A Focused Ultrasound Therapy System with Associated Positioning and Its Usage Method

Abstract

The present invention belongs to the technical field of surgical treatment, and relates to a focused ultrasound treatment system with associated positioning and a usage method. The system includes a focused ultrasound treatment device called a local system. The focused ultrasound treatment device at least includes a guiding device and a doctor's workstation. The doctor's workstation imports diagnostic results for the training of a local deep learning model to obtain an image template; and then, relying on the image template, several images transmitted by the guiding device in the focused ultrasound treatment device are matched and recognized, and images with a recognition probability greater than a threshold are obtained. The coordinates of the obtained images are the lesion positions. The system also includes a remote deep computing server, in which a remote deep learning model is constructed to identify the lesion positions based on the remote deep learning model, and one or more images with a relatively high lesion recognition probability are obtained; the system performs associated secondary recognition and positioning in combination with AI on the basis of clinical diagnosis, reduces the manual intervention time, greatly shortens the speed of doctors to find the target, and improves the treatment efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of surgical treatment, and relates to a correlation-positioned focused ultrasound treatment system and a use method thereof. Background Art

[0002] High-intensity ultrasound therapy is a non-invasive treatment that uses the focusability and penetration of ultrasound to act on diseased tissues. By using the characteristics of ultrasound that can penetrate human tissue and focus inside human tissue, multiple ultrasound beams are generated outside the body, and then the ultrasound beams are focused on the human lesions for treatment. In recent years, it has been widely used in the clinical treatment of many diseases.

[0003] The prior art discloses a focused ultrasound treatment system, including an ultrasound transmitting device, a signal acquisition device, an imaging device, a central processing device and a feedback control device. The ultrasound transmitting device transmits focused ultrasound to the target area or target site, and the signal acquisition device and the imaging device cooperate with the central processing device and the feedback control device to realize the positioning of the target site, adjust the relative position of the focused ultrasound focus and the target site in real time, realize better tracking of the treatment target, and adjust the ultrasound treatment dose in real time to improve the accuracy of treatment positioning and target tracking capabilities, and ensure the accuracy and stability of the dose required for treatment. In addition, during the treatment process, the temperature change parameters can also be determined, and the feedback control device adjusts the working parameters of the ultrasound transmitting device according to the echo and temperature change parameters, thereby effectively avoiding excessive local tissue temperature of the patient, improving the efficiency of treatment, and reducing the patient's pain and side effects.

[0004] There is also a disclosed remote high-intensity focused ultrasound treatment system, which includes an ultrasonic transducer motion control device, a motion control server, a B-ultrasound machine, a medical imaging server, a doctor's workstation and a nurse's workstation. The doctor's workstation and the nurse's workstation are connected to the motion control server and the medical imaging server through a TCP / IP network, and the status reading and the control of the surgical process are realized by sending instructions to the server. The medical imaging server is connected to the B-ultrasound machine, and the B-ultrasound image is digitized and sent to the doctor's workstation. The motion control server is connected to the ultrasonic transducer motion control device, and controls the movement of the ultrasonic transducer and the ultrasonic emission according to the instructions issued by the doctor's workstation. The system allows multiple doctor's workstations to be connected to the system using a TCP / IP network to realize remote consultation and remote.

[0005] Existing focused ultrasound treatment systems mainly use methods such as MRI guidance or ultrasound guidance for lesion localization. Based on the images from MRI or ultrasound, the equipment operator (usually a doctor) identifies and locates the lesion site. The operator directly operates or generates an instruction sequence according to the located lesion site to complete the treatment of the lesion site. Whether it is MRI guidance or ultrasound guidance, the final determination of the lesion site, that is, the treatment area, is determined by the operator. Due to the different experiences and levels of different operators, it is difficult to ensure that all the patient's lesion sites are correctly identified. Once the positioning is deviated or even incorrect, it cannot be guaranteed that the treated area completely covers the lesion, and in severe cases, it may lead to treatment accidents. In addition, the memory ability of the operator is limited and cannot complete the memory of dozens or even hundreds of images. Therefore, the efficiency of manual identification is extremely low.

[0006] For the guiding device of existing ultrasonic diagnostic equipment, it is impossible to have both a detection distance and image clarity. In high-intensity ultrasonic treatment equipment, once the treatment head is determined, its general usage range is also determined, usually reaching 10 - 40 cm. At this detection depth, the clarity of the ultrasonic probe is poor.

[0007] Existing MRI- and CT-guided high-intensity ultrasonic treatment equipment: Generally speaking, it has a relatively large volume and poor application flexibility. For equipment such as CT that contains ionizing radiation, it is impossible to achieve real-time communication between the equipment operator and the patient during use. MRI has relatively strict requirements for the use environment. The changes in the patient's body position and the ultrasonic coupling environment during treatment and examination lead to changes in the image morphology, quality, and position, resulting in positioning difficulties. The changes in the body position and the ultrasonic coupling environment are the main factors causing changes in the image morphology, quality, and position, but the relative position, size, and shape of the lesions in the patient's body have not actually changed significantly or have regular changes.

[0008] AI image recognition has been very mature in other fields such as security and autonomous driving. In recent years, AI recognition has gradually been introduced into the direction of medical auxiliary diagnosis. In the existing focused ultrasound treatment systems that use AI technology, the adopted solutions are all to directly identify the patient's lesions and organs through AI and complete the lesion delineation operation. Such AI diagnosis almost only exists under ideal conditions, and in the face of an actual very complex treatment environment, the recognition error rate is high. Therefore, the application of AI technology in this direction is not mature. Summary of the Invention

[0009] The purpose of the present invention is to provide a focused ultrasound treatment system with associated positioning and its usage method. Aiming at the fact that existing ultrasonic transducers only provide piezoelectric conversion of power ultrasound, but the ultrasonic treatment head needs to be combined with position measurement and integrated data synchronization to achieve effective treatment, it combines precise diagnostic means and AI recognition technology to achieve high-intensity, precise, and efficient focused ultrasound treatment.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] On the one hand, the present invention proposes a focused ultrasound therapy system for associated positioning, including a focused ultrasound therapy device called a local system. The focused ultrasound therapy device at least includes a guiding device and a doctor workstation. The doctor workstation includes a computing server. The doctor workstation imports the diagnostic results of a clinical diagnostic device and performs training based on a local deep learning model in the computing server to obtain an image template. Then, relying on the image template, a number of images transmitted by the guiding device in the focused ultrasound therapy device are matched and recognized to obtain images with a recognition probability greater than a threshold value, and the image coordinates obtained are the lesion positions.

[0012] As a possible implementation, the system further includes a remote deep computing server relying on a remote deep learning model. The remote deep computing server communicates with the doctor workstation through a network. The doctor workstation imports the diagnostic results and transmits them to the remote deep computing server for training of the remote deep learning model. The doctor workstation preprocesses a number of images transmitted by the guiding device and then transmits them to the remote deep computing server. The remote deep computing server performs training based on the remote deep learning model to obtain an image template. Then, relying on the image template, a number of images transmitted by the guiding device in the focused ultrasound therapy device are matched and recognized to obtain images with a recognition probability greater than a threshold value. The lesion area is determined based on the images, and then the lesion is treated and the remote deep learning model is optimized. The image with a recognition probability greater than the threshold value refers to the image corresponding to the coincidence of the center of the circumscribed ellipsoid of the lesion area and the center of the field of view of the guiding device on the premise that the similarity is greater than the threshold value. The image coordinates are the coordinate values of the position where the guiding device is located when the image is acquired.

[0013] As a possible implementation, after the remote system completes a period of interactive training, it regularly releases new versions of the remote deep learning model and the corresponding local deep learning model and updates them to the local system.

[0014] As a possible implementation, the preprocessing at least includes clipping, compression, enhancement, and feature extraction relying on a preprocessing model. The preprocessing model at least includes a CNN model and a Transformer model.

[0015] As a possible implementation, the CNN model includes SqueezeNet, GoogleNet, ShuffleNet, MobileNetV2, NasNetMobile, and GoogleNet models, and the Transformer model at least includes ViT, CaiT, and DeiT networks.

[0016] As a possible implementation, the number of images with an identification probability greater than the identification threshold is greater than or equal to 1 and less than or equal to 20; the identification threshold is greater than or equal to 80% and less than or equal to 1.

[0017] As a possible implementation, the system includes multiple local systems communicating with a remote depth computing server. The remote depth computing server receives data sent by each doctor workstation, trains and optimizes the remote deep learning model. Specifically: the data sent by each doctor workstation is sent into the remote deep learning model for training, and the training accuracy of the current round is statistically calculated. If the difference between this training accuracy and the training accuracy stored in the remote depth computing server is greater than or equal to the reduction threshold, the data sent by the current doctor workstation is discarded; otherwise, it is retained.

[0018] As a possible implementation, the focused ultrasound treatment system further includes a nurse workstation, a treatment bed, a monitoring and protection unit, a water treatment unit, a motion control unit, an ultrasonic power unit, and an ultrasonic transducer. The doctor workstation transmits data to the nurse workstation and the treatment bed respectively, and the doctor workstation sends control signals to the water treatment unit, the motion control unit, and the ultrasonic power unit; the functions performed by the local system further include establishing a patient case file, importing the patient's pre-treatment examination report data into the system, patient preprocessing, generating a treatment plan, and completing the treatment.

[0019] On the other hand, the present invention proposes a method for using an associated positioning focused ultrasound treatment system. This method utilizes the local system as described above. The focused ultrasound treatment device in it receives the diagnostic results from the clinical diagnostic device and performs calculations on the local deep learning model. The guiding device performs a pre-scan of M×N on the lesion site, where M is the number of scan lines and N is the number of scan columns, to obtain M×N pre-scan images of the lesion. Then, based on the local deep learning model, identification and search are performed among the M×N pre-scan images to generate a matching result, obtaining one or more images with a higher lesion identification probability, outputting the coordinate values of the corresponding images, and then judging the lesion site based on the coordinate values and performing treatment.

[0020] As an implementation method, a method for using an associated positioning focused ultrasound treatment system uses the local system as described above. The doctor workstation receives the diagnosis results and transmits them to the remote depth computing server for training of the remote deep learning model. The guiding device performs a pre-scan of M×N on the lesion site, where M is the number of scan lines and N is the number of scan columns, to obtain M×N pre-scan images of the lesion. The doctor workstation preprocesses and transmits a number of images transmitted by the guiding device to the remote depth computing server. The remote depth computing server trains based on the remote deep learning model to obtain an image template; then, relying on the image template, it matches and identifies a number of images transmitted by the guiding device in the focused ultrasound treatment system, obtains images with an identification probability greater than the threshold, and the obtained image coordinates are the lesion positions. After determining the lesion positions, treatment is carried out while optimizing the remote deep learning model and synchronizing the local deep learning model.

[0021] For the training of the local deep learning model and the remote deep learning model, the purpose of the training is to establish the correlation before and after changes in the body position and the ultrasound coupling environment, and to match the lesion sites before and after the changes.

[0022] Beneficial effects:

[0023] An associated positioning focused ultrasound treatment system and a using method proposed by the present invention have the following beneficial effects compared with the prior art:

[0024] 1. Different from the existing focused ultrasound treatment system that directly treats after positioning through the guiding device or directly uses AI to diagnose (identify) and treat the lesion site, this system combines AI for associated secondary identification and positioning on the basis of clinical diagnosis, reduces the time of manual intervention, greatly shortens the time for doctors to find the target tissue, and improves the speed of locating the lesion and the treatment efficiency.

[0025] 2. Based on prior knowledge, this system can continuously improve the matching degree and treatment effect of the treatment system through AI learning, and by optimizing the usage mode of AI, it exchanges for the practicability of the treatment process.

[0026] 3. The local system can work in a fully offline state, reduces the dependence on system resources through the local deep learning model, and can perform AI calculations locally to accelerate the doctor's positioning of the lesion site.

[0027] 4. The remote system can simultaneously process the computing requirements from various different local systems, further strengthen the applicability of the training model, improve the accuracy of finding the target lesion site, and jointly complete the optimization of the training model with local systems distributed globally. Description of the drawings

[0028] The accompanying drawings described herein are used to provide a further understanding of the present invention and form a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0029] Figure 1 It is a schematic diagram of the composition of the associated positioning focused ultrasound therapy system for the first embodiment;

[0030] Figure 2 It is a schematic diagram of the control relationship of the monitoring and protection unit;

[0031] Figure 3 It is a schematic diagram of the composition of the associated positioning focused ultrasound therapy system for the second embodiment;

[0032] Figure 4 It is a schematic diagram of the composition of the associated positioning focused ultrasound therapy system for the third embodiment.

[0033] Illustration description:

[0034] 1 - Doctor workstation; 2 - Nurse workstation; 3 - Treatment bed; 4 - Monitoring and protection unit; 5 - Water treatment unit; 6 - Guiding device; 7 - Motion control unit; 8 - Ultrasound power unit; 9 - Ultrasound transducer; 10 - Clinical diagnostic device; 11 - Remote depth calculation server. Detailed implementation manners

[0035] It should be noted that in the present invention, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0036] In a first aspect, the present invention provides a focused ultrasound therapy system for associated positioning, including a focused ultrasound therapy device referred to as a local system. The focused ultrasound therapy device at least includes a guiding device and a doctor workstation. The doctor workstation imports diagnostic results for training a local deep learning model to obtain an image template. Then, relying on the image template, a number of images transmitted by the guiding device in the focused ultrasound therapy device are matched and recognized to obtain images with a recognition probability greater than a threshold value, and the obtained image coordinates are the lesion positions. The system further includes a remote deep computing server in which a remote deep learning model is constructed. Based on the remote deep learning model, the lesion position is recognized to obtain one or more images with a relatively high lesion recognition probability. Different from the existing focused ultrasound therapy system that directly treats after positioning through the guiding device or directly uses AI to diagnose (recognize) and treat the lesion site, the system combines AI for associated secondary recognition and positioning on the basis of clinical diagnosis, reduces the manual intervention time, greatly shortens the time for doctors to find the target, improves the speed of positioning the tissue, and improves the treatment efficiency.

[0037] See Figure 1 , the focused ultrasound therapy device is a general term for 1-9. This device at least includes a guiding device 6 and a doctor workstation 1. The doctor workstation 1 includes a computing server. The doctor workstation 1 imports the diagnostic results of the clinical diagnostic device 10 and trains based on the local deep learning model in the computing server to obtain an image template. Then, relying on the image template, a number of images transmitted by the guiding device 6 in the focused ultrasound therapy device are matched and recognized to obtain images with a recognition probability greater than a threshold value, and the obtained image coordinates are the lesion positions.

[0038] The described focused ultrasound therapy system further includes a remote deep computing server relying on a remote deep learning model. The remote deep computing server communicates with the doctor workstation 1 through a network. The doctor workstation 1 imports diagnostic results and transmits them to the remote deep computing server for training the remote deep learning model. The doctor workstation 1 preprocesses a number of images transmitted by the guiding device 6 and then transmits them to the remote deep computing server. The remote deep computing server trains based on the remote deep learning model to obtain an image template. Then, relying on the image template, a number of images transmitted by the guiding device 6 in the focused ultrasound therapy device are matched and recognized to obtain images with a recognition probability greater than a threshold value. The lesion area is determined relying on the image, and then the lesion is treated and the remote deep learning model is optimized. The image with a recognition probability greater than a threshold value refers to the image corresponding to when the center of the circumscribed ellipsoid of the lesion area coincides with the center of the guiding device's field of view on the premise that the similarity is greater than the threshold value. The image coordinates are the coordinate values of the position where the guiding device is located when the image is acquired.

[0039] In specific implementation, the doctor workstation in the focused ultrasound treatment device includes a central AI computer, which receives the diagnostic results from the clinical diagnostic device 10 and is used for training the local deep learning model. The central AI computer matches and identifies a number of images sent by the guiding device 6 in the focused ultrasound treatment system based on the local deep learning model, obtains one or more images with a higher lesion recognition probability, and determines the lesion location for treatment based on these images.

[0040] Therefore, based on prior knowledge, the focused ultrasound treatment system provided by this solution can continuously improve the matching degree and treatment effect of the treatment system through AI lesion recognition. By optimizing the AI lesion recognition method, the practicability of the treatment process is exchanged. Compared with the auxiliary diagnosis of uncertain targets after extracting the positioning images and lesion features adopted by the existing focused ultrasound treatment systems, it has the advantages of clear features, sufficient reference, and determined position.

[0041] In specific implementation, the AI learning is realized through a local deep learning model (located on the central AI computer of the doctor workstation) and a remote deep learning model (located on the remote deep computing server in the remote system).

[0042] Computers have the ability to process abstract data more suitable than humans. Therefore, AI lesion recognition will bring more accurate positioning, thus ensuring complete coverage of the treatment range and further ensuring the treatment effect.

[0043] The number of images with a recognition probability greater than the recognition threshold is greater than or equal to 1 and less than or equal to 20. Usually, in specific implementation, the number of images with a recognition probability greater than the recognition threshold is 3 - 5. There are also more or fewer cases, such as: 1, 2, or more than a dozen. The situation of more than 20 is almost non-existent. The threshold is greater than or equal to 80% and less than or equal to 1. In specific implementation, the threshold is set to 85%. It can also be set to 80% (for the case where the number of images with a recognition probability greater than the recognition threshold is small) or 95% (for the case where the number of images with a recognition probability greater than the recognition threshold is large, the threshold is increased).

[0044] On the other hand, the present invention also provides a method for using an associated positioning focused ultrasound treatment system. This method utilizes a local system and a remote system. The doctor workstation receives the diagnostic results from the clinical diagnostic device and transmits them to the remote deep computing server for the training of the remote deep learning model. The guiding device performs a pre-scan of M×N on the patient's lesion site, where M is the number of scan lines and N is the number of scan columns, to obtain M×N pre-scan images of the patient's lesion. The doctor workstation collects and preprocesses the images presented by the guiding device and then transmits them to the remote deep computing server. The remote deep computing server identifies the lesion location based on the remote deep learning model, obtains one or more images with a higher lesion recognition probability, and determines the lesion location for treatment accordingly. At the same time, the remote deep learning model is optimized and the local deep learning model is synchronized.

[0045] Embodiment 1

[0046] As the first embodiment of the present invention, the first associated positioning focused ultrasound treatment system. For the embodiment, refer to Figure 1 , this system is a local system, including the following parts: doctor workstation 1, nurse workstation 2, treatment bed 3, monitoring and protection unit, water treatment unit 5, guiding device 6, motion control unit 7, ultrasonic power unit 8, and ultrasonic transducer 9. Among them, 1-3 are data and instruction processing parts, 4 is the safety guarantee part, and 5-9 are execution parts; 1-9 belong to the local focused ultrasound treatment equipment, and 10 is the local diagnostic device.

[0047] Doctor workstation 1: It is the control center of the focused ultrasound treatment system, including a central AI computer, which controls the generation of instructions for the entire system and is operated interactively by an operator (doctor). The doctor workstation 1 receives the diagnostic results from the clinical diagnostic device 10, and the doctor workstation 1 conducts data transmission with the nurse workstation 2 and the treatment bed 3 respectively. The doctor workstation 1 receives the position signals sent by the guiding device 6, and the doctor workstation 1 sends control signals to the water treatment unit 5, the motion control unit 7, and the ultrasonic power unit 8.

[0048] The information of each part is synchronized here. The doctor workstation 1 is also responsible for the retrieval and calculation of the local deep learning model, identifying the images sent by the guiding device 6, generating a treatment plan and implementing the treatment, as well as the storage and management of treatment cases. The doctor workstation 1 is equipped with a central AI computer, which can use a simplified model to identify the lesion without accessing the remote system.

[0049] The nurse workstation 2 provides an auxiliary operation interface for the ultrasonic treatment equipment, mainly providing the control of the commonly operable functions of the equipment to assist the operator in performing various steps such as positioning and treatment.

[0050] The treatment couch 3 bears the patient, provides an aqueous environment and a treatment platform for the ultrasonic transducer, and provides the commonly used operable function controls of the system, facilitating the operator to perform positioning and treatment.

[0051] The monitoring and protection unit 4, see Figure 2 , realizes the status monitoring function of each part, timely reminds the operator to handle when abnormal parameters are found, and suspends the treatment process when necessary to prevent accidents, mainly including ultrasonic voltage monitoring, movement position monitoring, water treatment water temperature monitoring, water treatment vacuum degree monitoring, water treatment flow monitoring, transducer water shortage protection, transducer water overflow protection; movement position, ultrasonic voltage; vacuum degree (water treatment).

[0052] The water treatment unit 5 prepares degassed water for the ultrasonic coupling path for the patient's treatment, and its main functions are decompression, filtration, degassing, heating, and provides a control interface and parameter display.

[0053] The guiding device 6 realizes the external positioning of the lesion site, transmits the lateral position image to the doctor workstation, and generally can adopt ultrasonic guidance, MRI guidance, CT guidance, etc. The treatment system determines the relative coordinates of the lesion site according to the image of the guiding device.

[0054] The motion control unit 7 is one of the actuators of the ultrasonic treatment equipment, responsible for bearing the ultrasonic transducer, executing the motion instructions generated by the doctor workstation, and controlling the guiding device to move to measure the lesion position, as well as auxiliary operations such as the lifting and rotation of the ultrasonic transducer.

[0055] The ultrasonic power unit 8 is one of the actuators of the ultrasonic treatment equipment, responsible for the main acoustic power output of the system and its signal processing, mainly including an ultrasonic control signal processing unit, an ultrasonic power driving unit, and an ultrasonic transducer matching unit.

[0056] The ultrasonic transducer 9 is the controlled element of the ultrasonic treatment equipment and is the ultrasonic power output unit of the system.

[0057] The clinical diagnostic equipment 10 includes various clinical ultrasonic diagnostic equipment, which is used to examine various organs inside the human body and can clearly display various sectional images and blood flow conditions of each organ and the surrounding organs.

[0058] The usage mode, content and functions of the local system include:

[0059] 1), Establishment of patient case files;

[0060] 2), Import the patient's pre-treatment examination report data into the system;

[0061] 3), Patient pre-treatment: that is, the preparatory work before treatment, anesthesia, body position fixation, etc.;

[0062] 4) AI lesion recognition: Place the treatment site of the patient within the detectable range of the guiding device 6. The real-time image presented by the guiding device 6 is transmitted to the central AI computer. The central AI computer preprocesses the image data. Since the option of not connecting to the remote system is selected, the report data obtained by the clinical diagnosis device 10 before treatment and the preprocessed real-time image of the patient are directly loaded into the local deep learning model on the central AI computer, and the recognition probability of the current image obtained by the guiding device 6 is calculated. The lesion contour with the highest calculated probability is pre-outlined, and at the same time, the lesion outlining areas with other probabilities are listed for comparison. The doctor selects a certain result based on the probability given by the AI and manually adjusts it on this basis to optimize the outlining area.

[0063] The local system can work in a fully offline state. By using the local deep learning model, it reduces the dependence on system resources, enabling local execution of AI calculations to speed up the doctor's positioning of the lesion site, and thus achieving fast AI lesion recognition.

[0064] For the preprocessing, specifically in implementation, an efficient and lightweight CNN model or Transformer model is used; for the CNN model, it includes SqueezeNet, GoogleNet, ShuffleNet, MobileNetV2, NasNetMobile, and GoogleNet models. The Transformer model at least includes ViT, CaiT, and DeiT networks. Tables 1 and 2 respectively show the comparison of the model size, number of parameters, pixels, and recognition accuracy of each CNN model and Transformer model in specific implementation.

[0065] Table 1 Comparison of the size, number of parameters, pixels, and accuracy performance of each CNN model

[0066] CNN model Model size Number of parameters (in millions) Pixel Accuracy SqueezeNet 5.2MB 1.24 227×227 85％ ShuffleNet 5.4MB 1.4 224×224 94.3％ MobileNetV2 13MB 3.5 224×224 95％ NasNetMobile 20MB 5.3 224×224 85.8％ GoogleNet 27MB 7 224×224 95.2％

[0067] The main feature of SqueezeNet is the use of the Fire module to reduce the number of parameters and reduce the model size by compressing the input feature map to a smaller dimension. A 1x1 convolutional kernel is used to further reduce the number of parameters; ShuffleNet introduces channel grouping and depthwise separable convolutions; MobileNetV2 introduces an inverted residual structure and a linear bottleneck. The inverted residual structure and the linear bottleneck make the model deeper and more effective, with better expressive ability and feature extraction ability; NasNetMobile is a network structure obtained through neural architecture search in Auto-ML, with better architecture design and feature extraction ability; GoogleNet uses the Inception module, which contains multiple parallel convolutional kernels (1x1, 3x3, 5x5) and pooling operations, that is, it captures multi-scale features by increasing the width and depth of the network.

[0068] Table 2 Comparison of the size, parameters, input format (pixels), model description and estimated accuracy performance of each Transformer model

[0069]

[0070] 5), Generate a treatment plan: After the doctor finally confirms the contour of the lesion site, the treatment system will complete the 3D reconstruction of the treatment body based on the image data, and retrieve different treatment strategies and treatment parameters according to the shape, size, and depth of the tumor, calculate the complete coordinates of the treatment body, and generate corresponding treatment instructions;

[0071] 6), Complete the treatment: The AI central computer uses the control interface to convey the instructions and coordinates to each unit of the execution part, controls the ultrasonic power output unit to emit an appropriate treatment dose, controls the motion unit to move the transducer to the specified coordinates, and controls the water treatment unit to output coupling water with an appropriate flow rate, temperature, and gas content, and finally completes the scanning treatment of the entire lesion;

[0072] 7), The monitoring and protection unit 4 monitors various instructions and parameters during the treatment process in real time: Once the monitoring results exceed the maximum acceptable threshold compared with the target expectation, operations such as reminder, alarm, or even treatment interruption will be provided according to the severity of the deviation from the expectation.

[0073] Advantages of the local system: The local system uses a high-performance computer as the central AI computer, which can load a pre-trained model - a local deep learning model for fast feature search and recognition. The system can work in a fully offline state and is used by institutions that do not provide an Internet environment or have a certain confidentiality level. The essence of the local system is to reduce the dependence on system resources through the local deep learning model, so as to be able to perform AI calculations locally and speed up the doctor's positioning of the lesion site.

[0074] Although the computing power and model of the local system cannot be compared with those of the remote system, it can also assist the operator in localizing the lesion to a certain extent with minimal resources.

[0075] Specifically, a method for using an associated positioning focused ultrasound treatment system utilizes the above local system. The central AI computer in it receives the diagnostic results from the clinical diagnostic device and performs calculations using the local deep learning model. The guiding device 6 performs a pre-scan of M×N on the patient's lesion site, where M is the number of scan lines and N is the number of scan columns, to obtain M×N pre-scan images of the patient's lesion. The central AI computer performs identification and search in the M×N images based on the local deep learning model and generates a matching result, obtaining one or more images with a higher lesion recognition probability, and then outputs the coordinate values of the corresponding images to assist the operator in judging the lesion site and performing treatment.

[0076] In specific implementation, the values of M and N range from greater than or equal to 2 to less than or equal to 500;

[0077] The pre-scanned image is obtained through pre-scanning. In specific implementation, it is obtained by using the fast scanning method. Starting from the rough positioning point found through pre-scanning, the image is scanned along the X and Y directions respectively, and k scanned images are obtained in each direction. The scanning preferably takes steps of 0.1 mm to 1 mm to obtain the pre-scanned image. The scanning range preferably can be 1.5 to 2.5 times the maximum length of the target lesion or the maximum range of the target lesion plus 20 to 50 mm. If a matching result can be recognized and found within these 2×k images, then re-scanning is performed in the other direction centered on the found target position to obtain n new scanned images;

[0078] The fast scanning method can optimize the pre-scanning speed and improve the positioning efficiency;

[0079] The values of k and n range from greater than or equal to 5 to less than or equal to 500. In specific implementation, they are different according to the size of the lesion tissue.

[0080] Embodiment 2

[0081] As the second embodiment of the present invention, an embodiment of the second associated positioning focused ultrasound treatment system is shown in Figure 3 , on the basis of the previous embodiment, a remote system is added. Specifically, a remote depth calculation server 11 is added, forming an architecture of a local system + remote system. The remote depth calculation server 11 is responsible for collecting the data preprocessed by the doctor workstation 1. The remote system consists of a super high-performance computer or a super high-performance computer cluster. The remote system has sufficient performance and time to complete the resources required for deep learning. Relying on the powerful computing power of the remote high-performance server 11, it is dedicated to model evaluation, training, optimization, and adjustment, generating a remote deep learning model and sending it to the doctor workstation 1 to provide synchronous updates to the local deep learning model. Specifically, the doctor workstation 1 collects and preprocesses the images presented by the guiding device 6. The preprocessing scheme can adopt conventional digital image signal processing technologies such as thresholding, edge detection, image filtering, image morphology operations, and image binarization. In specific implementation, image processing schemes such as OpenCV can be used. The preprocessed image signal is transmitted to the remote system (remote depth calculation server 11). The remote depth calculation server 11 relies on its powerful computing power to calculate by loading a remote deep learning model that has been trained and optimized for a long time, quickly outputs the result of the contour of the lesion site, and provides multiple possibilities in the form of percentage probabilities for the user to select. On the one hand, the remote depth calculation server 11 sends the calculation result back to the doctor workstation, and on the other hand, the data of this calculation will also continue to participate in model training and synchronously pre-train the local deep learning model for the local system.

[0082] After the remote system completes a period of interactive training, it regularly releases new versions of the remote deep learning model and the corresponding local deep learning model and updates them to the local system. The remote system can simultaneously process the computing requirements from various different local systems, further strengthening the applicability of the training model, improving the accuracy of finding the target lesion site, and jointly optimizing the training model with local systems distributed globally.

[0083] The training model is the remote deep learning model located in the remote system.

[0084] The preprocessing at least includes cropping, compression, enhancement, and feature extraction relying on the preprocessing model. Specifically in this embodiment, the preprocessing includes cropping, compression, enhancement, and feature extraction of images; the enhancement is a sharpening operation.

[0085] For feature extraction relying on the preprocessing model, the preprocessing model at least includes a CNN model and a Transformer model. Specifically in implementation, the CNN model is the MobileNetV2 model.

[0086] After the remote system completes a period of interactive training, it regularly releases new versions of the training model and the corresponding local deep learning model. The local system can update the new version of the training model to the local system under the operation of an engineer. The remote system directly updates and uses the model after releasing the new version of the training model.

[0087] The remote system uniquely marks the learning content submitted by the local system. If the subsequent model is determined to be inaccurate by the new learning content, the training contribution weight of the corresponding local system is reduced. If the subsequent model is determined to be accurate by the new learning content, the training contribution weight of the corresponding local system is increased. The use of the training contribution weight is a strategy for system adaptive optimization, aiming to distinguish the correctness of the selection of the lesion site by different device users participating in the training, and eliminate training contamination caused by invalid data, faulty devices, malicious use, and practice use.

[0088] The usage method and implementation functions of the remote model are as follows:

[0089] 1) Establishment of patient case files;

[0090] 2) Import the data of the patient's pre-treatment examination report into the system;

[0091] 3) Patient preprocessing: that is, the preparatory work before treatment, such as anesthesia and body position fixation;

[0092] 4) Lesion localization.

[0093] Place the treatment part required by the patient within the detectable range of the guiding device 6. The real-time image presented by the guiding device 6 is transmitted to the central AI computer. The central AI computer preprocesses the image data and uploads it to the remote deep computing server 11 together with the report data obtained by the clinical diagnosis device 10 before the patient's treatment. The remote deep computing server 11 loads the remotely deep learning model accumulated through long-term training, combines the patient's past data, obtains the current image recognition probability obtained by the guiding device 6, pre-outlines the lesion contour with the highest calculated probability, and simultaneously lists the lesion outlining areas with other probabilities for comparison. The doctor can select a certain result based on the probability given by the AI and manually adjust it on this basis to optimize the outlining area. At the same time, the selection and adjustment here will be transmitted back to the remote system as new learning input;

[0094] 5) Generate a treatment plan: After the doctor finally confirms the lesion site contour, the treatment system will complete the 3D reconstruction of the treatment body based on the image data, retrieve different treatment strategies and treatment parameters according to the shape, size, and depth of the tumor, calculate the complete coordinates of the treatment body, and generate corresponding treatment instructions;

[0095] 6) Complete the treatment: The AI central computer uses the control interface to convey the instructions and coordinates to each unit of the execution part, controls the ultrasonic power unit 8 to emit an appropriate treatment dose, the motion control unit 7 moves the ultrasonic transducer 9 to the specified coordinates, and controls the water treatment unit 5 to output coupling water with an appropriate flow rate, temperature, and gas content, and finally completes the scanning treatment of the entire treatment lesion;

[0096] 7) The monitoring and protection unit 4 monitors various instructions and parameters during the treatment process in real time: Once the monitoring result exceeds the acceptable maximum threshold from the target expectation, operations such as reminder, alarm, or even suspension of treatment will be provided according to the severity of the deviation from the expectation.

[0097] Advantages of the remote system: When the local system and the remote system are used in combination, the local system, as the link that can interact with the system operator (usually a doctor), provides training for the remote system according to the treatment area finally selected by the doctor, corrects, and optimizes the output error of the original model. The ultra-high computing power of the ultra-high performance computer or ultra-high performance computer cluster used by the remote system can improve the model calculation efficiency and reduce the calculation time.

[0098] Model sharing of the associated focused ultrasound treatment system: If the local connection network is unobstructed, it is handed over to the server; use the latest version of the local deep learning model to identify similar images locally. If the features do not change much, even if the recognition is wrong, there is still a doctor to check and proofread; it greatly shortens the speed at which the doctor locates the target and improves the treatment efficiency.

[0099] Example 3

[0100] As the third embodiment of the present invention, a third associated positioning focused ultrasound treatment system embodiment is described as follows. Refer to Figure 4 , this system includes a remote depth calculation server 11 and multiple discrete local systems.

[0101] The remote depth calculation server 11 is responsible for collecting the data preprocessed by each doctor workstation 1. Relying on the powerful computing power of the remote high-performance computer, it is dedicated to model evaluation, training, optimization, and adjustment, generating a remote deep learning model for each doctor workstation 1 to synchronously update the local deep learning model.

[0102] On the other hand, the calculated data will also continue to participate in model training and synchronously pre-train the model - the local deep learning model for all doctor workstations connected to the remote depth calculation server.

[0103] As the cloud center, the remote system connects distributed local systems, can handle the computing requirements from different local systems simultaneously, aggregates the preprocessed data uploaded from different local systems, provides new training content for the remote system, further strengthens the applicability of the training model, and improves the accuracy of finding the target lesion site.

[0104] The essence of the remote system is to utilize ultra-high-performance computing power to achieve more extensive and accurate training, loading, and use of deep training models, and jointly complete the optimization of the training model with various local systems distributed globally.

[0105] All relevant contents of each step involved in the above method embodiment can be cited in the function description of the corresponding functional module, and will not be elaborated here.

[0106] Although the present invention has been described in conjunction with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present invention. Accordingly, this specification and the drawings are merely exemplary illustrations of the present invention defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

[0107] The parts not elaborated in the present invention belong to the well-known technologies in the art. Although the illustrative specific embodiments of the present invention have been described above for the understanding of those skilled in the art of the present technology, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

Claims

1. A focused ultrasound therapy system with associated positioning, characterized in that: The invention comprises a focused ultrasound treatment device called a local system, wherein the focused ultrasound treatment device at least comprises a guiding device and a doctor workstation, wherein the doctor workstation comprises a computing server, wherein the doctor workstation imports the diagnosis result of the clinical diagnostic device and trains the image template based on the local deep learning model in the computing server; then, the image template is used to match and identify a plurality of images transmitted by the guiding device in the focused ultrasound treatment device, and an image with a recognition probability greater than a threshold is obtained, and the obtained image coordinates are the lesion position; The doctor workstation is the control center of the focused ultrasound therapy system, which controls the generation of instructions for the entire system, is also responsible for the retrieval and calculation of the local deep learning model, recognizes the images sent by the guidance device, generates treatment plans and implements treatment, and accesses and manages treatment cases; it can identify lesions without accessing the remote system; The local system places the desired treatment part of the patient within the detectable range of the guiding device, and the real-time image presented by the guiding device is transmitted to the computing server, which pre-processes the image data. Since the remote system is not connected, the report data obtained by the clinical diagnostic equipment before the treatment and the pre-processed real-time image are directly loaded on the computing server. The local deep learning model is pre-outlined for the lesion contour with the calculated maximum probability for the current image recognition probability obtained by the guiding device, and the lesion outline areas with other probabilities are listed for comparison. The doctor selects a certain result according to the probability given by the computing server loaded with the local deep learning model and manually adjusts it on this basis to optimize the outline area; The system is based on clinical diagnosis results and combined with AI to perform secondary identification of associations; The preprocessing adopts a CNN model or a Transformer model.

2. The focused ultrasound therapy system according to claim 1, characterized in that: The system also includes a remote deep computing server based on a remote deep learning model, which is part of the remote system. The remote system can simultaneously handle computing requirements from various local systems and optimize the training model together with local systems distributed around the world. The remote deep computing server communicates with the doctor's workstation through the network. The doctor's workstation imports the diagnosis results and transmits them to the remote deep computing server for training the remote deep learning model. The doctor's workstation pre-processes the images transmitted by the guiding device and transmits them to the remote deep computing server. The remote deep computing server obtains an image template through training based on the remote deep learning model; then, the image template is used to match and identify the images transmitted by the guiding device in the focused ultrasound treatment device to obtain an image with a recognition probability greater than a threshold, and the lesion area is determined based on the image, and then the lesion is treated and the remote deep learning model is optimized; the image with a recognition probability greater than the threshold refers to the image corresponding to the coincidence of the center of the circumscribed ellipsoid of the lesion area and the center of the field of view of the guiding device under the premise that the similarity is greater than the threshold; The image coordinates are the coordinate values ​​of the location of the guiding device when the image is acquired.

3. The focused ultrasound therapy system according to claim 2, characterized in that: After the remote system completes a period of interactive training, it regularly releases new versions of the remote deep learning model and the corresponding versions of the local deep learning model and updates them to the local system.

4. The focused ultrasound therapy system according to claim 2, characterized in that: The preprocessing includes at least cropping, compression, enhancement and feature extraction based on a preprocessing model; the preprocessing model includes at least a CNN model and a Transformer model.

5. The focused ultrasound therapy system according to claim 4, characterized in that: The CNN model includes SqueezeNet, GoogleNet, ShuffleNet, MobileNetV2, NasNetMobile and GoogleNet models, and the Transformer model includes at least ViT, CaiT and DeiT networks.

6. The focused ultrasound therapy system according to claim 1, characterized in that: The number of images whose recognition probability is greater than the recognition threshold is greater than or equal to 1 and less than or equal to 20; and the recognition threshold is greater than or equal to 80% and less than or equal to 1.

7. The focused ultrasound therapy system according to claim 2, characterized in that: The system includes multiple local systems that communicate with a remote deep computing server, which receives data sent by each doctor's workstation and trains and optimizes the remote deep learning model; Specifically: the data sent by each doctor workstation is sent to the remote deep learning model for training, and the training accuracy of the current round is counted. If the difference between this training accuracy and the training accuracy stored in the remote deep computing server is greater than or equal to the lowering threshold, the data sent by the current doctor workstation is discarded, otherwise it is retained.

8. The focused ultrasound therapy system according to any one of claims 1 to 7, characterized in that: The focused ultrasound treatment equipment also includes a nurse workstation, a treatment bed, a monitoring and protection unit, a water treatment unit, a motion control unit, an ultrasonic power unit and an ultrasonic transducer. The doctor workstation transmits data with the nurse workstation and the treatment bed respectively, and sends control signals to the water treatment unit, the motion control unit and the ultrasonic power unit. The functions performed by the local system also include establishing patient case files, importing patient pre-treatment examination report data into the system, patient pre-treatment, generating treatment plans and completing treatment.

9. A method for using a correlation positioning focused ultrasound treatment system, characterized in that: The method utilizes the focused ultrasound therapy device included in the associated positioning focused ultrasound therapy system described in claim 1 to receive the diagnosis results from the clinical diagnostic device and perform calculations on the local deep learning model, guiding the device to perform M×N pre-scans on the lesion site, wherein M is the number of scan rows and N is the number of scan columns, obtaining M×N lesion pre-scan images, and then performing identification and search in the M×N pre-scan images based on the local deep learning model and generating matching results, obtaining one or more images with a lesion recognition probability greater than a threshold, outputting the coordinate values ​​of the corresponding images, and then judging the lesion site based on the coordinate values ​​and performing treatment.

10. A method for using a correlation positioning focused ultrasound treatment system, characterized in that: The method utilizes the focused ultrasound therapy system described in any one of claims 2-8, the doctor workstation receives the diagnosis result and transmits it to the remote deep computing server for training the remote deep learning model, the guiding device performs M×N pre-scans on the lesion site, where M is the number of scan rows and N is the number of scan columns, and M×N lesion pre-scan images are obtained. The doctor workstation pre-processes the images transmitted by the guiding device and transmits them to the remote deep computing server, and the remote deep computing server performs training based on the remote deep learning model to obtain an image template; then, relying on the image template, the images transmitted by the guiding device in the focused ultrasound therapy system are matched and identified to obtain images with an identification probability greater than a threshold, and the obtained image coordinates are the lesion location. After the lesion location is determined, treatment is performed while optimizing the remote deep learning model and synchronizing the local deep learning model.

Images