Critical nursing ultrasonic image training system

Abstract

The invention relates to the technical field of medical training equipment, in particular to a critical care ultrasonic image training system, which comprises a clinical data acquisition module for acquiring critical clinical case data, ultrasonic image materials and diagnosis and treatment process specifications; the training scene construction module is used for constructing a high-fidelity three-dimensional training scene according to severe clinical case data, ultrasonic image materials and diagnosis and treatment process specifications; the virtual ultrasonic practical operation module simulates the operation of an ultrasonic probe based on a high-fidelity three-dimensional training scene, and generates an ultrasonic image corresponding to an anatomical part in real time; the intelligent skill assessment center assesses operation normalization and diagnosis accuracy through practical operation data and real-time generation of an ultrasonic image corresponding to an anatomical site in combination with an improved ResNet-LSTM model, and an assessment result is obtained; and the personalized teaching unit generates a customized training scheme and knowledge strengthening content according to the evaluation result in combination with the learning track of the student. Therefore, the problems of insufficient dynamic scene simulation, poor authenticity and the like in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of medical training equipment technology, specifically to a critical care ultrasound imaging training system. Background Technology

[0002] Critical care ultrasound, as a core diagnostic and treatment technology in critical care medicine, has become a key tool for differentiating the causes of shock, grading respiratory failure, assessing multiple organ function, and guiding bedside interventions due to its non-invasive, real-time, and portable advantages. Its application level directly determines the treatment efficiency, success rate, and prognosis of critically ill patients. In critical care settings, patients often present with complex characteristics such as hemodynamic instability, anatomical variations, and multiple organ failure, placing extremely high demands on the precision of probe manipulation, parameter adjustment adaptability, and accuracy of pathological image recognition in ultrasound operation. Novice medical staff require systematic and scenario-based training before they can independently perform clinical operations. In long-term training practice, factors such as insufficient diversity of critically ill cases, restrictions on high-risk operations, and limitations of simulation equipment performance can easily lead to problems such as incomplete coverage of training scenarios, delayed operational feedback, and low skill conversion efficiency. This results in novice medical staff being prone to probe positioning errors, image interpretation errors, and even diagnostic and treatment decision-making errors during clinical operations. As modern medicine develops towards precision, intelligence, and efficiency, higher demands are placed on ultrasound imaging training in critical care nursing, particularly in areas such as high-fidelity simulation of complex scenarios, real-time perception of the operation process, precise identification of skill gaps, and full-cycle training management. There is an urgent need for a training system that can deeply integrate clinical scenarios with virtual simulation, dynamically capture operation and image correlation, and intelligently drive training and assessment decisions to ensure efficient connection between training effectiveness and clinical application.

[0003] However, traditional training methods for ultrasound imaging in critical care nursing have significant shortcomings: The perception and feedback dimensions are singular, relying heavily on manual observation by instructors or local parameter acquisition from simple simulation equipment. This only allows for the assessment of superficial information such as probe displacement and basic image display, failing to develop a multi-dimensional, comprehensive understanding of operational standardization, parameter adjustment rationality, and image interpretation accuracy. It also makes it difficult to capture the dynamic correlation between operational behavior, image quality, and diagnostic results. Furthermore, virtual simulation applications remain at the static scenario stage. Virtual models built based on standard anatomical drawings cannot accurately reflect dynamic states such as patient positional changes and pathological variations in real time, resulting in significant deviations between virtual operations and clinical reality. The realism of the scenarios and the efficiency of skill transfer are limited; training and assessment strategies are disconnected, with training content often adopting a standardized curriculum without dynamic adjustments based on trainees' basic levels and operational weaknesses. Assessment relies heavily on subjective scoring, lacking quantitative evaluation of operational details and image interpretation logic, which easily leads to problems such as "training content not matching needs" and "assessment results not matching actual abilities." Furthermore, the lack of immersive and interactive 3D visualization methods makes it difficult for trainees to intuitively grasp the impact of operational details such as probe angle and pressure on image generation, resulting in slow improvement in the ability to identify features of complex pathological images. Overall, the technology faces multiple challenges, including insufficient dynamic scenario simulation, poor realism, and poor accuracy in training decisions. Summary of the Invention

[0004] This application provides a critical care ultrasound imaging training system to address the problems of insufficient dynamic scene simulation and poor realism in existing technologies.

[0005] The first aspect of this application provides a critical care ultrasound imaging training system, comprising: a clinical data acquisition module, a training scenario construction module, a virtual ultrasound practice module, an intelligent skills assessment center, and a personalized teaching unit; wherein, the clinical data acquisition module is used to collect critical care clinical case data, ultrasound image materials, and diagnostic and treatment process specifications; the training scenario construction module is used to construct a high-fidelity three-dimensional training scenario based on the critical care clinical case data, ultrasound image materials, and diagnostic and treatment process specifications; the virtual ultrasound practice module, based on the high-fidelity three-dimensional training scenario, simulates the operation logic of the ultrasound probe and generates ultrasound images of corresponding anatomical sites in real time; the intelligent skills assessment center is used to evaluate the standardization of operation and diagnostic accuracy by combining the practice data and the real-time generated ultrasound images of corresponding anatomical sites with an improved ResNet-LSTM model, and obtain skills assessment results; the personalized teaching unit is used to generate customized training plans and knowledge reinforcement content based on the skills assessment results and the learner's learning trajectory.

[0006] Preferably, the clinical data acquisition module includes a case data acquisition unit, an image material integration unit, and a standardization unit. The case data acquisition unit is used to collect basic information, disease severity, ultrasound examination indications, clinical diagnostic conclusions, and prognostic data of critically ill patients. The image material integration unit is used to collect ultrasound images of different anatomical locations and different pathological states, including standard and abnormal section images. The standardization unit is used to organize critical care ultrasound diagnosis and treatment guidelines, operating procedures, and quality control standards to form a structured diagnosis and treatment process document.

[0007] Preferably, the training scenario construction module includes a scenario modeling unit and an interaction logic configuration unit. The scenario modeling unit constructs a high-fidelity three-dimensional training scenario based on clinical data and anatomical standards, restoring the body position, anatomical structure details and pathological features of critically ill patients. The interaction logic configuration unit simulates the clinical ultrasound examination process, configures the pre-operation logic for body position adjustment, probe selection, and coupling agent use, sets operation constraint rules, and restores the clinical practice scenario.

[0008] Preferably, the virtual ultrasound operation module includes a probe simulation unit, an image rendering unit, and a real-time interactive feedback unit. The probe simulation unit simulates the operation of commonly used clinical ultrasound probes, performing multi-dimensional operations such as movement, rotation, pressing, and tilting, and switching between different ultrasound planes. The image rendering unit generates ultrasound images of corresponding anatomical sites in real time based on a physical acoustic model and clinical ultrasound image characteristics, simulating the image manifestations under different pathological conditions. The real-time interactive feedback unit provides feedback on the operation status through visual and auditory cues, and provides real-time warnings for violations.

[0009] Preferably, the intelligent skills assessment center includes a data extraction unit, a model evaluation unit, and a result generation unit. The data extraction unit is used to collect the trainees' practical data, real-time generated ultrasound image features, and diagnostic conclusions. The model evaluation unit uses an improved ResNet-LSTM model, extracting spatial features of ultrasound images and spatial trajectory features of operations through ResNet, and capturing the temporal logic of operations and diagnostic decision-making processes through LSTM for quantitative evaluation. The result generation unit outputs skills scores and a weakness analysis report, enabling trainees to clearly identify weaknesses in operation procedures, section recognition, and pathological diagnosis.

[0010] Preferably, the personalized teaching unit includes a learning trajectory analysis unit, a program generation unit, and a knowledge reinforcement module. The learning trajectory analysis unit records the student's learning time, number of practical exercises, assessment results, and error logs, analyzing the student's knowledge gaps and skill deficiencies. The program generation unit generates customized training programs based on the skill assessment results and learning trajectory, including targeted practical training, key theoretical learning points, and training cycle planning. The knowledge reinforcement module provides case studies, ultrasound imaging differential diagnosis guidelines, and video reinforcement content on operational skills, enabling students to learn as needed, synchronously track training effects, and dynamically adjust training programs.

[0011] A second aspect of this application provides a method for training critical care ultrasound imaging, comprising: acquiring critical care clinical case data, ultrasound imaging materials, and diagnostic and treatment procedure specifications; constructing a high-fidelity three-dimensional training scenario based on the critical care clinical case data, ultrasound imaging materials, and diagnostic and treatment procedure specifications; simulating the operation logic of an ultrasound probe based on the high-fidelity three-dimensional training scenario, generating ultrasound images of corresponding anatomical locations in real time, evaluating the operational standardization and diagnostic accuracy by combining practical data with the real-time generated ultrasound images of corresponding anatomical locations, and using an improved ResNet-LSTM model to obtain a skills assessment result; and generating a customized training plan and knowledge reinforcement content based on the skills assessment result and the trainee's learning trajectory.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement a critical care ultrasound imaging training method as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a critical care ultrasound imaging training method as described in the above embodiments.

[0014] The fifth aspect of this application provides a computer program product, including a computer program or instructions, for implementing a critical care ultrasound imaging training method as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects: This application's embodiments comprehensively collect critical clinical case data, ultrasound image materials, and standardized diagnostic and treatment procedures through a clinical data acquisition module, providing multi-dimensional standardized foundational data for training scenario construction and skills assessment. The training scenario construction module, based on anatomical standards and clinical data, constructs a high-fidelity 3D training scenario and achieves precise mapping between ultrasound images and 3D models, overcoming the limitations of insufficient realism in traditional training scenarios. The virtual ultrasound practice module accurately simulates the multi-dimensional operational logic of ultrasound probes and the clinical ultrasound image generation mechanism, combined with real-time interactive feedback functions, enhancing the immersion and standardization of practical training and strengthening the practical application of operational skills. The intelligent skills assessment center, using an improved ResNet-LSTM model, integrates practical data and image features to achieve quantitative assessment of operational standardization and diagnostic accuracy, enhancing the accurate identification of skill gaps. Personalized teaching units, based on skills assessment results and learning trajectory analysis, generate customized training plans and knowledge reinforcement content, helping trainees accurately fill knowledge gaps and skill deficiencies, effectively improving training relevance, learning efficiency, and clinical ultrasound diagnostic and treatment application capabilities. Thus, it solves the problems of insufficient dynamic scenario simulation and poor realism in existing technologies.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the structure of a critical care ultrasound imaging training system according to an embodiment of this application; Figure 2 This is a schematic diagram of a clinical data acquisition module provided according to an embodiment of this application; Figure 3 This is a schematic diagram of a training scenario construction module provided according to an embodiment of this application; Figure 4 This is a schematic diagram of a virtual ultrasound practice module provided according to an embodiment of this application; Figure 5 This is a schematic diagram of an intelligent skills assessment center provided according to an embodiment of this application; Figure 6 This is a schematic diagram of a personalized teaching unit provided according to an embodiment of this application; Figure 7 This is a flowchart of a critical care ultrasound imaging training system according to an embodiment of this application; Figure 8 This is a flowchart illustrating a critical care ultrasound imaging training method according to an embodiment of this application; Figure 9 This is a schematic diagram of a critical care ultrasound imaging training method according to an embodiment of this application; Figure 10 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The following description, with reference to the accompanying drawings, illustrates an embodiment of a critical care ultrasound imaging training system. Addressing the issue of insufficient dynamic scene simulation mentioned in the background section, this application provides a critical care ultrasound imaging training system. In this system, a clinical data acquisition module comprehensively collects critical care clinical case data, ultrasound image materials, and standardized diagnostic and treatment procedures, providing multi-dimensional standardized foundational data for training scenario construction and skills assessment. The training scenario construction module, based on anatomical standards and clinical data, constructs a high-fidelity three-dimensional training scenario and achieves precise mapping between ultrasound images and the three-dimensional model, overcoming the limitations of insufficient realism in traditional training scenarios. A virtual ultrasound practice module accurately simulates the multi-dimensional operational logic of ultrasound probes and the clinical ultrasound image generation mechanism. Combined with real-time interactive feedback, it enhances the immersion and standardization of practical training, strengthening the practical application of operational skills. An intelligent skills assessment center, utilizing an improved ResNet-LSTM model, integrates practical data and image features to achieve quantitative assessment of operational standardization and diagnostic accuracy, enhancing the accurate identification of skill gaps. Personalized teaching units, based on skills assessment results and learning trajectory analysis, generate customized training plans and knowledge reinforcement content to help trainees accurately fill knowledge gaps and skill deficiencies, effectively improving training relevance, learning efficiency, and clinical ultrasound diagnostic and treatment application capabilities. This solves the problems of insufficient dynamic scene simulation and poor realism in existing technologies.

[0020] Figure 1 This is a schematic diagram of the structure of a critical care ultrasound imaging training system provided in an embodiment of this application.

[0021] This application provides a critical care ultrasound imaging training system, the system 10 comprising: The system includes 100 clinical data acquisition modules, 200 training scenario construction modules, 300 virtual ultrasound practice modules, 400 intelligent skills assessment centers, and 500 personalized teaching units.

[0022] The system comprises the following modules: Clinical Data Acquisition Module 100, which collects data on severe clinical cases, ultrasound images, and standardized diagnostic and treatment procedures; Training Scenario Construction Module 200, which constructs a high-fidelity 3D training scenario based on the severe clinical case data, ultrasound images, and standardized diagnostic and treatment procedures; Virtual Ultrasound Practice Module 300, which simulates the operation logic of an ultrasound probe and generates ultrasound images of corresponding anatomical locations in real time based on the high-fidelity 3D training scenario; Intelligent Skills Assessment Center 400, which assesses the standardization of operations and diagnostic accuracy by combining practice data and real-time generated ultrasound images of corresponding anatomical locations with an improved ResNet-LSTM model, and obtains skills assessment results; and Personalized Teaching Unit 500, which generates customized training plans and knowledge reinforcement content based on the skills assessment results and the learner's learning trajectory.

[0023] It is understood that in this embodiment, the clinical data acquisition module comprehensively collects critical clinical case data, ultrasound image materials, and diagnostic and treatment process specifications, providing multi-dimensional standardized basic data for training scenario construction and skills assessment. The training scenario construction module, based on anatomical standards and clinical data, constructs a high-fidelity three-dimensional training scenario and achieves precise mapping between ultrasound images and three-dimensional models, overcoming the limitations of insufficient realism in traditional training scenarios. The virtual ultrasound practice module accurately simulates the multi-dimensional operation logic of ultrasound probes and the clinical ultrasound image generation mechanism, combined with real-time interactive feedback functions, enhancing the immersion and standardization of practical training and strengthening the practical application of operational skills. The intelligent skills assessment center, using an improved ResNet-LSTM model, integrates practical data and image features to achieve quantitative assessment of operational standardization and diagnostic accuracy, enhancing the accurate identification of skill gaps. Personalized teaching units, based on skills assessment results and learning trajectory analysis, generate customized training plans and knowledge reinforcement content, helping trainees accurately fill knowledge gaps and skill deficiencies, effectively improving training relevance, learning efficiency, and clinical ultrasound diagnostic and treatment application capabilities. Thus, it solves the problems of insufficient dynamic scenario simulation and poor realism in existing technologies.

[0024] In this embodiment of the application, the clinical data acquisition module 100 includes: Figure 2 As shown, there are three units: case data collection unit, image material integration unit, and standardization unit.

[0025] The case data collection unit is used to collect basic information, disease severity, ultrasound examination indications, clinical diagnosis conclusions and prognostic data of critically ill patients; the image material integration unit is used to collect ultrasound images of different anatomical sites and different pathological states, including standard and abnormal section materials; the standardization unit is used to organize the guidelines, operating procedures and quality control standards for critical care ultrasound diagnosis and treatment, and form a structured diagnosis and treatment process document.

[0026] It is understood that the embodiments of this application comprehensively collect basic information, disease severity, ultrasound examination indications, clinical diagnostic conclusions, and prognostic data of critically ill patients through the case data acquisition unit. With the help of the image material integration unit, standard and abnormal ultrasound images of different anatomical sites and different pathological states are collected. According to the standardization unit, the guidelines, operating procedures, and quality control standards for critical care ultrasound diagnosis and treatment are sorted out and a structured diagnosis and treatment process document is formed. This provides comprehensive, standardized, and high-quality basic data support for the construction of subsequent training scenarios, virtual practice simulation, and intelligent skills assessment. This not only ensures the authenticity and clinical authority of the training data, but also lays a core foundation for the precise alignment of training content with clinical practice, and improves the high fidelity of subsequent training scenarios, the clinical relevance of practical training, and the objective accuracy of skills assessment.

[0027] It should be noted that the standardization unit generates structured diagnostic and treatment process documents through multi-dimensional data integration and structured processing mechanisms. First, it collects unstructured / semi-structured raw data, including authoritative diagnostic and treatment guidelines in the field of critical care ultrasound (including key points for ultrasound assessment of disease classification, and definitions of indications / contraindications), clinical operation specifications (covering equipment parameter settings, probe selection, standard operating procedures, and image acquisition requirements), and quality control standards (including examination time limits, report completeness indicators, and operational compliance criteria). Second, it uses Natural Language Processing (NLP) technology to parse the raw data and extract core diagnostic and treatment elements (such as examination process nodes, judgment thresholds, intervention indications, and quality control standards). The system includes control and assessment points, and performs deduplication and standardization verification (unifying terminology definitions, data formats, and process logic). Subsequently, according to the clinical application scenarios of critical care ultrasound (such as shock assessment, respiratory failure diagnosis and treatment, and multi-organ function monitoring), the extracted structured elements are classified and integrated to construct a full-process logical framework of indication screening, preoperative preparation, operation steps, image analysis, result determination, quality control verification, and report output. Finally, the elements are associated and the process is linked to automatically generate a structured diagnosis and treatment process document that is hierarchical, logically coherent, and meets the needs of clinical practice. The document also embeds compliance prompts corresponding to the quality control standards to ensure the standardization and traceability of the diagnosis and treatment process.

[0028] In this embodiment of the application, the training scenario construction module 200 includes: as follows Figure 3 As shown, the scene modeling unit and the interaction logic configuration unit.

[0029] The scenario modeling unit constructs a high-fidelity 3D training scenario based on clinical data and anatomical standards, restoring the position, anatomical details and pathological features of critically ill patients; the interaction logic configuration unit simulates the clinical ultrasound examination process, configures the pre-operation logic for position adjustment, probe selection and coupling agent use, sets operation constraint rules, and restores the clinical practice scenario.

[0030] It is understood that the scenario modeling unit in this application constructs a high-fidelity 3D training scenario based on clinical data and anatomical standards, restoring the position, anatomical details, and pathological characteristics of critically ill patients. Combined with the interactive logic configuration unit, it simulates the clinical ultrasound examination process, configures the pre-operation logic, and sets operation constraint rules, allowing trainees to become familiar with the entire clinical practice process and master the key points of standardized operation in an immersive environment. This effectively avoids operational risks in real clinical scenarios, improves the relevance and effectiveness of training, and helps trainees quickly accumulate practical experience and shorten the clinical adaptation period.

[0031] It should be noted that a high-fidelity 3D training scenario is constructed based on clinical data and anatomical standards. Using anatomical standards as a foundation, core standardized information of normal human anatomy is extracted, including organ morphology (such as the inherent shape and size proportions of organs like the heart, lungs, and liver), tissue layers (such as the distribution and thickness of skin, muscles, fat, and fascia), vascular and nerve distribution (such as the course of arteries and veins, and nerve branch pathways), and skeletal and joint structures (such as bone morphology, joint connection methods, and range of motion), forming the basic anatomical framework of the 3D model. Simultaneously, multimodal clinical data from critically ill patients is collected, covering imaging data (such as CT, MRI, and ultrasound images), case diagnostic data (such as lesion type, lesion location, and lesion severity), and postural monitoring data (such as bedridden posture, limb positioning angles, and trunk flexion status). This data is then processed through image segmentation and feature extraction. Data acquisition and annotation processes are used to obtain precise data containing pathological features (such as lesion size, shape, boundary features, and tissue infiltration range), individual positional differences, and anatomical structural variations. Subsequently, the individual-specific information extracted from clinical data is registered, fused, and optimized with the standard anatomical framework to correct the differences between individual anatomical structures and standard models. This accurately maps the specific positional state of critically ill patients, refines the subtle features of anatomical structures (such as the minute direction of vascular branches and the fine texture of tissues), and visualizes the spatial distribution and morphological manifestation of pathological features. Ultimately, a high-fidelity 3D training scenario is constructed that can realistically reproduce the positional posture of critically ill patients, accurately replicate anatomical structural details, and clearly present pathological features. This provides a high-precision digital carrier that fits clinical practice for clinical ultrasound training, surgical planning, pathological mechanism research, and medical equipment development. This study simulates a clinical ultrasound examination process, configuring pre-operative logic for patient positioning, probe selection, and coupling agent application. Operational constraints are set to recreate a clinical scenario. First, the standard operating procedures and core process nodes for ultrasound examinations of critically ill patients are outlined, including preoperative preparation, patient positioning assessment and adjustment, ultrasound probe selection and matching, standardized use of coupling agent, probe operation and scanning, and image acquisition and analysis. The operational sequence, technical requirements, and clinically applicable scenarios for each process node are clarified, constructing a logical framework for the ultrasound examination process. Then, refined logical rules are configured for the three key pre-operative operations: patient positioning, probe selection, and coupling agent application. The patient positioning logic must cover common patient positions (such as supine, lateral, and semi-recumbent positions). The system should define the adjustment path and operation steps (such as trunk angle adjustment, limb placement and fixation, and pillow support adjustment) for the prone position, and associate them with the positional adaptation requirements of the examination site (e.g., cardiac examination requires a supine position combined with a left lateral decubitus position, and abdominal examination requires a supine position with knees bent). It should support automatic recommendation of the optimal position or manual precise adjustment of positional parameters based on the examination target. The probe selection logic should integrate the technical parameters (such as frequency range, detection depth, and resolution) and applicable scenarios (e.g., linear array probes for superficial tissue examinations, convex array probes for deep organ examinations, and phased array probes for cardiac examinations) of different types of ultrasound probes (e.g., linear array probes for superficial tissue examinations, convex array probes for deep organ examinations, and phased array probes for cardiac examinations). It should set intelligent matching rules between the probe and the examination site and purpose, and simultaneously simulate probe connection... The operation procedures and equipment feedback for connection and switching must be clearly defined; the logic for using coupling agent must specify the application site (skin surface of the examination area), application amount (uniform coverage of the examination area without bubbles), and application method (uniform application in a spiral or straight motion), as well as the timing relationship between coupling agent application and probe operation (probe scanning must be performed promptly after application to prevent the coupling agent from drying out); based on this, multi-dimensional operation constraint rules should be set, including operation sequence constraints (e.g., probe selection can only be performed after body positioning is completed, and probe scanning can only be started after coupling agent application is completed), range of motion constraints (e.g., the probe scanning angle should not exceed ±45° to avoid excessive pressure on the skin), and error operation feedback constraints (e.g., triggering a prompt "Please adjust" when body positioning is not properly adjusted). The system includes several steps: patient positioning to supine, exposing the subxiphoid region; displaying "The current probe is not suitable for lung ultrasound examination; a convex array probe is recommended" when the probe selection is incorrect; safety operation constraints (such as avoiding direct pressure from the probe on the patient's wound or drainage tube area); and finally, by integrating process logic, operating rules, and equipment interaction feedback (such as mechanical feedback during probe movement and real-time ultrasound image response), it fully recreates the entire process of clinical ultrasound examination from pre-examination preparation to core operations. It simulates the operational norms, decision-making logic, and emergency handling in a real clinical environment, enabling trainees to conduct practical training while adhering to clinical standards. This ensures that operational behavior is highly consistent with clinical reality, providing an interactive simulation environment for ultrasound skills training that closely matches clinical practice.

[0032] In this embodiment of the application, the virtual ultrasound practice module 300 includes: Figure 4 As shown, the probe simulation unit, image rendering unit, and real-time interactive feedback unit are included.

[0033] The probe simulation unit simulates the operation of commonly used clinical ultrasound probes, performing multi-dimensional operations such as movement, rotation, pressing, and tilting, and switching between different ultrasound planes; the image rendering unit generates ultrasound images of corresponding anatomical sites in real time based on physical acoustic models and clinical ultrasound image characteristics, simulating the image manifestations under different pathological conditions; the real-time interactive feedback unit provides feedback on the operation status through visual and auditory prompts, and provides real-time warnings for violations.

[0034] It is understood that the embodiments of this application replicate the multi-dimensional operations of clinical ultrasound probes, such as movement, rotation, pressing, and tilting, as well as the ultrasound section switching function, through the probe simulation unit. Combined with the advantages of the image rendering unit, which generates corresponding anatomical ultrasound images in real time based on physical acoustic models and clinical ultrasound image features, and simulates the image manifestations under different pathological conditions, the real-time interactive feedback unit provides visual and auditory prompts to provide feedback on the operation status and provides real-time warnings for violations. This allows trainees to repeatedly practice standardized operation skills of ultrasound probes in simulated scenarios without clinical risks, become familiar with the methods of obtaining ultrasound sections of different anatomical locations, and master the key points of ultrasound image recognition under various pathological conditions. Real-time feedback can promptly correct non-standard operating behaviors, enhance operational proficiency and clinical decision-making ability, effectively reduce the error rate in real clinical practice, and improve training efficiency and quality.

[0035] It should be noted that, based on a physical acoustic model and clinical ultrasound image features, ultrasound images of corresponding anatomical sites are generated in real time, simulating image manifestations under different pathological conditions. First, a physical acoustic mathematical model conforming to the propagation laws of clinical ultrasound is constructed, incorporating propagation characteristic parameters of sound waves in different human tissues (such as skin, muscle, fat, organs, bones, and blood vessels), including sound velocity, density, acoustic impedance, attenuation coefficient, and scattering coefficient. The physical laws of sound wave incidentness, reflection, refraction, scattering, and attenuation are clarified, and a simulation mechanism for the interaction between ultrasound signals and tissues is established, such as strong echoes generated by boundary reflections, diffuse echoes formed by tissue scattering, and far-field grayscale darkening caused by sound wave attenuation. Simultaneously, a large-scale clinical ultrasound image feature library is constructed, covering… Standard imaging features of normal anatomical sites (such as homogeneous intermediate echogenicity of the liver, clear corticomedullary boundary of the kidney, and echogenicity of the cardiac chambers) and specific imaging manifestations of various pathological conditions are extracted. These include inflammation (such as patchy hypoechoic areas of pneumonia, and thickened gallbladder wall with diffuse hyperechoic areas of cholecystitis), space-occupying lesions (such as irregular hyperechoic masses of liver cancer, and anechoic areas with clear capsules of cysts), effusions (such as anechoic dark areas of pleural effusion, and free fluid dark areas of ascites), and stones (such as hyperechoic masses with posterior acoustic shadowing of gallstones). Key parameters such as grayscale range, echogenicity features, boundary morphology, internal echo homogeneity, and blood flow signal distribution (for color Doppler ultrasound) are extracted for each pathological condition. Subsequently, combined with the high-fidelity 3D human body model constructed by the scene modeling unit, the operating parameters of the probe simulation unit (such as probe position, angle, depth, and scanning mode) are acquired in real time to locate the anatomical location, tissue layer, and pathological area corresponding to the current scan. The propagation path, signal reflection intensity, propagation time, and attenuation degree of the sound wave emitted from the probe in the target tissue are calculated through a physical acoustic model. Simultaneously, feature parameters corresponding to the anatomical location and pathological state in the image feature library are called up to convert the ultrasound signal obtained from the physical simulation into grayscale image data, restoring the grayscale gradient, echo distribution, artifact characteristics (such as side lobe artifacts and reverberation artifacts), and cross-sectional morphology (such as the anatomical structure presentation of transverse, longitudinal, and oblique sections) of the clinical ultrasound image. During operation... During the process, with the dynamic changes of multi-dimensional operations such as probe movement, rotation, and pressing, the model adjusts the acoustic wave propagation parameters and the tissue matching relationship of the scanning area in real time. The images are updated synchronously to fit the anatomical structure presented from the current scanning perspective. At the same time, it accurately reproduces the image differences of different pathological states, such as the changes in echo intensity of solid lesions, the changes in the dark area range corresponding to the amount of effusion, and the richness of blood flow signals in tumor lesions. This ensures that the generated ultrasound images not only conform to the laws of acoustic physics, but also are highly consistent with the ultrasound image characteristics of different anatomical sites and different pathological states in actual clinical diagnosis. It provides trainees with a real ultrasound image interpretation scenario, helps them master the key points of differentiating normal and abnormal images, and improves the professionalism and accuracy of ultrasound diagnostic skills.

[0036] Physical acoustic model formula:

[0037]

[0038]

[0039]

[0040] Where ∇²p is the Laplace operator for p; p is the pressure; and c is the wave velocity; Let p be the second partial derivative of p with respect to time t; β be the correlation coefficient; and ρ0 be the initial density. Let q be the first partial derivative of q with respect to time t; q be the source term; p(z) be the pressure at position z; p0 be the initial value of p at position z=0; e be the base of the natural logarithm; a(f0) be the attenuation coefficient related to frequency f0; z be the position coordinates; R be the reflection coefficient; ρ2 be the density of medium 2; c2 be the wave velocity in medium 2; ρ1 be the density of medium 1; c1 be the wave velocity in medium 1. Total strength; For reference strength; λ is the intensity-related parameter; r is the wavelength; θ is the distance between the observation point and the source; cos²θ is the square of the cosine of angle θ.

[0041] The construction of a large-scale clinical ultrasound image feature database consists of four steps: First, collecting ultrasound images covering all anatomical locations, normal / various pathological states (inflammation, space-occupying lesions, effusion, etc.), and different equipment / populations to ensure data comprehensiveness; second, performing standardized preprocessing such as denoising and format unification on the original images to eliminate interference factors; third, having a professional medical team label the anatomical location, lesion type, and specific ultrasound features (echo, boundaries, etc.) of the images according to unified standards, and associating them with clinical information and tissue physical parameters; fourth, extracting clinically visible features and deeply hidden features through manual and AI methods, storing them in a structured classification, and establishing a dynamic update mechanism to continuously supplement new cases and optimize data, ultimately forming a core database supporting ultrasound image generation and pathological simulation.

[0042] For example, in the practical training of vascular ultrasound examination nursing, the probe simulation unit provides linear array probes (suitable for superficial vessels such as the carotid artery and dorsalis pedis artery) and convex array probes (suitable for deep vessels such as the iliac artery) for selection. Trainees need to simulate the entire operation in conjunction with the nursing process: when selecting the linear array probe to examine the carotid artery, assist the patient to lie supine and provide verbal reassurance; after positioning, move the probe laterally and longitudinally to find the positioning point and cover the examination area; rotate to obtain long / short axis sections; control the pressure and fine-tune the tilt angle; and observe and communicate simultaneously. When examining superficial vessels such as the dorsalis pedis artery in diabetic foot patients, switch to the linear array probe to assist in positioning and skin preparation; perform the same operation, but reduce the pressure and strengthen communication for elderly patients. When examining the deep iliac artery, switch to the convex array probe to assist in lateral decubitus position; apply coupling gel and perform the positioning operation. Throughout the process, the simulation unit responds in real time to switching sections and simulating patient reactions; trainees need to adjust their techniques and intervene, ultimately enabling trainees to master the skills of probe selection, patient positioning, communication, and pressure control, improving the accuracy of cooperation and patient comfort.

[0043] In this embodiment of the application, the intelligent skill assessment center 400 includes, as follows: Figure 5 As shown, there are data extraction unit, model evaluation unit, and result generation unit.

[0044] The data extraction unit is used to collect trainees' practical data, real-time generated ultrasound image features, and diagnostic conclusions. The model evaluation unit uses an improved ResNet-LSTM model to extract the spatial features of ultrasound images and the spatial trajectory features of the operation through ResNet, and captures the operation timing logic and diagnostic decision-making process through LSTM for quantitative evaluation. The results generation unit outputs skill scores and a weakness analysis report, enabling trainees to clearly identify their weaknesses in operation procedures, section recognition, and pathological diagnosis.

[0045] It is understood that the embodiments of this application comprehensively collect trainees' practical data, real-time ultrasound image features, and diagnostic conclusions through the data extraction unit. Relying on the improved ResNet-LSTM model adopted by the model evaluation unit, ResNet accurately extracts the spatial features of ultrasound images and the trajectory features of the operation space. LSTM effectively captures the temporal logic of the operation and the diagnostic decision-making process, and scientifically and quantitatively evaluates the trainees' ultrasound practical skills. Then, the results generation unit outputs an intuitive skill score and a detailed shortcoming analysis report, which clearly identifies the trainees' weaknesses in terms of operation procedure standardization, section recognition accuracy, and pathological diagnosis accuracy. This breaks through the subjectivity and one-sidedness of traditional evaluation, allowing trainees to clearly recognize their own skill shortcomings and carry out targeted reinforcement training to effectively improve the standardization of practical operation and diagnostic accuracy. It also provides the training provider with objective data support for the trainees' skill level, helping to optimize training content, adjust teaching focus, and improve the overall relevance and effectiveness of training.

[0046] It should be noted that the improved ResNet-LSTM model

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] in, Spatial characteristics of the residual block output; Input features for the residual blocks; This is a convolution operation; For batch normalization operation; Weights for attention mechanisms; For activation functions; The final spatial characteristics after fusion; For feature fusion weights; Spatial features of ultrasound images extracted by ResNet; Operational space trajectory features extracted for ResNet; for The output of the forget gate at all times; Use the Sigmoid activation function; Here is the forget gate weight matrix; for LSTM hidden state at time step; for Spatial characteristics after time-lapse integration; for and splicing operation; Forget gate bias term; for Input gate output at all times; The input gate weight matrix; for The state of candidate cells at any given time; It is the hyperbolic tangent activation function; This is the candidate cell state weight matrix; This refers to the candidate cell state bias term; for LSTM cell state at any given time; for LSTM cell state at any given time; for Output gate output at all times; This is the output gate weight matrix; for The final hidden state of the LSTM at time point; To quantify the evaluation results; The normalized activation function; This is the output layer weight matrix; The hidden state of the LSTM at the last time step; This is the output layer bias term.

[0056] In this embodiment of the application, the personalized teaching unit 500 includes, for example: Figure 6 As shown, the learning trajectory analysis unit, the scheme generation unit, and the knowledge reinforcement module are included.

[0057] The learning trajectory analysis unit records students' learning time, number of practical exercises, assessment results, and error records, analyzing students' knowledge gaps and skill deficiencies. The program generation unit generates customized training programs based on skill assessment results and learning trajectories, including targeted practical training, key theoretical learning points, and training cycle planning. The knowledge reinforcement module provides case analysis, ultrasound imaging differential diagnosis guidelines, and operation skill video reinforcement content, enabling students to learn as needed, track training effects synchronously, and dynamically adjust training programs.

[0058] It is understood that the embodiments of this application record the student's learning time, number of practical exercises, evaluation results, and error records through the learning trajectory analysis unit, and locate knowledge gaps and skill deficiencies. Based on the scheme generation unit, a customized training scheme is generated that includes targeted practical training, key theoretical learning points, and training cycle planning, in combination with the evaluation results and learning trajectory. Then, the knowledge reinforcement module provides reinforcement content such as case analysis, ultrasound image differential diagnosis guidelines, and operation skill videos. At the same time, the training effect is tracked and the scheme is dynamically adjusted, so that students can learn accurately as needed, efficiently make up for weaknesses, and effectively improve learning efficiency, mastery of ultrasound theoretical knowledge, and practical skills.

[0059] It should be noted that, based on the individualized learning needs of trainees, the program provides three core reinforcement content categories: case studies (covering typical and complex ultrasound clinical scenarios to help trainees deeply integrate theoretical knowledge with practical application), ultrasound imaging differential diagnosis guidelines (systematically outlining the imaging characteristics, key points of differentiation, and diagnostic approaches for different diseases to reduce the risk of misdiagnosis and missed diagnosis), and operation skills videos (intuitively demonstrating standardized operating procedures, key techniques, and common problem correction methods to standardize practical actions). Trainees can choose learning content independently based on their knowledge gaps, skill deficiencies, and learning progress, and receive supplementary training as needed. At the same time, the module tracks trainees' learning progress and training effect feedback (such as practical operation pass rate, theoretical test scores, etc.), and dynamically optimizes and adjusts the content focus, training intensity, and schedule of the customized training program based on the tracking data.

[0060] For example, a trainee participating in a specialized ultrasound training program focused on cardiovascular ultrasound. The learning trajectory analysis unit collected and integrated the trainee's comprehensive learning data in real time through the system backend: The trainee invested a total of 28 hours in the core module of cardiac ultrasound, including 15 hours of theoretical knowledge learning (including sub-chapter on valve anatomy and hemodynamics principles) and 13 hours of practical exercises. This included 12 practical ultrasound training sessions covering key areas such as the mitral and aortic valves (including 5 simulated phantom model exercises and 7 clinical case simulation exercises). The trainee failed to meet the standards in 3 practical assessments (due to non-standard acquisition of the mitral valve short-axis section, blood flow spectrum measurement angle deviation exceeding the threshold, and incomplete observation of valve morphology). In the theoretical test, the trainee answered 3 core questions incorrectly in the comprehensive cardiovascular ultrasound assessment, all focusing on the differentiation of valvular lesions using ultrasound imaging (including 1 question comparing M-mode ultrasound features of mitral stenosis and regurgitation, and 2 two-dimensional ultrasound case analysis questions differentiating between chordae tendineae rupture and prolapse). The system also recorded that the trainee repeatedly reviewed theoretical videos related to valvular lesions during the learning process but still failed to grasp the key points. The learning trajectory analysis unit, through cross-correlation and in-depth analysis of these multi-dimensional data, not only quantifies and presents the learners' learning behaviors and achievements, but also accurately uncovers the core issues behind the data. Knowledge gaps are concentrated in the identification logic of ultrasound imaging features of valvular lesions (especially the differences in morphology and hemodynamics under different pathological states), while skill deficiencies lie in the standardized operating procedures of mitral valve ultrasound examination (such as section positioning techniques and standardized methods for obtaining measurement parameters). At the same time, it was found that the scenarios in which practical skills were not up to standard highly overlapped with the knowledge points of theoretical errors. This provides comprehensive, accurate, and practical data support for the subsequent solution generation unit to design customized training content such as "specialized theoretical courses on the identification of valvular lesions + standardized practical training for mitral valves + one-on-one error correction guidance".

[0061] This application proposes a critical care ultrasound imaging training system. Through a clinical data acquisition module, it comprehensively collects critical care clinical case data, ultrasound image materials, and standardized diagnostic and treatment procedures, providing multi-dimensional standardized foundational data for training scenario construction and skills assessment. The training scenario construction module, based on anatomical standards and clinical data, constructs a high-fidelity three-dimensional training scenario and achieves precise mapping between ultrasound images and the three-dimensional model, overcoming the limitations of insufficient realism in traditional training scenarios. The virtual ultrasound practice module accurately simulates the multi-dimensional operational logic of ultrasound probes and the clinical ultrasound image generation mechanism. Combined with real-time interactive feedback, it enhances the immersion and standardization of practical training, strengthening the practical application of operational skills. The intelligent skills assessment center, utilizing an improved ResNet-LSTM model, integrates practical data and image features to achieve quantitative assessment of operational standardization and diagnostic accuracy, enhancing the accurate identification of skill gaps. Personalized teaching units, based on skills assessment results and learning trajectory analysis, generate customized training plans and knowledge reinforcement content, helping trainees accurately fill knowledge gaps and skill deficiencies, effectively improving training relevance, learning efficiency, and clinical ultrasound diagnostic and treatment application capabilities. This solves the problems of insufficient dynamic scene simulation and poor realism in existing technologies.

[0062] The following will illustrate a critical care ultrasound imaging training system through a specific embodiment, such as... Figure 7 As shown, it includes: The nursing ultrasound training center of the intensive care unit (ICU) of a top-tier hospital had long faced numerous challenges with traditional training models: fixed anatomical structures in physical ultrasound models made it impossible to simulate complex pathological conditions; scarce and unrepeatable clinical case resources; and reliance on instructors' subjective judgment for practical assessments, lacking quantitative standards, resulting in low training efficiency and uneven skill mastery among trainees. To address these issues, the center introduced a critical care ultrasound imaging training system. This system, through multi-module collaboration, automates the entire process from data acquisition to personalized training, significantly improving the ultrasound operation and diagnostic capabilities of critical care nurses.

[0063] Deployment and Data Processing Flow of Clinical Data Acquisition Module The clinical data acquisition module, serving as the core data support source of the system, comprises a case data acquisition unit, an image material integration unit, a standardization unit, and a data transmission and storage sub-unit, constructing a comprehensive data resource library covering "cases-images-standards." The case data acquisition unit interfaces with the Hospital Information System (HIS), Electronic Medical Record System (EMR), and Intensive Care Information System (ICUIS) to collect structured data from critically ill patients across the hospital over the past five years. This data includes basic information (age, gender, underlying diseases), disease severity (APACHE II score, SOFA score), ultrasound examination indications (differentiation of shock causes, organ function assessment, etc.), clinical diagnostic conclusions (sepsis, acute respiratory distress syndrome, etc.), and prognostic data (length of hospital stay, outcome). A total of 1200 valid cases have been collected, encompassing 32 typical critical illness scenarios such as severe infection, cardiovascular failure, and multiple organ dysfunction. The image data integration unit utilizes the hospital's ultrasound department's PACS system to collect ultrasound images of different anatomical locations (heart, lungs, abdomen, blood vessels) and different pathological states. This includes 45 standard sections (such as the four-chamber view of the heart and the B-line view of the lungs) and 68 abnormal sections (such as pericardial effusion and pulmonary consolidation). Each section is labeled with the lesion location, characteristic description, and diagnostic conclusion. Simultaneously, the unit records practical procedures using a high-definition ultrasound diagnostic instrument (model: GE Vivid E95), generating 150 video clips with a resolution of 1920×1080 and a frame rate of 25fps. The standardization unit, comprised of a working group of 3 critical care medicine experts and 2 ultrasound physicians, reviewed various authoritative literature and extracted standardized diagnostic and treatment procedures for 18 types of critical care scenarios, forming a structured document containing operational steps, precautions, and quality control standards. All data is processed by the data transmission and storage subunit: transmitted to the local server via a medical-grade Ethernet (transmission rate 1000Mbps), and after data cleaning (removing duplicate cases and correcting image annotation errors) and format conversion (unifying images to DICOM format and documents to PDF format), it is stored in an encrypted database (using the AES-256 encryption algorithm) to ensure data security and standardization, providing high-quality data support for the construction of subsequent training scenarios.

[0064] Implementation details of the training scenario construction module The training scenario construction module creates high-fidelity, high-fidelity virtual training scenarios through the collaborative operation of scenario modeling units and interactive logic configuration units, achieving "digital replication of clinical scenarios". The scene modeling unit is based on a standard CAD model of human anatomy and combines high-precision 3D laser scanning technology (scanning device: FaroFocus S70, scanning accuracy ±0.1mm) to acquire point cloud data of the surface and internal anatomical structures of critically ill patients in typical positions (supine, semi-recumbent, and lateral positions). Through point cloud registration, mesh reconstruction (using the Poisson reconstruction algorithm) and texture mapping technology, a full-size 3D human body model containing key organs such as the heart, lungs, liver, and kidneys is constructed. The geometric dimensions of the model deviate from the real human body by ≤0.3mm. For different pathological states, the model parameters are adjusted to simulate the characteristics of the lesions. For example, in the pericardial effusion scenario, the virtual volume of the pericardial cavity is adjusted (adjustable from 0-500ml) and the ultrasound appearance of the fluid-filled dark area is simulated. In the pulmonary consolidation scenario, the virtual density of lung tissue is modified (from the normal 0.3g / cm³ to 0.8g / cm³) to restore the corresponding image characteristics. The interactive logic configuration unit is based on the clinical ultrasound examination process, constructing a full-process interactive logic of "preoperative preparation - body positioning - probe operation - image acquisition": In the preoperative preparation stage, it simulates coupling agent application (highlighted in the virtual operation area) and probe selection (automatically recommending phased array or linear array probes based on the examination site); in the body positioning stage, it supports dragging virtual human joints to switch positions (response time ≤ 0.3 seconds), and simultaneously displays the impact of body position on ultrasound section acquisition (e.g., optimization of lower lobe lung section display in a semi-recumbent position); regarding operation constraint rule settings, it simulates the contact force feedback between the probe and the body surface through a physics engine (pressure range adjustable from 0-5N, vibration warning issued when exceeding the safe range), limiting excessive probe tilt (triggered warning when tilt angle > 45°), ensuring standardized operation. This module receives operation data from the virtual ultrasound operation module in real time, dynamically updating the organ position, probe posture, and other states in the scene, ensuring precise synchronization between the virtual scene and the actual operation, and reproducing the real clinical examination environment.

[0065] Virtual Ultrasound Practice Module Operation Mechanism The virtual ultrasound hands-on module, serving as the core carrier for students' practical training, consists of a probe simulation unit, an image rendering unit, and a real-time interactive feedback unit, achieving a real-time closed loop of "operation-image-feedback." The probe simulation unit employs a combination of hardware and software: On the hardware side, it is equipped with a multi-degree-of-freedom ultrasound probe simulator (model: PhantomPremium1.5), possessing four degrees of freedom: translation, rotation, pressing, and tilting. It has a built-in force feedback sensor (measurement accuracy ±0.01N), simulating the grip feel and operational resistance of different probes (phased array probes for cardiac examinations and linear array probes for vascular examinations). On the software side, a coordinate mapping algorithm (using Kalman filtering to optimize positioning accuracy, with a positioning error ≤0.2mm) synchronously maps the physical operations of the hardware simulator to the probe model in the virtual scene, achieving zero-delay linkage between "hardware operation and virtual synchronization." This allows students to obtain different ultrasound sections by adjusting the probe position and angle (e.g., adjusting the probe angle to switch from the parasternal long-axis section to the short-axis section). The image rendering unit, based on a physical acoustic model and a clinical image feature library, enables real-time generation of ultrasound images. It employs an improved wave equation (considering changes in tissue density during severe illness, with sound velocity ranging from 1450-1580 m / s) to calculate the sound wave propagation process. Combined with an attenuation formula (the attenuation coefficient is dynamically adjusted according to the pathological type, such as a 30% increase in attenuation coefficient for lung consolidation compared to normal tissue) to simulate signal attenuation, and calculates the reflected signal at the tissue interface using the reflection coefficient formula, ultimately generating ultrasound images (frame rate ≥ 30fps, resolution 1024×768). Specifically designed for the unique characteristics of severe illness scenarios, the dynamic image generation algorithm is optimized, such as simulating the impact of cardiac motion on ultrasound images when the heart rate of a shock patient is too fast (120-150 beats / min), generating dynamically fluctuating cardiac cross-sectional images. The real-time interactive feedback unit provides operational guidance through multimodal prompts: Visually, standard sectional areas are marked with green dashed lines in the virtual scene, while flashing red indicates the probe is deviating from the target area; key anatomical structures (such as the mitral valve and hepatic veins) are marked with arrows on the image. Auditorily, a "beep" sound confirms correct operation, while a voice prompt (such as "Probe pressure too high, please reduce pressure") plays when an operation is incorrect (e.g., excessive probe pressure, blurred sectional view). Furthermore, an operation log recording function automatically records data such as the trainee's sectional acquisition success rate (e.g., success rate of acquiring a four-chamber view of the heart) and operation time, providing a basis for subsequent evaluation. This module supports repeated hands-on training without relying on physical patients or models, significantly reducing training costs.

[0066] Intelligent Skills Assessment Central Algorithm Implementation The intelligent skills assessment center, serving as the core of quantitative evaluation of training effectiveness, comprises a data extraction unit, a model evaluation unit, and a result generation unit, achieving full automation of "data acquisition - feature extraction - quantitative evaluation." The data extraction unit collects trainee practical and diagnostic data in real time: practical data includes probe operation trajectory (sampling frequency 100Hz, recording X / Y / Z axis coordinates and angle changes), operation duration, number of position adjustments, and number of slices acquired; image feature data uses image segmentation algorithms (using the U-Net++ model, with a segmentation accuracy of 94%) to extract the anatomical structural integrity of ultrasound images (e.g., whether the mitral and aortic valves are clearly displayed in the heart slice) and grayscale distribution characteristics (e.g., the grayscale value range of fluid-filled dark areas); diagnostic conclusion data includes the trainee's judgment of the lesions in the images and the diagnostic basis they filled in. The model evaluation unit uses an improved ResNet-LSTM hybrid model for quantitative evaluation: The ResNet part adopts an 18-layer residual network structure, using 3 convolutional layers (kernel sizes 3×3, 5×5, 3×3), batch normalization layers, and ReLU activation function to extract 128-dimensional spatial features (such as the integrity of the cross-sectional structure and the clarity of lesion features) from ultrasound images and 64-dimensional spatial features (such as the smoothness of operation and the accuracy of probe positioning) from the operation trajectory; The LSTM part uses 2 hidden layers (256 neurons per layer) and introduces an attention mechanism (the attention coefficient is dynamically calculated using the softmax function) to capture the operational temporal logic (such as the rationality of the process of body position adjustment → probe placement → cross-sectional optimization) and the diagnostic decision-making process (such as the time distribution of thinking for observing images → identifying features → making a diagnosis); After the model was trained with 80,000 sets of labeled data (including the practical data of 500 trainees and the expert evaluation results), the evaluation accuracy reached 95% and the evaluation time was ≤1 second. The results generation unit generates a multi-dimensional assessment report based on the model output: the skills score uses a 100-point scale, covering three dimensions: operational standardization (40 points), slice recognition accuracy (30 points), and pathological diagnosis correctness (30 points); the weakness analysis report uses a combination of text and charts to clearly point out the trainees' weaknesses (such as "low success rate of obtaining short-axis cardiac slices, requiring more training in probe angle control" and "insufficient ability to differentiate between pulmonary consolidation and pleural effusion by imaging"), and links them to corresponding knowledge points and operational points, providing precise guidance for personalized teaching.

[0067] Application effect of personalized teaching units Personalized teaching units achieve precise instruction based on assessment results. These units consist of a learning trajectory analysis unit, a solution generation unit, and a knowledge reinforcement module, constructing a closed-loop teaching system of "analysis-solution-reinforcement-optimization." The learning trajectory analysis unit integrates students' full-cycle learning data through data mining technology: recording the learning time for each module (accurate to the minute, e.g., 25 hours for cardiac ultrasound), the number of practical exercises (e.g., 32 lung ultrasound practical exercises, 10 of which failed to meet the standard), assessment results (changes in scores and rankings), and error records (e.g., repeated errors on 3 questions about vascular ultrasound differential diagnosis in the theoretical test). Clustering algorithms (K-Means algorithm, with a clustering accuracy of 92%) are used to analyze the data, accurately identifying knowledge gaps (e.g., insufficient recognition of ultrasound imaging features related to severe infections) and skill deficiencies (e.g., improper operation of vascular ultrasound probe pressure), and establishing individual learning files for each student. The program generation unit generates customized training programs based on the analysis results: targeted practical training is designed to address skill deficiencies (e.g., "precise acquisition of cardiac sections" training is arranged for trainees with insufficient probe angle control, 2 sets per day, 10 times per set); theoretical learning focuses are clarified to address knowledge gaps (e.g., a special course on "ultrasound differential diagnosis of severe lung diseases" is recommended for trainees with insufficient image identification ability); the training cycle is designed in a step-by-step manner (e.g., 1 week for the basic stage to master standard section acquisition, 2 weeks for the advanced stage to strengthen pathological image identification, and 1 week for the consolidation stage to conduct a comprehensive simulation assessment). The knowledge enhancement module provides diverse learning resources: the case analysis library contains 100 typical severe cases (including textual case information, ultrasound images, diagnostic procedures, and expert commentary), supporting searches by disease type and difficulty level; the ultrasound imaging differential diagnosis guide is presented in an interactive document format, allowing users to click and view comparative images of different diseases (such as comparison of imaging characteristics between pericardial effusion and pleural effusion) and key differentiation points; the operation skills video library contains 50 high-definition videos (1920×1080 resolution) demonstrated by experienced ultrasound physicians, showcasing standard operating procedures (such as "Ultrasound-guided central venous catheterization procedures") and common error corrections (such as "Correction methods for blurred sections caused by excessive probe tilt"). The module tracks learner learning data in real time (such as video completion rate and case practice accuracy rate). When a learner's pass rate for a certain type of training is ≥90%, the difficulty of the program is automatically adjusted; if the pass rate is <60%, basic training content is increased and the training cycle is extended, achieving "on-demand teaching and dynamic optimization" to ensure training effectiveness.

[0068] In summary, this application's embodiments provide solid support by integrating case, image, and standardized data into a high-quality resource library through a clinical data acquisition module; a training scenario construction module creates a high-fidelity virtual scenario to replicate the clinical environment, laying the foundation for practical training; a virtual ultrasound practice module, through hardware and software synergy, supports repeated training for trainees without relying on physical patients or models, significantly reducing training costs; an intelligent skills assessment center, relying on an improved ResNet-LSTM model, automatically collects and analyzes practical and diagnostic data, quantitatively outputting multi-dimensional assessment reports, avoiding the subjectivity of manual assessment, and accurately identifying weaknesses; personalized teaching units generate customized training plans based on assessment results and learning trajectories, dynamically optimizing training content and cycles by combining diversified reinforcement resources, achieving on-demand teaching. This comprehensively addresses the pain points of traditional training, such as fixed physical model structures, scarce case resources, and lack of standardized assessments, significantly improving trainees' ultrasound operation standardization, image differential diagnosis accuracy, and skill mastery efficiency, ensuring a balanced improvement in training quality, and providing efficient support for the training of critical care ultrasound professionals.

[0069] Next, referring to the accompanying drawings, a training method for intensive care ultrasound imaging based on an embodiment of this application is described.

[0070] like Figure 8 As shown, this method for training critical care ultrasound imaging includes the following steps: In step S101, critical clinical case data, ultrasound imaging materials, and diagnostic and treatment process specifications are acquired.

[0071] It is understood that the embodiments of this application, by acquiring critical care clinical case data, ultrasound image materials, and standardized diagnosis and treatment procedures, enable the training scenarios to accurately replicate the real clinical environment, and the virtual ultrasound practice to generate clinically relevant image manifestations and operational feedback. This allows for real-time capture of the diagnosis and treatment logic, pathological image characteristics, and standardized operational points of different critical care scenarios. This provides high-quality data support for the high-fidelity construction of subsequent training scenarios, precise operation of virtual practice, intelligent skills quantitative assessment, and personalized training program customization. It also ensures precise control over the clinical relevance of training content and standardized support for the entire training process. This addresses the pain points of traditional training, such as scarce case resources, limited scenarios, and inconsistent operational standards, improving the relevance and practical adaptability of training, and ensuring that trainees master ultrasound operation and diagnostic skills that meet clinical needs.

[0072] In step S102, a high-fidelity three-dimensional training scenario is constructed based on severe clinical case data, ultrasound imaging materials, and diagnostic and treatment process specifications.

[0073] Among them, the high-fidelity 3D training scenario refers to a virtual environment for critical care ultrasound training, which is constructed based on critical care clinical data, anatomical standards and treatment guidelines, and is built through 3D modeling, rendering and other technologies. It includes highly realistic human anatomical structures, pathological features and clinical operation logic and supports dynamic interaction.

[0074] It is understood that the embodiments of this application, by accurately reproducing human anatomical structures, different critical illness pathological features, and clinical ultrasound examination operation logic, provide trainees with a practical training platform that closely resembles the real clinical environment through dynamic interactions such as body positioning and probe operation. This not only solves the pain points of traditional training, such as fixed anatomical structures of physical models, inability to simulate complex pathological states, and scarcity of clinical case resources, but also ensures the standardization of training through real-time synchronization of operational constraint rules and scenario states. It allows trainees to repeatedly practice the entire ultrasound examination process under different critical illness scenarios, deepen their understanding of anatomical relationships, pathological imaging features, and operational logic, improve their practical skills and clinical scenario response capabilities, achieve a high degree of adaptation between training scenarios and clinical reality, and ensure the practicality and standardization of trainees' skill mastery.

[0075] For example, after a trainee in critical care ultrasound training completed a practical exercise in a high-fidelity 3D training scenario of "severe infection complicated by pulmonary consolidation," the improved ResNet-LSTM model first extracted bimodal spatial features from the trainee's practical data using ResNet—identifying spatial features such as solid echoes in the pulmonary consolidation area, B-line distribution, and pleural thickening from the real-time generated ultrasound images (judging 92% of the section integrity), and extracting spatial features such as positioning accuracy (e.g., positioning deviation of the lower lobe posterior basal segment section ≤0.5mm), pressure stability (avoiding excessive pressure that could cause chest pain in the patient), and body position assistance adaptability from the probe operation trajectory data; then, the LSTM captured the temporal logic and analyzed "body position adjustment (assisting in a semi-recumbent position and guiding deep breathing, taking 10 seconds) → probe placement (positioning in the right lower abdomen posterior axillary region)." The model evaluates the rationality of the nursing operation process of "line area (6 seconds) → section optimization (fine-tuning the tilt angle to avoid rib obstruction, 13 seconds)" and the diagnostic decision sequence of "image observation (12 seconds) → feature recognition (combining the patient's respiratory cycle to determine the range of consolidation, 9 seconds) → diagnostic conclusion input (6 seconds)". The final model outputs a skill score of 80 points. The shortcoming analysis clearly points out that "the body position was not raised enough to raise the head of the bed by 30°, resulting in incomplete display of the lower lobe section" and "the diagnosis did not combine the patient's shallow and rapid breathing (32 breaths / min) to determine the correlation of pulmonary ventilation function". Compared with the traditional manual assessment, which only gives a vague conclusion of "unskilled operation", this model realizes the quantification and precise traceability of the assessment, and provides a clear basis for the subsequent customized "lower lobe section body position and breathing coordination guidance special training".

[0076] In step S103, based on the high-fidelity 3D training scenario, the operation logic of the ultrasound probe is simulated, and ultrasound images of the corresponding anatomical sites are generated in real time. By combining the practical data and the real-time generated ultrasound images of the corresponding anatomical sites with the improved ResNet-LSTM model, the standardization of operation and diagnostic accuracy are evaluated, and the skills assessment results are obtained.

[0077] Among them, the improved ResNet-LSTM model is a deep learning model adapted to ultrasound skills assessment scenarios. It extracts spatial features of ultrasound images and probe operation trajectory through ResNet, captures the temporal logic of operation and diagnostic decision-making process through LSTM, and integrates dual-modal spatial-temporal information to realize the quantitative assessment of trainees' practical skills and diagnostic capabilities.

[0078] It is understood that the embodiments of this application accurately extract the spatial features of ultrasound images and probe operation trajectories generated in real time in a high-fidelity 3D training scenario using ResNet, effectively capture the trainee's operational temporal logic and diagnostic decision-making process with the help of LSTM, and integrate bimodal spatial and temporal information to conduct a comprehensive quantitative evaluation of operational standardization and diagnostic accuracy. This avoids the pain points of traditional manual evaluation, such as subjectivity, single evaluation dimensions, and inability to link the operation process with the diagnostic results. It can also accurately locate the trainee's weaknesses in areas such as slice acquisition, temporal operation logic, and pathological image recognition, providing precise data support for the customization of subsequent personalized training programs, improving the relevance of training and the efficiency of trainee skill improvement, and ensuring the objectivity and guiding value of the evaluation results.

[0079] For example, during a practical training session on gastric ultrasound nursing for a trainee in a high-fidelity 3D training scenario for "chronic gastritis complicated by gastric mucosal thickening," the scenario presented a highly realistic virtual human body in a fasting supine position, showcasing the stomach and surrounding anatomical structures (clearly distinguishing the antrum, body, pylorus, and their relationships with adjacent organs such as the liver, pancreas, and duodenum). It also precisely simulated the echogenicity changes in the thickened gastric mucosa area (corresponding to homogeneous enhanced echogenicity in ultrasound images). The trainee first needed to complete core nursing procedures: assisting the virtual patient in verifying their fasting status (simulating 8 hours of clinical fasting for confirmation), instructing them to drink 300ml of warm water as a contrast agent, and informing them to "remain seated for 5 minutes after drinking water to facilitate gastric filling." Subsequently, the trainee adjusted their position (knees bent and abdomen relaxed in the supine position, left lateral decubitus position for observing the gastric fundus, and sitting position to screen for gastroesophageal reflux-related manifestations), while simultaneously using "position adjustment" as a guide. If abdominal distension occurs, speak up immediately to strengthen communication; then, use the handheld probe simulator to control the pressure on different gastric areas (gentle pressure on the antrum to avoid stimulating spasms, and slightly more force on the fundus to overcome gas interference) and fine-tune the tilt angle (within ±20°) to find the boundary between the thickened and normal areas of the gastric mucosa. When the probe pressure is too strong (>4N) or the tilt is excessive (>35°), a red flashing prompt is triggered. The scenario can also be switched to the "chronic gastritis with gastric retention" scenario to compare the difference between the echo of gastric fluid and the appearance of the mucosa. This method does not rely on rare clinical cases or fixed physical models, and allows for repeated practice of core nursing skills such as pathological identification, body positioning, drinking guidance, and communication and reassurance in a clinically close interactive environment. It can efficiently familiarize patients with the anatomical relationships, pathological features, and nursing cooperation skills in the gastric ultrasound scenario, and improve the standardization of practical operation and patient adaptability.

[0080] In step S104, a customized training plan and knowledge reinforcement content are generated based on the skills assessment results and the learner's learning trajectory.

[0081] Understandably, this application's embodiments, based on the quantitative shortcomings identified in the skills assessment results, such as operational standardization and diagnostic accuracy, and combined with historical data such as learning time, number of practical operations, and error records from the learning trajectory, generate customized training plans and knowledge reinforcement content. This approach precisely addresses trainees' skill deficiencies and knowledge gaps through targeted practical training, focused theoretical learning, and planned training cycles. Furthermore, it meets individual learning needs with adaptable reinforcement content such as case studies, ultrasound imaging differential diagnosis guidelines, and operational skill videos. Simultaneously, it tracks training effectiveness and dynamically adjusts the plan, allowing trainees to concentrate on overcoming weaknesses, avoiding ineffective repetitive learning, and efficiently improving their depth of ultrasound theoretical understanding, practical skills standardization, and clinical diagnostic adaptability.

[0082] According to the embodiments of this application, a training method for critical care ultrasound imaging is proposed. A clinical data acquisition module comprehensively collects critical care clinical case data, ultrasound image materials, and standardized diagnostic and treatment procedures, providing multi-dimensional standardized basic data for training scenario construction and skills assessment. The training scenario construction module, based on anatomical standards and clinical data, constructs a high-fidelity three-dimensional training scenario and achieves precise mapping between ultrasound images and the three-dimensional model, overcoming the limitations of insufficient realism in traditional training scenarios. A virtual ultrasound practice module accurately simulates the multi-dimensional operation logic of ultrasound probes and the clinical ultrasound image generation mechanism. Combined with real-time interactive feedback, it enhances the immersion and standardization of practical training, strengthening the practical application of operational skills. An intelligent skills assessment center, using an improved ResNet-LSTM model, integrates practical data and image features to achieve quantitative assessment of operational standardization and diagnostic accuracy, enhancing the accurate identification of skill gaps. Personalized teaching units, based on skills assessment results and learning trajectory analysis, generate customized training plans and knowledge reinforcement content, helping trainees accurately fill knowledge gaps and skill deficiencies, effectively improving training relevance, learning efficiency, and clinical ultrasound diagnostic and treatment application capabilities. This solves the problems of insufficient dynamic scene simulation and poor realism in existing technologies.

[0083] The following will illustrate a training method for intensive care ultrasound imaging through a specific embodiment, such as... Figure 9 As shown, it includes: A nursing ultrasound training center in the intensive care unit of a tertiary hospital was selected as the application scenario. Using 1200 critical care clinical cases (including 32 typical scenarios such as sepsis and acute respiratory distress syndrome) as the core training vehicle, a comprehensive training system was established. During the data acquisition phase, a multi-source data acquisition network was constructed: case data was collected through interfaces of the Hospital Information System (HIS), Electronic Medical Record System (EMR), and Intensive Care Information System (ICUIS) to collect basic patient information (age 18-85 years, underlying diseases including cardiovascular disease, diabetes, etc.), disease severity (APACHE II score 4-32), ultrasound examination indications (shock identification, organ function assessment, etc.), and prognostic data (hospital stay 3-60 days). Data anonymization technology (removing names, ID numbers, and other private information) was used to ensure compliance. Ultrasound image materials were collected using the ultrasound department's PACS system, covering cardiac... The ultrasound system covers six major anatomical sites, including the heart, lungs, and abdomen, and includes 45 standard sections (such as the four-chamber view of the heart and the B-line view of the lungs) and 68 abnormal sections (such as pericardial effusion and pulmonary consolidation). Each section is labeled with the location of the lesion, grayscale characteristics, and diagnostic conclusion. Simultaneously, 150 practical videos (1920×1080 resolution, 25fps) were recorded using a GE Vivid E95 high-definition ultrasound diagnostic instrument. The standardized diagnosis and treatment process was developed by a working group consisting of 3 critical care medicine experts and 2 ultrasound physicians. They reviewed various authoritative literature and extracted standardized operating procedures for 18 types of critical care scenarios (including 12 key steps such as body positioning, probe selection, and section acquisition). Data processing employs a three-tiered workflow of "cleaning-labeling-standardization": duplicate cases are removed using the Python Pandas library (duplicate rate <3%), and image labeling errors are corrected (error rate <2%). The LabelImg tool is used to annotate anatomical structures in ultrasound images (labeling accuracy reaches 96%). Image data is converted to the DICOM standard format, case data is organized into JSON format, and standard documents are exported as PDF format. Data is transmitted to a local server via a dedicated medical Ethernet network (transmission rate 1000Mbps), stored in a MySQL database using AES-256 encryption, and high-frequency practical videos are stored in an InfluxDB time-series database (retaining 90 days of training data), thus constructing a complete data traceability and security system.

[0084] The construction of high-fidelity 3D training scenarios adopts a three-step method of "anatomical modeling - pathological simulation - interactive configuration" to achieve digital replication of clinical scenarios. The anatomical modeling stage adopted a dual-source fusion scheme of "CAD benchmark + laser scanning": a standard CAD anatomical model (containing 206 bones and 12 core organs) was imported from the Chinese Digital Human Body Database. Topology optimization was performed using Blender software, and non-critical features such as blood vessels with a diameter <2mm were deleted. The Catmull-Clark subdivision algorithm was used to simplify the surface, reducing the number of polygons in the model from 8 million to 2 million. A FaroFocus S70 laser scanner was used to perform 360° scanning on the human body model in three typical critical care positions (supine, semi-recumbent, and lateral decubitus) to obtain point cloud data (point density 100 points / mm²). The model was registered with the CAD model using the ICP algorithm of GeomagicWrap software (registration error <0.03mm). The anatomical gaps of key parts such as the thoracic cavity and abdominal cavity were corrected (e.g., the actual gap between the lung and chest wall was 0.8mm, while the model originally set it to 0.5mm). Finally, a 1:1 scale geometric twin was constructed in Unity3D, with the geometric accuracy of key anatomical structures (such as the mitral valve and hepatic vein) ≤0.3mm. The pathological simulation dynamically adjusts model parameters based on clinical data: In the pericardial effusion scenario, the volume of pericardial fluid is adjusted through a virtual volume control module (continuously adjustable from 0-500ml), and the gray value distribution of the fluid-filled dark area is simulated (50-80HU) using the ultrasound reflection coefficient formula; In the pulmonary consolidation scenario, the virtual density parameter of lung tissue is modified (gradually adjusted from the normal 0.3g / cm³ to 0.8g / cm³), and the ultrasound attenuation coefficient is adjusted simultaneously (30% higher than normal tissue) to restore the solid echo appearance of the consolidation area; For the shock scenario, a dynamic heart rate adjustment function is set (adjustable from 120-150 beats / min) to simulate the dynamic effect of rapid cardiac motion on ultrasound images. The interactive logic configuration aligns with clinical operating procedures: In the preoperative preparation phase, a highlighted prompt guides the virtual application of coupling gel (application range error ≤2mm), automatically recommending probe types based on the examination site (phased array probes recommended for cardiac examinations, linear array probes recommended for vascular examinations); In the body positioning phase, mouse dragging of virtual human joints enables posture switching with a response time ≤0.3 seconds, synchronously displaying the impact of body position on the cross-section (e.g., a 20% increase in the display range of the lower lobe lung cross-section in a semi-recumbent position); Operational constraints are implemented using the NVIDIA PhysX physics engine, simulating the contact force feedback between the probe and the body surface (pressure range adjustable from 0-5N, accuracy ±0.01N). When the probe tilt angle >45° or the pressure >5N, a red flashing warning and vibration feedback (vibration frequency 2Hz) are triggered to ensure standardized operation. Scene rendering utilizes Unity3D's URP rendering pipeline, configured with real-time global illumination (baking accuracy 512 texels / m), and employs a PBR material system to simulate the optical properties of tissues such as skin and muscle, maintaining shadow quality and scene smoothness at 60fps.

[0085] Virtual ultrasound hands-on training relies on a closed-loop system of "hardware simulation - image rendering - real-time feedback" to achieve immersive training. The hardware equipment adopts a multi-degree-of-freedom ultrasound probe simulator linked with the virtual scene: equipped with a Phantom Premium 1.5 probe simulator, it has four degrees of freedom of operation: translation, rotation, pressing, and tilting, and has a built-in force feedback sensor (measurement accuracy ±0.01N). It can simulate the holding resistance of three commonly used clinical probes: phased array, linear array, and convex array (phased array probe resistance 1.2N, linear array probe resistance 0.8N). The Kalman filter algorithm realizes the coordinate mapping between hardware operation and virtual probe (positioning error ≤0.2mm), supports continuous movement of the probe on the virtual body surface (movement resolution 0.1mm), and adjusting the probe angle can realize the switching of the section (e.g., the switching response time between the parasternal long axis section and the short axis section is ≤0.1 seconds). Image rendering is generated in real time based on a physical acoustic model: an improved ultrasonic dynamic equation is used to calculate sound wave propagation (tissue sound velocity range 1450-1580m / s), and the attenuation formula (α(f)=0.5f+0.1, f is the ultrasonic frequency) is combined to simulate signal attenuation. The reflection coefficient formula is used to calculate the reflection signal at different tissue interfaces, and finally an ultrasound image is generated (resolution 1024×768, frame rate ≥30fps). Dynamic rendering algorithms are optimized for critical care scenarios, such as simulating the lung movement of mechanically ventilated patients (respiratory rate 12-20 breaths / min), generating lung ultrasound images that fluctuate with respiration, and the boundaries of the fluid-filled dark area show slight deformation with respiration. Real-time interactive feedback employs a multimodal prompting mechanism: visually, standard slicing ranges are marked with green dashed lines, red arrows indicate probe adjustment direction, and key lesion areas (such as pericardial effusion dark areas) are marked with yellow boxes on the image; auditorily, a "beep" sound confirms correct operation, and a voice prompt, "Please adjust the probe angle to maintain a clear slice," plays when the slice is blurry; the operation log automatically records data such as slice acquisition success rate (e.g., success rate of four-chamber cardiac slice acquisition), operation time (average time for a single scenario ≤ 3 minutes), and number of violations, with a sampling frequency of 100Hz, providing comprehensive data support for subsequent evaluation. The training process supports scenario switching (e.g., switching from "pericardial effusion" to "pulmonary consolidation with pleural effusion") and repeated drills, without relying on physical patients or models, and a single device can support 20 trainees per day for rotation training.

[0086] The intelligent skills assessment uses an improved ResNet-LSTM model to achieve full automation of "data extraction - feature fusion - quantitative assessment". The data extraction unit simultaneously collects three types of core data: practical data includes probe operation trajectory (X / Y / Z axis coordinates and angle changes), number of body position adjustments (standard ≤3 times), and number of slices acquired (≥5 effective slices in a single scene); image feature data uses the U-Net++ segmentation model (segmentation accuracy 94%) to extract 16-dimensional features such as anatomical structural integrity (e.g., whether the heart slice clearly shows the mitral valve and aortic valve) and lesion feature clarity (e.g., the identification of the boundary of the fluid-filled dark area); diagnostic conclusion data includes the lesion type, diagnostic basis, and confidence score filled in by the trainee. The model structure adopts an architecture of "ResNet spatial feature extraction + LSTM temporal modeling + attention mechanism": The ResNet part uses an 18-layer residual network, which extracts 128-dimensional spatial features from ultrasound images and 64-dimensional spatial features (including indicators such as positioning accuracy and operation smoothness) from the operation trajectory through 3 convolutional layers (convolutional kernel size 3×3, 5×5, 3×3), batch normalization layers and ReLU activation function; The LSTM part uses 2 hidden layers (256 neurons per layer) with a dropout coefficient of 0.3 to capture the temporal logic of operation (such as the rationality of the process of body position adjustment → probe placement → section optimization) and the diagnostic decision process (such as the time distribution of image observation → feature recognition → diagnostic conclusion); An attention mechanism is introduced (the attention coefficient is dynamically calculated through the softmax function) to strengthen the weight of key temporal segments (such as the section optimization stage) and core features (such as lesion boundary features). Model training was completed on an NVIDIA Tesla V100 graphics card, using 80,000 sets of labeled data (including practical data from 500 trainees and expert evaluation results) for 50 epochs. The Adam optimizer learning rate was 0.001, and the mean squared error (MSE) loss function was eventually reduced to 0.002. The test set evaluation accuracy reached 95%, and the single-example evaluation time was ≤1 second. The evaluation results generated a multi-dimensional report: the skill score used a 100-point scale, with scores assigned to operational standardization (40 points), section recognition accuracy (30 points), and pathological diagnosis correctness (30 points); the weakness analysis report used text and line graphs to show the weak points, such as "success rate of obtaining cardiac short-axis sections 65% (standard ≥80%), probe angle control needs to be strengthened" and "accuracy rate of differentiating pulmonary consolidation from pleural effusion 70%, it is recommended to strengthen image feature learning," and associated with the corresponding knowledge point numbers and operation key point videos.

[0087] Personalized training program generation and optimization employ a closed-loop system of "trajectory analysis - program customization - dynamic adjustment". The learning trajectory analysis unit integrates trainees' full-cycle data through data mining: recording the learning time of each module (e.g., an average of 25 hours of learning for the cardiac ultrasound module), the number of practical exercises (an average of 8 practical exercises per scenario), the scores of each assessment (tracking the score change trend over 30 days), and the record of incorrect questions (e.g., repeatedly making mistakes on 3 vascular ultrasound differential diagnosis questions); using the K-Means clustering algorithm (clustering accuracy of 92%) to classify the data, accurately locate knowledge gaps (e.g., "insufficient recognition of ultrasound imaging features of severe infections") and skill deficiencies (e.g., "improper operation of vascular ultrasound probe pressure"), and establish a personal learning file for each trainee. The program generation unit customizes a tiered training plan based on the analysis results: The basic stage (1 week) includes specialized training in standard section acquisition, 2 sets per day, 10 repetitions per set, accompanied by a 4-hour theoretical course on "Basic Ultrasound Section Anatomy"; the advanced stage (2 weeks) designs scenario-based training to address weaknesses, such as "Intensive Training on Precise Cardiac Section Positioning" for trainees with insufficient probe angle control, and "Case Studies in Differentiation of Severe Case Pathological Imaging" (3 typical case analyses per day) for trainees with weak diagnostic abilities; the consolidation stage (1 week) conducts comprehensive scenario assessments (including 3 types of complex severe cases). The knowledge enhancement module provides diverse resources: a database of 100 typical severe cases (searchable by disease type and difficulty level), including text cases, ultrasound images, and expert commentary; an interactive differential diagnosis guide, allowing users to view image comparisons of different diseases (e.g., grayscale comparison of pericardial effusion and pleural effusion); and 50 high-definition operation skill videos (1920×1080 resolution), demonstrating standard operating procedures and error correction methods (e.g., "Adjustment techniques for blurred sections caused by excessive probe tilt"). The dynamic optimization mechanism adjusts in real time based on training results: when the trainee's pass rate for a certain type of training is ≥90% (e.g., the success rate of obtaining the four-chamber view of the heart is ≥90%), the difficulty of the scenario is automatically increased (e.g., adding a complex scenario of "pericardial effusion combined with arrhythmia"); if the pass rate is <60%, the number of basic training sets is increased (from 2 sets per day to 3 sets) and corresponding micro-lessons on knowledge points (10 minutes / lesson) are pushed. The training effect is verified through "assessment + clinical suitability assessment": the pass rate of the graduation assessment has increased from 75% in traditional training to 92%, the time for obtaining standard sections in trainees' clinical practice has been shortened by 40%, and the accuracy of pathological diagnosis has increased by 35%; a dual-terminal platform of "local VR training + remote Web learning" has been built. Locally, the HTC VivePro2 headset is used to achieve immersive training (positioning accuracy ±1mm), and remotely, the WebGL lightweight solution supports mobile access. Administrators can view the training progress and effect in real time, realizing intelligent management and control of the entire training process.

[0088] In summary, this application's embodiments, through comprehensive multi-source data acquisition and standardized processing, combined with high-fidelity digital twin modeling and improved ResNet-LSTM model evaluation, significantly improve the training completion rate, enhance critical care ultrasound practical skills, and achieve a standard section acquisition accuracy of ±0.3mm. Virtual interactive training and dynamic optimization schemes reduce the time for trainees to acquire standard sections in clinical practice by 40% and improve the accuracy of pathological diagnosis by 35%. Simultaneously, personalized training plans ensure targeted training. A dual-end collaborative platform and intelligent full-process management improve training coverage efficiency and reduce reliance on physical resources; dynamic feedback and adaptive difficulty adjustment enable precise identification of trainees' skill gaps and closed-loop optimization of training effectiveness, controlling the incidence of non-standard practical problems, and comprehensively improving the efficiency, quality, and clinical adaptability of critical care ultrasound training.

[0089] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0090] When the processor 1002 executes the program, it implements the critical care ultrasound imaging training method provided in the above embodiments.

[0091] Furthermore, electronic devices also include: Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0092] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0093] The memory 1001 may include high-speed RAM (Random Access Memory) and may also include non-volatile memory, such as at least one disk storage.

[0094] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0095] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0096] The processor 1002 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0097] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for training critical care ultrasound imaging.

[0098] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-described method for training critical care ultrasound imaging.

[0099] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0100] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0101] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0102] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0103] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0104] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A critical care ultrasound imaging training system, characterized in that, include: The module includes a clinical data acquisition module, a training scenario construction module, a virtual ultrasound practice module, an intelligent skills assessment center, and personalized teaching units; among them, The clinical data acquisition module is used to collect data on severe clinical cases, ultrasound images, and standardized diagnosis and treatment procedures. The training scenario construction module is used to construct a high-fidelity three-dimensional training scenario based on the critical clinical case data, ultrasound image materials, and diagnosis and treatment process specifications. The virtual ultrasound practice module, based on the high-fidelity 3D training scenario, simulates the operation logic of the ultrasound probe and generates ultrasound images of the corresponding anatomical parts in real time. The intelligent skills assessment center is used to assess operational standardization and diagnostic accuracy by combining practical data with real-time generated ultrasound images of corresponding anatomical sites and an improved ResNet-LSTM model, thereby obtaining skills assessment results. The personalized teaching unit is used to generate customized training plans and knowledge reinforcement content based on the skills assessment results and the student's learning trajectory.

2. The critical care ultrasound imaging training system according to claim 1, characterized in that, The clinical data acquisition module includes a case data acquisition unit, an image material integration unit, and a standardization unit. The case data acquisition unit is used to collect basic information, disease severity, ultrasound examination indications, clinical diagnostic conclusions, and prognostic data of critically ill patients. The image material integration unit is used to collect ultrasound images of different anatomical locations and different pathological states, including standard and abnormal section images. The standardization unit is used to organize critical care ultrasound diagnosis and treatment guidelines, operating procedures, and quality control standards to form a structured diagnosis and treatment process document.

3. The critical care ultrasound imaging training system according to claim 1, characterized in that, The training scenario construction module includes a scenario modeling unit and an interaction logic configuration unit. The scenario modeling unit constructs a high-fidelity three-dimensional training scenario based on clinical data and anatomical standards, restoring the body position, anatomical structure details and pathological features of critically ill patients. The interaction logic configuration unit simulates the clinical ultrasound examination process, configures the pre-operation logic for body position adjustment, probe selection, and coupling agent use, sets operation constraint rules, and restores the clinical practice scenario.

4. The critical care ultrasound imaging training system according to claim 1, characterized in that, The virtual ultrasound operation module includes a probe simulation unit, an image rendering unit, and a real-time interactive feedback unit. The probe simulation unit simulates the operation of commonly used clinical ultrasound probes, allowing for multi-dimensional operations such as movement, rotation, pressing, and tilting, and switching between different ultrasound planes. The image rendering unit generates ultrasound images of corresponding anatomical sites in real time based on a physical acoustic model and clinical ultrasound image characteristics, simulating the image manifestations under different pathological conditions. The real-time interactive feedback unit provides feedback on the operation status through visual and auditory cues, and provides real-time warnings for violations.

5. The critical care ultrasound imaging training system according to claim 1, characterized in that, The intelligent skills assessment center includes a data extraction unit, a model evaluation unit, and a result generation unit. The data extraction unit collects the trainees' practical data, real-time generated ultrasound image features, and diagnostic conclusions. The model evaluation unit uses an improved ResNet-LSTM model, extracting spatial features of ultrasound images and spatial trajectory features of operations through ResNet, and capturing the temporal logic of operations and the diagnostic decision-making process through LSTM for quantitative assessment. The result generation unit outputs skills scores and a weakness analysis report, enabling trainees to clearly identify weaknesses in operation procedures, section recognition, and pathological diagnosis.

6. The critical care ultrasound imaging training system according to claim 1, characterized in that, The personalized teaching unit includes a learning trajectory analysis unit, a program generation unit, and a knowledge reinforcement module. The learning trajectory analysis unit records students' learning time, number of practical exercises, assessment results, and error logs, analyzing their knowledge gaps and skill deficiencies. The program generation unit generates customized training programs based on skill assessment results and learning trajectories, including targeted practical training, key theoretical learning points, and training cycle planning. The knowledge reinforcement module provides case studies, ultrasound imaging differential diagnosis guidelines, and video reinforcement content on operational skills, enabling students to learn as needed, track training effectiveness, and dynamically adjust training programs.

7. A training method for intensive care ultrasound imaging, characterized in that, The method includes: Acquire data on severe clinical cases, ultrasound images, and standardized diagnostic and treatment procedures; Based on the aforementioned critical clinical case data, ultrasound imaging materials, and standardized diagnosis and treatment procedures, a high-fidelity 3D training scenario was constructed. Based on the high-fidelity 3D training scenario, the operation logic of the ultrasound probe is simulated, and ultrasound images of the corresponding anatomical parts are generated in real time. By combining the practical data with the real-time generated ultrasound images of the corresponding anatomical parts and the improved ResNet-LSTM model, the standardization of operation and diagnostic accuracy are evaluated, and the skills assessment results are obtained. Based on the skills assessment results and the learners' learning trajectories, customized training plans and knowledge reinforcement content are generated.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the critical care ultrasound imaging training method of claim 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the critical care ultrasound imaging training method of claim 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When a computer program or instruction is executed, it implements the critical care ultrasound imaging training method of claim 7.

Images