Medical software system assisted by artificial intelligence

Abstract

The invention provides a medical software system assisted by artificial intelligence. The medical software system comprises a software function module and a special big language model; the specially-called large language model serves as a medium for interaction between a user and the software function module, an operation instruction is generated according to interaction information input by the user, a corresponding function interface of the software function module is called, and an expected function is achieved. According to the method, the specially called large language model is used as a medium for interaction between the user and the software function module, the interaction information between the user and the software function module is converted into the operation instruction matched with the software function module, the corresponding function interface is called, and the function expected by the user is achieved; a user does not need to directly face a complex software interface and various operation processes, and the learning and using threshold of a medical software system is lowered.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an artificial intelligence-assisted medical software system. Background Art

[0002] Medical software can perform complex functions in various scenarios, reducing the workload of medical staff. However, with the advancement of medical technology, the complexity and precision requirements of surgeries are constantly increasing, and the use of medical software is becoming increasingly complex. This requires people to spend a lot of time learning how to operate the software, which poses a huge challenge to doctors. For example, surgical planning software can assist doctors in planning surgical plans and improve their work efficiency. However, current surgical planning software has a complex interface, a multi-step process, and is difficult to operate. Different surgeries require different surgical planning software, and each type of surgery must be learned separately, which requires a long learning curve and a complex learning process. Senior physicians are less willing to learn.

[0003] To address the shortcomings of existing medical software, such as the complexity of operation and the need to spend a lot of time learning how to operate the software, the present invention provides an artificial intelligence-assisted medical software system to improve the interaction performance between doctors and surgical planning systems, and reduce the learning and use threshold of medical software systems. Summary of the Invention

[0004] The present invention provides an artificial intelligence-assisted medical software system to address the defects of the existing medical software in that the operation is complicated and a lot of time is required to learn how to operate the software.

[0005] In a first aspect, the present invention provides an artificial intelligence-assisted medical software system, comprising: a software function module, a specially tuned large language model;

[0006] The specially tuned large language model serves as a medium for interaction between the user and the software function module, and generates operation instructions based on the interaction information input by the user to call the corresponding functional interface of the software function module to achieve the expected function.

[0007] Optionally, the specially tuned large language model is obtained by:

[0008] Generate interactive corpus between users and large language models based on the usage scenarios of medical software systems;

[0009] The basic large language model is trained and fine-tuned using the interactive corpus to obtain the specially tuned large language model.

[0010] Optionally, the specially tuned large language model can provide the user with guidance information for using the software function module, guiding the user to input interactive information for generating operation instructions.

[0011] Optionally, after the software function module executes the corresponding function, the specially tuned large language model also feeds back information related to the execution result.

[0012] Optionally, the software function module is used to process medical images, and the software function module includes a plurality of function interfaces, each of which is used to implement different functions at different process nodes;

[0013] The specially tuned large language model generates operation instructions that conform to the process nodes based on the interaction information with the user, and calls the corresponding functional interface or functional interface combination according to the operation instructions to execute the corresponding function.

[0014] Furthermore, the specially tuned large language model performs any one or more of the following steps to guide the process of processing medical images:

[0015] Get the patient image file path entered by the user;

[0016] Identify the sequence type of the image files under the patient image file path;

[0017] Display the surgical types that can be supported by existing image file types for users to choose;

[0018] According to the target surgery type selected by the user, the sequence type of the image file to be used is determined, and the corresponding image template is matched from the existing image files of the sequence type;

[0019] Call the function interface and perform processing operations based on the image template.

[0020] Furthermore, the user inputs the patient image file path through the instruction interaction window, or the user inputs the patient image file path through the large language model interaction window.

[0021] Optionally, identifying the sequence type of the image files under the patient image file path includes processing the image files under the patient image file path by calling a pre-trained image classification model to identify the sequence type of each image file.

[0022] Furthermore, the image files in the patient image file path are processed by calling a pre-trained image classification model to identify the sequence type of each image file, including:

[0023] For each image file, a number of slices are taken out and respectively input into the image classification model to obtain classification results of the slices;

[0024] The classification results of each slice are summarized by majority voting to determine the sequence type of the image file.

[0025] Optionally, the types of surgeries that can be supported by each combination of image file sequence types are predetermined, and the specially tuned large language model can determine the types of surgeries that it can support based on the existing image file sequence types.

[0026] Optionally, the process of guiding the processing of medical images further includes: if the existing image file sequence type cannot support the target surgery type, prompting the required additional image file sequence type until the existing image file sequence type can support the target surgery type.

[0027] Optionally, the step of determining the sequence type of image files required based on the target surgery type selected by the user and matching corresponding image templates from existing image files of the sequence type includes:

[0028] When there is only one file in a certain type of image files required for the target type of surgery, directly use this file as a template for the image files of this type;

[0029] When there are two or more sub-files in a certain type of image file required for the target type of surgery, description information of these files is displayed for the user to select one as a template for the image file of this type.

[0030] Optionally, when the specially tuned large language model generates at least two pieces of information for the user to select and enter the next instruction, it also generates a link or button for each piece of information for the user to click and select.

[0031] Optionally, the specially tuned large language model also refers to historical information of previous interactions with the user when responding to interaction information input by the user.

[0032] Optionally, the system further includes an instruction interaction window and a large language model interaction window;

[0033] The user can input interaction information through the large language model interaction window at any process node, or directly input interaction instructions from the instruction interaction window.

[0034] Furthermore, when the user directly inputs an operation instruction through the instruction interaction window, the specially adjusted large language model also obtains the process node after the user operates the instruction interaction window and the status information after the operation, so as to output corresponding prompt information according to the current process node and status information.

[0035] In a second aspect, the present invention also provides an embodied intelligent robot, comprising any of the aforementioned artificial intelligence-assisted medical software systems.

[0036] In a third aspect, the present invention further provides a computer program product comprising computer-executable instructions, which, when executed, are used to implement the functions of the artificial intelligence-assisted medical software system as described in any of the foregoing items.

[0037] The artificial intelligence-assisted medical software system provided by the present invention has at least the following

[0038] Beneficial effects:

[0039] 1. A specially tuned large language model is used as a medium for interaction between users and software functional modules. The interactive information between users is converted into operation instructions adapted to the software functional modules, and the corresponding functional interfaces are called to realize the functions expected by users. This allows users to avoid facing complex software interfaces and numerous operating procedures directly, thus lowering the learning threshold for the use of medical software systems and reducing learning time.

[0040] 2. In some implementations, the medical software system is used to process medical images. The specially tuned large language model has the flexibility to intervene from any process node, and can guide users to input relevant information and instructions based on user needs and the current process node.

[0041] 3. Pre-trained image classification models can more accurately identify medical image types, avoiding the risk of misidentification caused by identifying file names and specific file fields.

[0042] 4. By selecting non-edge and non-center slices to identify image types and combining them with majority voting, the accuracy of image type identification is improved.

[0043] 5. Users can directly operate the command interaction window at any node, or input the desired operation through a specially tuned large language model, taking into account the efficiency and convenience of the surgical planning system.

[0044] 6. After the user directly enters the operation instruction through the instruction interaction window, the specially tuned large language model also obtains the process node after the user operates the instruction interaction window and the status information after the operation. It can accurately cut into the current process node and current status, and output prompt text in a targeted manner to guide the user's next operation or help the user understand the current status.

[0045] 7. When the specially tuned large language model generates an information list for users to select and enter the next instruction, it also generates a link or button for each option for users to click and select, avoiding the defect of low text input efficiency and improving interaction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 This is a schematic diagram of the structure of an artificial intelligence-assisted medical software system provided by the present invention;

[0048] Figure 2 This is a schematic diagram of a process for processing medical images guided by a specially tuned large language model in an artificial intelligence-assisted medical software system provided by the present invention;

[0049] Figure 3 This is an example of a user interface in an artificial intelligence-assisted medical software system provided by the present invention. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0051] The following combination Figure 1-Figure 3 Describe an artificial intelligence-assisted medical software system of the present invention, such as Figure 1 As shown, an artificial intelligence-assisted medical software system of the present invention includes a software function module 10 and a specially tuned large language model 20;

[0052] The specially tuned large language model 20 serves as a medium for interaction between the user and the software function module 10. It generates operation instructions based on the interaction information input by the user to call the corresponding functional interface of the software function module 10 to achieve the expected function.

[0053] Specifically, the existing software interface is complex, and the steps and processes for realizing specific functions are numerous, and the threshold for getting started is high. In this solution, a specially tuned large language model 20 is used as the interaction medium between the user and the software functional module 10. The specially tuned large language model 20 has natural language processing capabilities, can understand the interactive information input by the user, and generate one or more functional interfaces required for the operation instruction call according to user needs to achieve the expected function. The user inputs the interactive information in the form of character input or voice input, and the voice is converted into characters with the help of voice recognition. It is understandable that in order to achieve the ultimate expected goal, the user may need to interact with the specially tuned large language model 20 multiple times, and generate operation instructions to call the same or different functional modules multiple times.

[0054] For example, if a user needs to implement requirement A, the specially tuned large language model 20 prompts the user to perform steps A1, A2, and A3. When entering step A1, the specially tuned large language model 20 prompts the user to enter the information required for step A1. After the user enters a certain information, the specially tuned large language model 20 determines that the input information generates a control instruction (for example, the information is wrong / incomplete), and cannot generate an operation instruction. It also prompts the reason for the information error / missing and gives information example a. 11 After reading the prompt information, the user re-enters the new information. The specially tuned large language model 20 determines that the newly input information can generate a control instruction (the information is correct and complete), and generates an operation instruction based on it to call the functional interface P1 to perform the corresponding processing operation; enter step A2, the specially tuned large language model 20 prompts the user to enter the information a 21 、a 22 、a 23 、a 24 Select the information that meets their needs, the user selects a 23 After that, the specially tuned large language model 20 generates an operation instruction call and a 23 Corresponding functional interface P 23 Execute the corresponding processing operation. Entering step A3, the specially tuned large language model 20 combines the historical interaction information input by the user (such as the interaction information entered for the first time in step A1) to directly generate an operation instruction. At this time, no interaction with the user is required. The operation instruction is directly generated to call the functional interface P3 to perform the corresponding processing operation.

[0055] Of course, the above description is just one example. In specific scenarios, after the user inputs or selects different information, the subsequent steps may be different. Some steps may not need to be executed, and some steps may be executed directly without the user inputting any interactive information. In short, the specially tuned large language model 20 can attempt to understand the existing information input by the user (i.e., all information input before the current moment) based on the specific software function design, flexibly interact with the user, guide the user to input interactive information, or generate operation instructions based on the interactive information input by the user to call the corresponding functional interface to achieve the expected function.

[0056] This embodiment uses a specially tuned large language model as a medium for interaction between users and software functional modules, converts the interactive information between users into operation instructions adapted to the software functional modules, calls the corresponding functional interfaces, and realizes the functions expected by users, so that users do not need to directly face complex software interfaces and numerous operating procedures, thereby lowering the learning and use threshold of medical software systems.

[0057] It should also be noted that the present invention can be applied to different medical software usage scenarios, such as the use of multimodal medical image fusion software, the use of surgical navigation registration software, and the use of patient three-dimensional model reconstruction software. In some implementations, the artificial intelligence-assisted medical software system supports the use of a single software. Taking the scenario of patient three-dimensional model reconstruction software as an example, the artificial intelligence-assisted medical software system includes various functional modules of the three-dimensional model reconstruction software (for example, image reading, multimodal registration, image segmentation, image fusion and other functional modules). The specially tuned large language model can guide the user to interact with the software functional modules, generate operation instructions based on the interaction information, and call the corresponding functional interface of a functional module to achieve the corresponding function. In some implementations, the artificial intelligence-assisted medical software system supports the use of two or more software. The artificial intelligence-assisted medical software system also includes the functional modules of these software. The specially tuned large language model can guide the user to interact with the functional modules of these software to achieve the expected functions.

[0058] Furthermore, some software may have some of the same functional modules. Integrating these same functional modules in the AI-assisted medical software system can reduce system resource usage. For example, the first type of software supported by the medical software system has four functional modules: A, B, C, and D, and the second type of software supported by the medical software system has five functional modules: A, B, E, F, and G. Both types of software have the same functional modules A and B. These two types of software can be integrated into one software, merging the redundant functional modules. In other words, the integrated medical software system includes all functional modules of the two types of software without omission or duplication, which can reduce system resource usage.

[0059] Of course, it's also possible to integrate these two types of software without using other mechanisms to speed up the medical software. For example, after the first type of software calls function module A to achieve its purpose, the GPU storage space allocated for running function module A is released. If the second type of software then calls the same function module A to achieve its purpose, function module A can be loaded and run in situ in the GPU space, speeding up the medical software.

[0060] Based on the previous embodiment, in some embodiments, the specially tuned large language model 20 is obtained by:

[0061] Generate interactive corpus between users and large language model 20 based on the usage scenarios of medical software system;

[0062] The basic language model is trained and fine-tuned using the interactive corpus to obtain a specially tuned large language model20.

[0063] Specifically, different medical software systems implement different functions. Even software that implements the same function may have different operational requirements and operational processes due to different designs. This embodiment provides specialized training for specific medical software systems, enabling the specially tuned large language model 20 to better "interface" with software functions and better convert user interaction information into operational instructions that match the software function module 10.

[0064] It should be noted that the present invention can be applied to different scenarios, such as the use of surgical navigation registration software, the use of surgical planning software, and the use of patient three-dimensional model reconstruction software. For a specific medical software system, the interactive corpus between the user and the large language model generated in the use scenario of the medical software system can be used to train the basic language model, so that the trained specially tuned large language model 20 has the ability to process the interactive information in the application scenario. Basic language models include the OpenBioLLM large language model, the BioGPT large language model, and the deepseek large language model (deepseek R1 series model, deepseek MOE series model, etc.). Furthermore, on the basis of the basic language model, the interactive corpus of one or more medical software systems can be used for training and fine-tuning, so that the specially tuned large language model 20 can be used as an interactive medium for the above-mentioned one or more medical software systems.

[0065] The following uses surgical navigation software as an example to briefly describe the user's interaction process with the medical software system through the specially tuned large language model 20:

[0066] User input of “patient registration” requirements;

[0067] The model will first prompt you to import patient images (including surgical plans). Please enter the patient image path.

[0068] The model understands the path entered by the user and calls the corresponding functional interface to obtain the patient images under the path;

[0069] The model prompts you to select a registration method, which includes: laser point cloud registration and marker point registration (assuming the user selects marker point registration).

[0070] The model prompts the user to install the registered probe on the robot arm and prompts for confirmation after the probe is installed;

[0071] After the user confirms, the model prompts to select a marker point in the patient's 3D image;

[0072] After the user selects the marker points, the model prompts the user to use the probe to collect the marker points on the patient's body surface according to the order of the marker points.

[0073] The model prompts the user to click Confirm after the point collection is completed, and then performs registration calculations;

[0074] After completing the registration calculation, the model prompts you to select verification points to verify the registration accuracy;

[0075] After verification, the model prompts you to click "Navigate" to enter the navigation process.

[0076] Based on the aforementioned embodiments, in some embodiments, the specially tuned large language model 20 can provide the user with guidance information for using the software function module 10, guiding the user to input interactive information for generating operation instructions.

[0077] Specifically, in an ideal situation, users can always input relevant information completely and correctly, which the specially tuned large language model 20 will convert into operational instructions, without the need for interactive guidance from the large model. Users can even input the interactive information required for multiple process nodes at once, and the large model will generate one or more operational instructions based on this information to call the corresponding functional interfaces to implement the desired functions. However, in general, users have a low level of proficiency in the software and require interactive guidance from the specially tuned large language model 20.

[0078] For example, in the usage scenario of surgical navigation software, the user inputs the "patient registration" requirement, and the specially tuned large language model 20 prompts the user to enter the patient image first. The user enters the storage path of the patient's X-ray film, and the specially tuned large language model 20 prompts "The user is currently entering a two-dimensional projection image, and the user cannot perform patient registration. Please enter the patient's three-dimensional image, such as a three-dimensional magnetic resonance image, a three-dimensional CT (that is, containing multiple continuous tomographic images)."

[0079] This embodiment uses a specially tuned large language model to provide users with prompt information about software function modules, guiding users to input interactive information that can generate operation instructions, thereby lowering the learning and use threshold of the software.

[0080] Based on the aforementioned embodiments, in some embodiments, after the software function module 10 executes the corresponding function, the specially tuned large language model 20 also feeds back information related to the execution result.

[0081] Taking the surgical navigation software scenario as an example, the specially tuned large language model 20 prompts the user to enter the patient image first. The user then enters the storage path of the 3D magnetic resonance image. The specially tuned large language model 20 then calls the image reading interface to read file F from the storage path. After confirming that file F is available, the specially tuned large language model 20 provides feedback, stating that "File F has been used as the base image for patient registration," to help the user understand the current status of the software. Feedback can also be provided through other modules, for example, displaying the loaded patient image in the workspace and displaying the image file's descriptive attributes (such as file type, file name, and file size) in the information bar.

[0082] Based on the foregoing embodiments, in some embodiments, the software function module 10 is used to process medical images, and the software function module 10 includes several functional interfaces, which are used to implement different functions at different process nodes; the specially adjusted large language model 20 generates operation instructions that conform to the process nodes based on the interaction information with the user, and calls the corresponding functional interface or functional interface combination according to the operation instructions to execute the corresponding function.

[0083] Specifically, software module 10 is used for medical image processing, for example, medical image segmentation, brain map segmentation, multimodal image registration, and other tasks such as path planning and ablation plan planning based on medical images. Software module 10 includes several functional interfaces that enable different processing functions. A specially tuned large language model 20 serves as the interface between the user and software module 10. It determines the medical image processing flow based on user needs, guides the user to input interactive information according to the flow, converts the interactive information into operational instructions that conform to the flow nodes, and calls the corresponding functional interfaces to implement the corresponding functions.

[0084] This embodiment can gradually guide the interaction process with the user and guide the processing operations on medical images, thereby lowering the learning and usage threshold.

[0085] Based on the aforementioned embodiments, in some embodiments, the specially tuned large language model 20 performs any one or more of the following steps to guide the process of processing medical images:

[0086] S1. Obtain the patient image file path input by the user;

[0087] S2. Identify the sequence type of the image file under the patient image file path;

[0088] S3. Display the surgical types that can be supported by the existing image file types for the user to choose;

[0089] S4. Determine the sequence type of the image files to be used based on the target surgery type selected by the user, and match the corresponding image template from the existing image files of the sequence type;

[0090] S5. Call the function interface and perform processing operations based on the image template.

[0091] Specifically, S1 to S4 are common steps in many medical image processing processes, requiring image selection, image template determination, and target surgery type determination. S5 may call different functional interfaces based on different needs. For example, for cerebral hemorrhage drainage, the interface called may be cerebral hemorrhage path planning; for laser ablation, the interface called may be fiber path planning; for resection target area planning, the interface called may be brain map segmentation; and for stereotactic electroencephalography (SEEG), the interface called may be SEEG electrode path planning.

[0092] Of course, there are no strict timing requirements for the above steps, and not every step must be executed. The large language model can flexibly call and execute the corresponding steps based on the interactive information it understands. For example, in step S3, the user enters the plan for the type of surgery they want to perform, but the currently available images are insufficient to support the plan for this type of surgery. The user is prompted to supplement the images and re-execute step S1. For example, if the user enters the target type of surgery they want to perform at the beginning of the interaction, S3 can be skipped. Not every step of the above steps requires guidance from the specially tuned large language model 20. For example, the user can directly manually execute some of the above steps, and only operate through the specially tuned large language model 20 in some steps.

[0093] Furthermore, in some embodiments, after S4 and before S5, the process includes: selecting an image template as a reference according to a user instruction, and calling the first functional interface to perform registration of images of various modalities.

[0094] In some other embodiments, after S4 and before S5, the step includes: calling a second functional interface to pre-process the image file of the target type according to a user instruction.

[0095] In some other embodiments, after S4 and before S5, the process includes: calling a third function interface to perform image segmentation on the target type of image file according to a user instruction.

[0096] Based on the aforementioned embodiment, in some embodiments, in S1 , the user inputs the patient image file path through the specially tuned large language model 20 interactive window.

[0097] Specifically, in some implementations, the user enters a character-based path in an interactive window of the specially tuned large language model 20. In other implementations, the specially tuned large language model 20 provides a file selection pop-up window, through which the user enters the image file or the image file path, thus avoiding the drawbacks of character-based paths being difficult to enter and prone to errors.

[0098] Based on the aforementioned embodiment, in some embodiments, in S1 , the user inputs the patient image file path through the command interaction window.

[0099] Specifically, in this embodiment, the user does not directly input the path of the image file through the specially tuned large language model 20, but inputs the path of the patient image file through the command interaction window of the software.

[0100] Based on the aforementioned embodiment, in some embodiments, in S2, the image files in the patient image file path are processed by calling a pre-trained image classification model to identify the sequence type of each image file.

[0101] Specifically, image files, such as DICOM files, NifTI files, and TIFF files, can be classified into specific sequence types, such as T1WI, T2WI, DTI, and CT. Conventional methods identify the sequence type of an image file by reading a specific field in the image file. This method can lead to misjudgments, resulting in an unrecognizable sequence type. For example, tampering with this field can lead to misjudgment of the sequence type, while a missing field can render the sequence type unrecognizable.

[0102] This embodiment processes the image files under the patient image file path through a pre-trained image classification model, which can avoid misjudgment or inability to determine the sequence type of the image files and improve the accuracy.

[0103] Based on the previous embodiment, in some embodiments, the image files in the patient image file path are processed by calling a pre-trained image classification model to identify the sequence type of each image file, including:

[0104] For each image file, several slices are taken out and input into the image classification model to obtain the classification results of several slices;

[0105] The classification results of each slice are summarized by majority voting to determine the sequence type of the image file.

[0106] Specifically, image classification models, such as CNN models and transformer models, improve classification accuracy by selecting multiple slices from the image file and inputting them into the image classification model for comprehensive voting to determine the sequence type. Preferably, an odd number of slices is selected to facilitate the ultimate determination of a single, majority sequence type. Furthermore, when selecting slices, avoid selecting slices at the edge of the image to prevent selecting pure background slices.

[0107] Based on the aforementioned embodiments, in some embodiments, the types of surgeries that can be supported by each combination of image file sequence types are predetermined, and the specially tuned large language model 20 can determine the types of surgeries that it can support based on the existing image file sequence types in step S3.

[0108] Specifically, the surgical types supported by each combination of image file sequence types can be derived from the physician's experience. These surgical types can be directly written into the specially tuned large language model 20, or trained on the specially tuned large language model 20 after generating corresponding training data, enabling it to determine the supported surgical types based on the existing image file sequence types.

[0109] Based on the aforementioned embodiments, in some embodiments, the process of guiding the processing of medical images further includes: if the existing image file sequence type cannot support the target surgery type, prompting the required additional image file sequence type until the existing image file sequence type can support the target surgery type.

[0110] Specifically, for example, in S3, the user does not select from the displayed surgical types, but instead enters a target surgical type. Upon determining that a modality is missing for performing the target surgical type, the specially tuned large language model 20 further prompts the user to supplement image files of the corresponding sequence type. For another example, if the user initially enters the target surgical type, and after acquiring patient imaging files, determines that a modality is missing for performing the target surgical type, the specially tuned large language model 20 further prompts the user to supplement image files of the corresponding sequence type.

[0111] Based on the foregoing embodiment, in some embodiments, S4 includes:

[0112] When there is only one file in a certain type of image files required for the target type of surgery, directly use this file as a template for the image files of this type;

[0113] When there are two or more sub-files in a certain type of image file required for the target type of surgery, description information of these files is displayed for the user to select one as a template for the image file of this type.

[0114] Specifically, when only one image file exists for a specific type of surgery, using that file as a template can improve efficiency and reduce redundant interactions. When two or more separate files exist for a specific type of surgery, the user can display descriptive information about these files, making it easier for the user to select the appropriate file as a template. This descriptive information includes, for example, image file size, resolution, and acquisition time.

[0115] Based on the aforementioned embodiment, in some embodiments, when the specially tuned large language model 20 generates at least two pieces of information for the user to select and enter the next instruction, it also generates a link or button for each piece of information for the user to click and select.

[0116] For example, the specially tuned large language model 20 outputs the following information:

[0117] T1 data A, size is [256, 256, 192], spacing is [1.0, 1.0, 1.2] .

[0118] T1 data B, size is [256,256,32], spacing is [1.0,1.0,5] .

[0119] Please select a template.

[0120] By clicking on the attached link, users can directly select the corresponding image file as a template, thereby improving interaction efficiency.

[0121] Based on the aforementioned embodiments, in some embodiments, the specially tuned large language model 20 also refers to historical information of previous interactions with the user when responding to interaction information input by the user.

[0122] The historical information here refers to all the information entered during the process of using the medical software system to process medical information for a patient. For example, when interacting in step S1, the user inputs "Please perform cerebral hemorrhage path planning on the patient image in the path aaa / bbb / ccc". The interactive information includes the target surgery type. After executing step S2, based on the historical information entered by the user in step S1, it is determined that there is no need to execute step S3 for information display, and the target surgery type is directly determined to be "cerebral hemorrhage path planning".

[0123] This embodiment improves the interaction efficiency between the specially tuned large language model and the user, and improves the operating efficiency of the medical software system.

[0124] Reference Figure 3 ,Based on the aforementioned embodiment, in some embodiments, the system further includes an instruction interaction window, a large language model interaction window;

[0125] Users can enter interactive information through the large language model interaction window at any process node, or directly enter interactive instructions from the instruction interaction window.

[0126] It is understandable that Figure 3 This is just an example of the layout of the command interaction window and the large language model interaction window. The size and position of the command interaction window and the large language model interaction window can be flexibly set. For example, the command interaction window can be set and / or placed in the workspace, or the large language model interaction window can be embedded in the command interaction window. For example, the large language model interaction window can be set to a callable mode, hiding the large language model interaction window in the normal state, and the user can call out the large language model interaction window through a preset command.

[0127] Based on the previous embodiment, in some embodiments, when the user directly inputs an operation instruction through the instruction interaction window, the specially adjusted large language model 20 also obtains the process node after the user operates the instruction interaction window and the status information after the operation, so as to output corresponding prompt information based on the current process node and status information.

[0128] Specifically, the user manually selects the target surgery type as "cerebral hemorrhage path planning" in the command interaction window. The specially adjusted large language model 20 reads the status after the user operation, outputs the prompt information "The execution of cerebral hemorrhage path planning has been selected", and determines that the next process node is template matching, and outputs the prompt information "There are already CT files A and CT files B, please select one as the CT image template."

[0129] In this embodiment, the specially tuned large language model also obtains the process nodes after the user operates the instruction interaction window and the status information after the operation, so that the user can seamlessly switch between "directly operating the instruction interaction window" and "seeking help from the specially tuned large language model", making the process of using the medical software system smoother.

[0130] The following describes an embodied intelligent robot provided by the present invention. The embodied intelligent robot described below and the artificial intelligence-assisted medical software system described above can be referenced to each other.

[0131] The present invention provides an embodied intelligent robot, comprising any of the aforementioned AI-assisted medical software systems. Specifically, users can utilize a specially tuned large language model to conveniently access the functional services provided by the embodied intelligent robot, lowering the barrier to entry for learning and using the embodied intelligent robot.

[0132] The present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can perform the functions of the above-mentioned artificial intelligence-assisted medical software systems.

[0133] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An artificial intelligence-assisted medical software system, characterized in that: include: Software functional modules and specially tuned large language models; The specially tuned large language model serves as a medium for interaction between the user and the software function module, and generates operation instructions based on the interaction information input by the user to call the corresponding functional interface of the software function module to achieve the expected function.

2. The artificial intelligence-assisted medical software system according to claim 1, characterized in that: The specially tuned large language model is obtained in the following manner: Generate interactive corpus between users and large language models based on the usage scenarios of medical software systems; The basic large language model is trained and fine-tuned using the interactive corpus to obtain the specially tuned large language model.

3. The artificial intelligence-assisted medical software system according to claim 1, characterized in that: The specially tuned large language model can provide users with guidance information for using the software function module, guiding users to input interactive information for generating operation instructions.

4. The artificial intelligence-assisted medical software system according to claim 1, characterized in that: After the software function module executes the corresponding function, the specially tuned large language model also feeds back information related to the execution result.

5. The artificial intelligence-assisted medical software system according to claim 1, characterized in that: The software function module is used to process medical images, and the software function module includes several functional interfaces, which are respectively used to implement different functions at different process nodes; The specially tuned large language model generates operation instructions that conform to the process nodes based on the interaction information with the user, and calls the corresponding functional interface or functional interface combination according to the operation instructions to execute the corresponding function.

6. The artificial intelligence-assisted medical software system according to claim 5, characterized in that: The specially tuned large language model performs any one or more of the following steps to guide the process of processing medical images: Get the patient image file path entered by the user; Identify the sequence type of the image files under the patient image file path; Display the surgical types that can be supported by existing image file types for users to choose; According to the target surgery type selected by the user, the sequence type of the image file to be used is determined, and the corresponding image template is matched from the existing image files of the sequence type; Call the function interface and perform processing operations based on the image template.

7. The artificial intelligence-assisted medical software system according to claim 6, characterized in that: The user inputs the patient image file path through the instruction interaction window, or the user inputs the patient image file path through the large language model interaction window.

8. The artificial intelligence-assisted medical software system according to claim 6, characterized in that: The identifying the sequence type of the image files under the patient image file path includes processing the image files under the patient image file path by calling a pre-trained image classification model to identify the sequence type of each image file.

9. The artificial intelligence-assisted medical software system according to claim 8, characterized in that: The image files in the patient image file path are processed by calling a pre-trained image classification model to identify the sequence type of each image file, including: For each image file, a number of slices are taken out and respectively input into the image classification model to obtain classification results of the slices; The classification results of each slice are summarized by majority voting to determine the sequence type of the image file.

10. The artificial intelligence-assisted medical software system according to claim 6, characterized in that: The types of surgeries that can be supported by each combination of image file sequence types are predetermined, and the specially tuned large language model can determine the types of surgeries that it can support based on the existing image file sequence types.

11. The artificial intelligence-assisted medical software system according to claim 6, characterized in that: The process of guiding the processing of medical images further includes: if the existing image file sequence type cannot support the target surgery type, prompting the image file sequence type that needs to be supplemented until the existing image file sequence type can support the target surgery type.

12. The artificial intelligence-assisted medical software system according to claim 6, characterized in that: The process of determining the sequence type of image files required based on the target surgery type selected by the user and matching the corresponding image template from the existing image files of the sequence type includes: When there is only one file in a certain type of image files required for the target type of surgery, directly use this file as a template for the image files of this type; When there are two or more sub-files in a certain type of image file required for the target type of surgery, description information of these files is displayed for the user to select one as a template for the image file of this type.

13. The artificial intelligence-assisted medical software system according to claims 1-12, characterized in that: When the specially tuned large language model generates at least two pieces of information for the user to select and input the next instruction, it also generates a link or button for each piece of information for the user to click and select.

14. The artificial intelligence-assisted medical software system according to claims 1-12, characterized in that: When responding to interaction information input by the user, the specially tuned large language model also refers to historical information of previous interactions with the user.

15. The artificial intelligence-assisted medical software system according to claims 1-12, characterized in that: The system also includes a command interaction window and a large language model interaction window; The user can input interaction information through the large language model interaction window at any process node, or directly input interaction instructions from the instruction interaction window.

16. An embodied intelligent robot, characterized in that: An artificial intelligence-assisted medical software system comprising the artificial intelligence-assisted medical software system according to any one of claims 1 to 15.

17. A computer program product comprising computer-executable instructions, characterized in that: When executed, the instructions are used to implement the functions of the artificial intelligence-assisted medical software system as described in any one of claims 1 to 15.

Images