organize the patient's B-ultrasound, CT and MRI image sample data; pre-process the B-ultrasound, CT and MRI image sample data respectively; use a multimodal image registration algorithm (such as a registration algorithm based on feature points, contours or mutual information) to register the B-ultrasound, CT and MRI image data under each modality to obtain registered image data, which is used to solve the problems of inconsistent grayscale and large resolution differences between images caused by different imaging principles, and adjust the image position through an optimization algorithm (such as iterative closest point ICP, gradient descent method, etc.) to achieve high-precision alignment of image data at different time points and imaging modes; according to the registration result, based on an image fusion algorithm (such as weighted fusion, multi-resolution fusion, etc.), fuse the information of B-ultrasound, CT, and MRI medical image modalities, thereby organically combining the advantages of B-ultrasound real-time, CT density information, and MRI tissue structure details, and then construct a liver and tumor three-dimensional model containing rich information; specifically, the fused image data can be converted into an intuitive three-dimensional model by using three-dimensional reconstruction and rendering technology, which is convenient for doctors to perform surgery. The user interaction module 160 is also used to provide the three-dimensional model of the liver and tumor to the surgeon through an interactive interface, and supports rotation, scaling, and cutting operations, so that the surgeon can observe the lesion from different angles.

[0054] Furthermore, the cryoablation treatment execution module is also integrated with a temperature monitoring and feedback mechanism to monitor the temperature changes in the treatment area in real time, ensure that the cryoablation process is carried out within a safe range, and adjust the treatment parameters as needed.

[0055] Furthermore, the machine learning model structure and training are as follows Figure 2 As shown in the figure, the deep learning model uses a convolutional neural network (CNN) or its variants, and is trained through a series of ultrasound image data sets annotated with ablation needle tip position information to improve the accuracy and robustness of needle tip positioning. The machine learning model includes a feature extraction layer, a maximum pooling layer, an attention mechanism layer, a needle tip positioning layer, and a fully connected layer. The accuracy of needle tip positioning is improved through convolutional neural networks and attention mechanisms. Among them,

[0056] The feature extraction layer uses multiple convolution kernels to extract the spatial local features of the needle tip and the surrounding tissue edge image, and outputs the activated feature vector;

[0057] The maximum pooling layer reduces the size of the feature vector and retains important features, and its input is the activated feature vector output by the feature extraction layer, and its output is the downsampled feature vector;

[0058] The attention mechanism layer enhances the feature representation capability by assigning higher weights to important features. Its input is the pooled feature vector and its output is the weighted feature vector.

[0059] The needle tip positioning layer generates position information by detecting and locating the position of the needle tip in real time. Its input is the weighted feature vector and its output is the predicted bounding box.

[0060] The fully connected layer outputs the predicted value by integrating the features. Its input is the predicted bounding box feature vector output from the needle tip positioning layer, and its output is the spatial position coordinates of the needle tip.

[0061] Preferably, in some embodiments, the data processing and calculation process of each layer of the machine learning model is as follows:

[0062] The feature extraction layer uses multiple convolution kernels to extract the spatial local features of the image, such as the edge image of the needle tip and the surrounding tissue. The input is the original ultrasound image (shape H×W×C, where H is the height, W is the width, and C is the number of channels). The output is the activated feature vector (shape H ′ ×W ′ ×K, where K is the number of convolution kernels). The activation function is ReLU, which introduces nonlinearity to improve the model's expressiveness and avoid gradient vanishing. The activation function is σ(x)=max(0,x), where x is the output value of the convolution layer. If x is less than 0, 0 is output, otherwise x is output.

[0063] The formula for convolution feature extraction is:

[0064]

[0065] Parameter meaning:

[0066] ·F i,j,k : The value of the kth eigenvector at position (i, j)

[0067] I: Input image

[0068] ·W m,n,c,k : The weight of the kth convolution kernel at position (m,n) and channel c

[0069] b k : Bias of the kth convolution kernel

[0070] σ: activation function

[0071] M, N: height and width of the convolution kernel

[0072] C: number of channels of the input image

[0073] ·W m,n,c,k : The weight of the convolution kernel, size is M×N×C×K

[0074] K: the number of convolution kernels (i.e. the number of channels of the output feature vector)

[0075] The maximum pooling layer reduces the size of the feature vector, retains important features, and reduces overfitting. The input is the activated feature vector (shape is H′×W′×K), and the output is the downsampled feature vector (shape is H″×W″×K).

[0076] formula:

[0077]

[0078] ·P i,j,k : The value of the pooled feature vector at position (i, j)

[0079] A: The activated feature vector, with a shape of H′×W′×K

[0080] p,q: height and width of the pooling window

[0081] ·s m ,s n : The step size of the pooling window in height and width

[0082] Dimension changes:

[0083]

[0084] H″: the height of the feature vector after pooling

[0085] W″: width of the feature vector after pooling

[0086] The attention mechanism layer enhances the feature representation capability by assigning higher weights to important features. The input is a pooled feature vector (with a shape of H″×W″×K), and the output is a weighted feature vector (with a shape of H″×W″×K)

[0087] formula:

[0088] Linear transformation: transform the input feature vector into query matrix Q, key matrix K and value matrix V:

[0089] Q=W Q X,K=W K X,V=W V X

[0090] ·W Q ,W K ,W V : weights of query, key and value matrices, shape d model ×d k

[0091] X: input feature vector, shape is H″×W″×K

[0092] Q, K, V: query, key, and value matrices, with shape H″×W″×d k

[0093] Calculate attention weights: Calculate attention weights by scaling the dot product:

[0094]

[0095] · The scaled dot product attention score matrix is ​​of shape H″×W″×W″

[0096] Softmax: Softmax function, used to normalize weights to ensure that the sum of weights is 1

[0097] V: value matrix

[0098] Weighted feature vector: Use the attention weight weighting matrix to get the weighted feature vector Z:

[0099] Z = Attention(Q,K,V)

[0100] Z: weighted feature vector, shape is H″×W″×d k

[0101] The needle tip positioning layer generates position information by detecting and locating the position of the needle tip in real time. The input is a weighted feature vector (with a shape of H″×W″×K), and the output is a predicted bounding box (with a shape of N×d, where N is the number of samples and d is the feature dimension).

[0102] Predict bounding box parameters and class probabilities at each cell of the feature vector:

[0103]

[0104] C x,y =Softmax(S x,y )

[0105] ·B x,y : BoundingBoxes at (x,y)

[0106] ·t x ,t y ,t w ,t h : Bounding box offset and scaling parameters predicted by the network

[0107] ·c x ,c y : Offset of grid cells

[0108] ·p w ,p h: The width and height of the prediction box

[0109] σ: Sigmoid function, ensuring that the center point of the bounding box is within the grid cell

[0110] ·S x,y : Classification score

[0111] ·C x,y : Category probability distribution

[0112] The optimizer uses the AdamW optimizer to deal with the overfitting problem. The optimizer provides adaptive learning rate, momentum optimization, and weight decay regularization. The AdamW formula is:

[0113] m t =β1m t-1 +(1-β1)g t

[0114]

[0115]

[0116] Parameter meaning:

[0117] ·m t : First moment estimate

[0118] ·v t : Second moment estimate

[0119] ·g t :gradient

[0120] β1,β2: Hyperparameters used to control the decay rate of the first-order and second-order moment estimates

[0121] · Bias-corrected moment estimates

[0122] ·θ t : parameter vector

[0123] η: learning rate

[0124] ∈: small value, to prevent division by zero

[0125] λ: weight decay parameter

[0126] The model is trained using a triplet loss function to optimize the alignment between the predicted spatial coordinates and the true spatial coordinates. It consists of three main parts: bounding box regression loss, confidence loss, and classification loss. The bounding box regression loss is used to evaluate the difference between the predicted bounding box and the true bounding box, including the errors of the center point coordinates and width and height:

[0127]

[0128] · Bounding Box Regression Loss

[0129] ·λ coord : Weight factor for bounding box regression loss

[0130] S: grid size

[0131] B: Number of bounding boxes predicted per grid cell

[0132] · Indicates whether the jth bounding box of the i-th grid cell contains an object

[0133] ·(x i ,y i ,w i ,h i ): ground truth bounding box parameters

[0134] · Predicted bounding box parameters

[0135] The confidence loss is used to evaluate the difference between the predicted confidence (i.e., whether there is an object in the bounding box) and the actual situation:

[0136]

[0137] · Confidence loss

[0138] ·λ noobj : Weight factor for no-object confidence loss

[0139] ·C i : True confidence (1 if an object exists, 0 otherwise)

[0140] · Prediction Confidence

[0141] The classification loss is used to evaluate the difference between the predicted class probability distribution and the true class:

[0142]

[0143] · Classification Loss

[0144] ·p i (c): True classification probability (1 if the object belongs to category c, otherwise 0)

[0145] · Predicted Classification Probability

[0146] The fully connected layer outputs the predicted value by integrating the features. Its input is the feature vector output from the needle tip positioning layer (shape is N×d, where N is the number of samples and d is the feature dimension), and its output is the spatial position coordinates of the needle tip (shape is N×3, i.e. (x, y, z)).

[0147] formula:

[0148] y=Wx+b

[0149] y: output vector (the position coordinate of the needle tip)

[0150] W: weight matrix (shape is d×3)

[0151] x: Input feature quantity (shape is N×d)

[0152] b: bias vector (shape 1×3)

[0153] The working principle of the system of the present invention is:

[0154] When in use, the patient first lies on the lying board with the patient's abdomen facing the side of the B-ultrasound scanning device. When the patient is ready, the medical staff controls the B-ultrasound scanning device through the B-ultrasound system to perform ultrasound detection on the patient. The system collects ultrasound images generated in real time during the operation and performs image preprocessing. It can extract clear tissue boundaries from complex ultrasound images. The main purpose of preprocessing is to provide high-quality input images for subsequent image analysis and needle tip positioning, thereby improving the overall performance of the algorithm. After the image processing is completed, it enters the needle tip positioning stage based on machine learning. In this stage, a deep learning model is trained, which can accurately identify and predict the position of the cryoablation needle in the liver. The surgeon plans the insertion path of the ablation needle based on the results of the B-ultrasound scan, then marks the insertion, and then adjusts the insertion angle. During the operation, the ablation needle is inserted through the positioning tube, and the cryoablation needle is controlled to implement cryoablation treatment based on the spatial position of the needle tip fed back by the machine learning model. High-pressure argon gas expands rapidly to absorb the surrounding heat, causing the temperature of the surrounding tissue to drop rapidly, leading to the formation of ice crystals inside and outside the tumor cells, cell rupture, and damage to the intima of the tumor venules and arterioles, and thrombosis. Tumor cells die from ischemia and hypoxia.

[0155] The above is an introduction to the system embodiment. The following is a method embodiment to further illustrate the solution of the present invention.

[0156] Figure 3 FIG. 2 is a flow chart of an intelligent navigation method 200 for liver cancer cryoablation based on artificial intelligence according to an embodiment of the present disclosure. Figure 3 As shown, an intelligent navigation method 200 for liver cancer cryoablation based on artificial intelligence comprises:

[0157] S210: collecting original image data of the liver cancer patient in real time under the guidance of the B-ultrasound scanning device;

[0158] S220: performing preprocessing operations on the collected original image to extract clear tissue boundaries;

[0159] S230: performing feature segmentation on the processed image based on a deep learning algorithm, and accurately predicting and locating the spatial position of the cryoablation needle tip in the patient's liver;

[0160] S240: Based on the needle tip positioning results and the instructions input by medical staff, the optimal ablation needle penetration path and angle are planned, and real-time navigation information is provided to ensure that the ablation needle accurately reaches the center of the tumor;

[0161] S250: Based on the guidance of the path planning and navigation module, high-pressure argon gas is controlled to expand rapidly to achieve cryoablation treatment, thereby inactivating tumor cells by lowering the temperature of surrounding tissues to below freezing point.

[0162] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described steps can refer to the corresponding process in the aforementioned system embodiment, and will not be repeated here.

[0163] In order to verify the advantages of the intelligent navigation method for liver cancer cryoablation based on artificial intelligence provided by the embodiment of the present invention over the traditional convolutional neural network, the hardware environment shown in Table 1 and the software environment shown in Table 2 were built, and the experimental parameters shown in Table 3 were set for comparative experiments. The comparative experimental results are shown in Table 4. The ablation efficacy was evaluated by indicators such as complete remission rate (CR) and local tumor progression rate (LTP), and the placement accuracy from the needle tip to the target center, the placement time (from skin puncture to the final position) and the operation completion time were compared. The experiment was conducted on a human model (CIRS057A), and 24 needle placements were independently completed by six interventional radiologists, divided into three needle guidance groups: the present invention, US-based fusion, and CT-guided free hand.

[0164] Table 1 Hardware environment

[0165] hardware Specification processor Intel Core i9-11900K, 3.5GHz Memory 64GB DDR4 RAM Graphics Processor NVIDIA GeForce RTX3090, 24GB VRAM storage 2TB NVMe SSD Robot Operating System ROS (Robot Operating System) Ultrasound equipment High-resolution ultrasound imaging equipment Respiratory monitoring equipment High-precision breathing monitor 3D display device Microsoft HoloLens holographic glasses Model Human Model (CIRS057A)

[0166] Table 2 Software environment

[0167] software Version operating system Ubuntu 20.04LTS Deep Learning Frameworks TensorFlow 2.5.0,PyTorch 1.9.0 Image processing libraries OpenCV4.5.3 Numerical computing library NumPy 1.21.1, SciPy 1.7.0 Visualization Tools Matplotlib 3.4.2, Seaborn 0.11.1 Robot Operating System Framework ROS Noetic Ninjemys

[0168] Table 3 Experimental parameters

[0169]

[0170]

[0171] Table 4 Comparative experimental results

[0172] index The present invention US-based Fusion CT-guided free-hand Complete remission rate (CR) 100％ 100％ 100％ Local tumor progression (LTP) Low middle middle Placement accuracy high middle middle Placement time (minutes) Shortest (average 6 minutes) Medium (average 12 minutes) Longest (average 15 minutes) Operation completion time (minutes) Shortest (average 28 minutes) Moderate (average 41 minutes) Longest (average 49 minutes)

[0173] As shown in Table 4, the present invention performs well in terms of needle placement accuracy, placement time, and surgery completion time, and outperforms the other two methods in the main indicators of ablation efficacy (such as complete remission rate, disease-free survival rate, and overall survival rate).

[0174] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0175] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence is characterized by: include: An image acquisition module, used for acquiring original ultrasound image data of liver cancer patients in real time under the guidance of a B-ultrasound scanning device; An image preprocessing module is used to perform preprocessing operations on the collected original ultrasound images to extract clear tissue boundaries; A machine learning model for feature segmentation of processed images based on a deep learning algorithm and for accurate prediction and localization of the spatial position of the cryoablation needle tip within the patient’s liver; The path navigation module plans the optimal ablation needle insertion path and angle based on the needle tip positioning results and the instructions input by medical staff, and provides real-time navigation information to ensure that the ablation needle accurately reaches the center of the tumor; The treatment execution module, based on the guidance of the path planning and navigation module, controls the rapid expansion of high-pressure argon gas to achieve cryoablation treatment, thereby inactivating tumor cells by lowering the temperature of surrounding tissues to below freezing point.

2. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 1, characterized in that: Also includes: User interaction module; The user interaction module is used to display the ultrasound image and the predicted needle tip position, the insertion path planning and navigation information in real time, and receive the operation instructions of the medical staff.

3. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 2, characterized in that: Also includes: A three-dimensional model construction module is used to collect and organize the patient's B-ultrasound, CT and MRI image sample data; pre-process the B-ultrasound, CT and MRI image sample data respectively; register the B-ultrasound, CT and MRI image data under each modality using a multimodal image registration algorithm to obtain registered image data; according to the registration result, based on the image fusion algorithm, fuse the information of the B-ultrasound, CT and MRI medical image modalities, and then construct a liver and tumor three-dimensional model containing rich information; The user interaction module is also used to provide the liver and tumor three-dimensional model to medical staff through an interactive interface, and supports rotation, scaling, and cutting operations.

4. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 1, characterized in that: The image preprocessing module also includes filtering, denoising and edge detection algorithms.

5. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 1, characterized in that: The cryoablation treatment execution module is also integrated with a temperature monitoring and feedback mechanism to monitor the temperature changes in the treatment area in real time, ensure that the cryoablation process is carried out within a safe range, and adjust the treatment parameters as needed.

6. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 1, characterized in that: The deep learning model adopts a convolutional neural network (CNN) or its variants and is trained through a series of ultrasound image data sets annotated with ablation needle tip position information; the machine learning model includes a feature extraction layer, a maximum pooling layer, an attention mechanism layer, a needle tip positioning layer and a fully connected layer.

7. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 6, characterized in that: in, The feature extraction layer uses multiple convolution kernels to extract spatial local features of the needle tip and the surrounding tissue edge image, and outputs an activated feature vector; The maximum pooling layer reduces the size of the feature vector and retains important features, and its input is the activated feature vector output by the feature extraction layer, and its output is the downsampled feature vector; The attention mechanism layer enhances the feature representation capability by assigning higher weights to important features, and its input is the pooled feature vector and its output is the weighted feature vector; The needle tip positioning layer generates position information by detecting and locating the position of the needle tip in real time, and its input is a weighted feature vector and its output is a predicted bounding box; The fully connected layer outputs a prediction value by integrating features, and its input is the predicted bounding box feature vector output from the needle tip positioning layer, and its output is the spatial position coordinates of the needle tip.

8. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 7, characterized in that: in, The tip localization layer predicts bounding box parameters and class probabilities at each cell of the feature vector: C x，y =Softmax(S x，y ) Among them, B x，y represents the prediction box at coordinates (x, y); t x , t y , t w , t h represents the bounding box offset and scaling parameters predicted by the network; c x , c y Indicates the offset of the grid unit; p w , p h Represents the width and height of the prediction box; σ: Sigmoid function, ensuring that the center point of the bounding box is within the grid cell; S x，y represents the classification score; S x，y Represents the class probability distribution.

9. The intelligent navigation system for liver cancer cryoablation based on artificial intelligence according to claim 1, characterized in that: Also included are optimizers and triplet loss functions for training machine learning models to optimize the alignment between predicted and true space coordinates; Among them, the optimizer adopts the AdamW optimizer, which provides adaptive learning rate, momentum optimization and weight decay regularization; Triple loss function in, is the bounding box regression loss, is the confidence loss, is the classification loss.

10. An intelligent navigation method for liver cancer cryoablation based on artificial intelligence, characterized in that: include: Under the guidance of B-ultrasound scanning device, the original ultrasound image data of liver cancer patients is collected in real time; Perform preprocessing operations on the collected original ultrasound images to extract clear tissue boundaries; The processed images are segmented based on deep learning algorithms, and the spatial position of the cryoablation needle tip in the patient's liver is accurately predicted and located; Based on the needle tip positioning results and the instructions input by medical staff, the optimal ablation needle penetration path and angle are planned, and real-time navigation information is provided to ensure that the ablation needle accurately reaches the center of the tumor; Under the guidance of the path planning and navigation module, high-pressure argon gas is controlled to expand rapidly to achieve cryoablation treatment, which inactivates tumor cells by lowering the temperature of surrounding tissues to below freezing point.

Images