Radiotherapy equipment science popularization display interaction system based on image recognition

Abstract

The invention relates to the technical field of medical science popularization display, and discloses a radiotherapy equipment science popularization display interaction system based on image recognition, and the system comprises a data collection and construction module which is used for collecting an entity image, a three-dimensional model and a technical document of radiotherapy equipment, and constructing a knowledge base; the image recognition module is used for inputting the acquired image into a deep learning model, extracting key semantic features of the radiotherapy equipment in the image, retaining an optimal frame and removing redundancy based on an improved clustering-non-maximum suppression algorithm, and realizing accurate recognition and category judgment of the radiotherapy equipment; the science popularization interaction module is used for matching the content of the knowledge base according to the identification result, displaying equipment information in a multi-modal form and adapting the content depth based on the user portrait; and the deployment integration module supports online platform and offline terminal deployment and triggers identification and interaction functions through image scanning or uploading. According to the method, high-precision image recognition and intelligent interactive display are realized, and the method has good universality, robustness and practical value.

Description

Technical Field

[0001] The present invention relates to the technical field of medical science popularization display, and in particular to an interactive system for popularization display of radiotherapy equipment based on image recognition. Background Art

[0002] With the advancement of modern medical technology, radiotherapy has become a core method for cancer treatment. Representative equipment includes proton therapy devices, linear accelerators, CyberKnife, and other high-precision devices. Despite the widespread use of these devices in the medical field, a significant gap in public understanding of their structural principles, working methods, and technical advantages remains, hindering the effectiveness of scientific dissemination and doctor-patient communication.

[0003] In recent years, image recognition and deep learning technologies have achieved remarkable results in medical image analysis, particularly in object detection and image semantic understanding. However, directly applying these technologies to radiotherapy equipment recognition and interactive science education scenarios still faces the following technical challenges:

[0004] The equipment structure is complex and diverse, and images are often accompanied by occlusion, angle changes and blurred boundaries, making it difficult for traditional detection algorithms to accurately locate and identify targets; the target ratio of radiotherapy equipment is often small, and the traditional IoU loss is insensitive to small targets during training, resulting in unstable gradients and affecting model convergence; the recognition task requires not only accurate classification, but also driving the subsequent multimedia interactive content display, which places higher requirements on detection accuracy and efficiency.

[0005] Therefore, it is urgent to build an integrated image recognition and science popularization display solution with precision, stability and interactive driving capabilities for radiotherapy equipment identification scenarios. Summary of the Invention

[0006] This invention aims to propose a complete and innovative algorithm framework for image recognition and popular science display of radiotherapy equipment. Through comprehensive improvements in loss function design, search strategy optimization, feature extraction structure innovation and target screening algorithm, this framework successfully constructs a high-precision image recognition and intelligent interactive display system for radiotherapy equipment. It has good versatility, robustness and practical value, and is suitable for various scenarios such as medical science exhibition halls, virtual exhibition halls, and intelligent guide systems.

[0007] To achieve the above objectives, the following technical solutions are adopted:

[0008] An interactive system for popularizing and displaying radiotherapy equipment based on image recognition, including:

[0009] The data acquisition and construction module is used to collect physical images, 3D models, and technical documentation of radiotherapy equipment, and build an image database containing equipment type labels and a knowledge base containing multi-dimensional popular science content;

[0010] The image recognition module is used to input the collected images into a deep learning model. Through a neural network combined with an improved feature extraction and target positioning algorithm, it extracts the key semantic features of the radiotherapy equipment in the image. Based on an improved clustering-non-maximum suppression algorithm, it retains the optimal frame and removes redundancy to achieve accurate recognition and classification of radiotherapy equipment.

[0011] The popular science interaction module matches knowledge base content based on recognition results, displays device information in a multimodal manner, and adapts content depth based on user profiles;

[0012] Deploy integrated modules, support deployment on online platforms and offline terminals, and trigger recognition and interaction functions through image scanning or uploading.

[0013] Furthermore, the image recognition module includes:

[0014] The target positioning unit adopts a dynamic anchor box generation mechanism and optimizes positioning accuracy by fusing the F1 score with a dynamic difficulty perception loss function;

[0015] The feature extraction unit uses a multi-scale feature extraction network with a dense connection structure and residual skip connection mechanism, combined with sub-pixel convolution upsampling to preserve image details;

[0016] The result screening unit, based on the improved clustering-non-maximum suppression algorithm, removes redundant prediction boxes through clustering grouping and confidence penalty mechanism, and outputs the final recognition result.

[0017] Furthermore, the target positioning unit includes:

[0018] The classification target calculation subunit uses the F1 score as the classification optimization target to balance precision and recall and solve the imbalance problem of positive and negative samples;

[0019] The regression target calculation subunit uses a dynamic difficulty-aware loss function as the regression optimization target. It dynamically adjusts the gradient weight of the bounding box regression based on the sample abnormality and the monotonic focusing coefficient to enhance the learning ability of small objects and samples with blurred edges.

[0020] The dual-objective fusion mechanism combines the F1 score with the dynamic difficulty perception loss function to form a multi-objective joint loss function. Gradient backpropagation is performed based on the joint loss function to synchronously update the parameters of the classification network and regression network, achieving coordinated optimization of classification and positioning results.

[0021] Anchor frame optimization strategy uses a nonlinear convergence factor to dynamically adjust the search range, performing global exploration in the early stages of iteration and local refinement and convergence in the later stages;

[0022] The anchor box generation mechanism constrains the predicted offset through the Sigmoid function and dynamically outputs the device bounding box coordinates.

[0023] Furthermore, the dynamic difficulty-aware loss function dynamically adjusts the gradient weight of the bounding box regression according to the sample abnormality and the monotonic focusing coefficient, including:

[0024] Calculate the abnormality β of the current sample, which is the ratio of the IoU loss value of the current sample to the average IoU loss value;

[0025] When β>1, it is determined to be a difficult sample and the gradient gain is amplified to strengthen the learning of small targets and fuzzy edge samples;

[0026] When β<1, it is judged as a simple sample and the gradient gain is suppressed to prevent overfitting;

[0027] The gradient gain of difficult samples is exponentially amplified by the monotonic focusing coefficient γ.

[0028] Furthermore, the feature extraction unit includes:

[0029] Data augmentation unit, which performs rotation, scaling, and copy-paste processing on the input image;

[0030] The encoding-decoding structure extracts multi-scale semantic features in the downsampling stage and restores spatial resolution through sub-pixel convolution in the upsampling stage;

[0031] The residual connection channel superimposes the shallow features with the reconstructed feature map to prevent the gradient from disappearing.

[0032] Furthermore, the result screening unit includes:

[0033] Calculate the overlap between prediction boxes based on the intersection-over-union matrix;

[0034] Iterative suppression mechanism, which uses a mask matrix to remove redundant boxes that overlap with high-confidence boxes in rounds;

[0035] The confidence penalty mechanism uses an exponential decay function to reduce the false suppression rate of occluded targets.

[0036] Furthermore, the science popularization interaction module includes:

[0037] Multi-dimensional knowledge display unit, providing equipment structure principles, treatment process animation, technology comparison charts and AR scene fusion demonstration;

[0038] Rich media output unit, supporting video playback, synchronous output of images, text and voice, and natural language question and answer interaction;

[0039] The user portrait adaptation unit distinguishes between teenagers, patients and medical staff, and dynamically adjusts the content presentation format and depth.

[0040] Furthermore, the AR scene fusion demonstration is specifically as follows:

[0041] Through image anchor alignment technology based on feature point matching or plane detection, a 3D model or dynamic path animation can be stably superimposed on the recognized device image or real scene;

[0042] For proton therapy devices, the movement trajectory of the beam from the accelerator to the tumor target is visually displayed.

[0043] Furthermore, the deployment integration module includes:

[0044] Offline deployment: touch terminals and AR devices are configured in the physical exhibition hall to support QR code scanning and linkage with physical exhibits;

[0045] Online integration provides image upload and recognition services through WeChat mini-programs and virtual exhibition halls.

[0046] Furthermore, the radiotherapy equipment includes a proton therapy device, a linear accelerator, a CyberKnife and a Gamma Knife.

[0047] Compared with the prior art, the present invention achieves the following beneficial effects:

[0048] 1. This paper proposes a PWIoU loss function: Based on the traditional IoU, it introduces a "dynamic difficulty perception mechanism". Through the combined effect of "abnormality" and "monotonic focusing coefficient", it dynamically adjusts the gradient contribution, significantly enhancing the model's learning ability for difficult samples such as small targets and blurred edges, alleviating the problem of gradient instability during training, and effectively suppressing overfitting of simple samples, thereby improving overall detection robustness and accuracy.

[0049] 2. This invention improves the Gray Wolf Optimization Algorithm: To prevent the anchor frame search process from falling into the local optimum, an improved Gray Wolf Optimization strategy is proposed, and a "dynamic nonlinear convergence factor" based on an exponential function is designed to implement an adaptive adjustment mechanism that changes the search range from global coarse exploration to local fine convergence, greatly improving the anchor frame training effect, which is particularly suitable for image positioning tasks of medical equipment with complex structures.

[0050] 3. This invention optimizes the image feature extraction structure: by introducing a "dense connection structure" and a "residual skip connection mechanism" in the network design, combined with "multi-level sub-pixel convolution upsampling", it achieves multi-scale deep semantic modeling while retaining image edges and texture details, effectively improving the model's adaptability to multi-angle and multi-resolution input images, and providing high-quality feature expression for subsequent recognition and matching.

[0051] 4. This paper constructs an improved clustered non-maximum suppression algorithm: For scenes with occlusion and densely populated targets, an integrated target screening scheme combining clustering, iterative suppression, and a score penalty mechanism is designed. This algorithm not only effectively removes redundant candidate boxes but also enhances the ability to retain occluded true targets through an exponential decay mechanism. It also supports parallel matrix calculations, ensuring high model accuracy while maintaining high inference efficiency.

[0052] In summary, the present invention has successfully constructed a high-precision image recognition and intelligent interactive display system for radiotherapy equipment through comprehensive improvements in loss function design, search strategy optimization, feature extraction structure innovation and target screening algorithm. It has good versatility, robustness and practical value, and is suitable for various scenarios such as medical science exhibition halls, virtual exhibition halls, and intelligent guide systems.

[0053] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0055] Figure 1 This is a module diagram of an interactive system for popularizing and displaying radiotherapy equipment based on image recognition according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the architecture of an interactive system for popular science display of radiotherapy equipment based on image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0059] Figure 1 This is a module diagram of an interactive system for popularizing and displaying radiotherapy equipment based on image recognition according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the architecture of an interactive system for popularizing and displaying radiotherapy equipment based on image recognition according to an embodiment of the present invention. Figure 1 and Figure 2 As shown, an interactive system 100 for popularizing and displaying radiotherapy equipment based on image recognition includes:

[0060] The data acquisition and construction module 110 is used to collect physical images, three-dimensional models and technical documents of radiotherapy equipment, and build an image database containing equipment type labels and a knowledge base containing multi-dimensional popular science content; among them, radiotherapy equipment includes proton therapy equipment, linear accelerators, CyberKnife and Gamma Knife.

[0061] S1. Demand Analysis and Data Collection

[0062] S1.1: Target users and application scenarios

[0063] In order to build a radiotherapy equipment image recognition and popular science display system with practical application value and social communication power, we first need to clarify the system's core service objects and typical application scenarios.

[0064] The target users of the system mainly include the following three categories:

[0065] Public users: especially teenagers, patients and their families, have significant blind spots in their knowledge of radiotherapy and are important targets for popular science publicity.

[0066] Medical practitioners: including radiation oncologists, nurses, equipment engineers, etc., who can use the system to assist in patient communication and equipment demonstrations.

[0067] Popular science education institutions: such as health education centers, hospital publicity departments, museums and science and technology museums, for daily or special displays of knowledge related to radiotherapy equipment.

[0068] Combining user portraits and survey results, the system needs to meet the following core requirements:

[0069] Equipment identification and interactive presentation: Based on image recognition technology, it automatically identifies radiotherapy equipment (such as proton therapy devices, linear accelerators, CyberKnife, etc.) and links multimedia resources to realize interactive knowledge display.

[0070] Multi-dimensional information presentation: supports displaying relevant knowledge content from multiple dimensions such as structural principles, treatment processes, indication analysis, and equipment comparison to enhance user understanding.

[0071] Immersive communication experience: Through image recognition and AR / VR technology, a more immersive visual communication method is provided to improve user participation and memory rate.

[0072] Doctor-patient communication assistance: used for communication between doctors and patients, improving patients' trust and acceptance of treatment equipment, and alleviating information asymmetry between doctors and patients.

[0073] The system covers two main application scenarios: offline and online:

[0074] Offline scenarios: In places such as hospital radiotherapy centers and health science exhibition halls, interactive display terminals (touch screens, AR glasses, etc.) are deployed to guide users in scanning and identification using physical devices or models.

[0075] Online scenarios: including the science column on the hospital's official website, embedded pages in WeChat public accounts, WeChat mini-programs, and virtual exhibition halls (VR version), which meet the needs of remote access and convenient science popularization, and support users to upload or scan codes to identify images through mobile phones to obtain matching content.

[0076] S1.2: Medical Device Data Collection

[0077] Building the image recognition model and knowledge-driven engine required for the system relies on rich image and text data. Data collection is divided into three levels:

[0078] (1) Equipment material collection

[0079] Image data: Organize on-site shooting to collect high-definition image materials of various radiotherapy equipment at different angles and operating conditions, ensuring that the images contain key structural features, brand logos, etc.

[0080] 3D modeling: Based on real equipment images and structural drawings, we create highly realistic 3D models (such as the Elekta Unity accelerator and IBA proton therapy system) for AR / VR scene display.

[0081] Technical data collection: The system collects equipment manuals, clinical descriptions, application reports, etc., extracts key technical parameters (such as acceleration method, radiation type, dose range, etc.), and builds a structured attribute library.

[0082] Document compilation: Collect and organize various popular science materials on radiotherapy published by authoritative organizations, such as health lecture manuscripts, public service announcement film commentary, etc., to facilitate the formation of multimodal knowledge content matching images.

[0083] (2) User behavior research

[0084] Using methods such as questionnaires, in-depth interviews, and online log analysis, we investigated the understanding of radiotherapy equipment by different user types, their information needs, and content preferences. For example, adolescents preferred animated explanations and anthropomorphic metaphors; medical professionals focused on technical specifications and performance indicators; and general patients were more concerned with the treatment process and safety. The survey results will guide subsequent content labeling, interaction format selection, and user profile development.

[0085] S1.3: Image library and knowledge base construction

[0086] In order to support subsequent recognition model training, image matching, and knowledge display, it is necessary to establish a structured image database and knowledge content library:

[0087] (1) Image database construction

[0088] Image resource integration: Centrally manage the collected radiotherapy equipment images (including physical images, operation interfaces, component close-ups, etc.) and the generated 3D modeling resources;

[0089] Image annotation and classification: Fine-grained labeling of images, including device type, manufacturer name, key parts (such as beam pipes, radiation heads), etc.

[0090] Data update mechanism: Establish an update mechanism to regularly introduce new equipment images (such as robotic radiotherapy systems, high-energy gamma knife, etc.) to keep the data fresh and forward-looking.

[0091] (2) Construction of a popular science knowledge base

[0092] Knowledge content structuring: Structuring the collected popular science materials and technical documents to establish a multi-dimensional knowledge graph (equipment introduction, principle explanation, indication matching, FAQs, etc.);

[0093] Credible content integration: Introducing authoritative information sources (such as radiotherapy guidelines, medical journals, and hospital clinical cases) to ensure the scientific nature and accuracy of knowledge content;

[0094] Content adaptation label design: Based on the depth of user type label content (such as patient perspective, medical perspective, and youth perspective), differentiated push of knowledge content is achieved.

[0095] After the completion of this module, it will provide a solid data and knowledge foundation for subsequent image recognition training and interactive display, ensuring that the system has comprehensive support capabilities in terms of recognition accuracy, content authority, and user interaction experience.

[0096] Image recognition module 120 is used to input the collected images into a deep learning model. Through a neural network combined with an improved feature extraction and target positioning algorithm, it extracts the key semantic features of the radiotherapy equipment in the image. Based on an improved clustering-non-maximum suppression algorithm, it retains the optimal frame and removes redundancy to achieve accurate recognition and classification of the radiotherapy equipment.

[0097] S2. Image Recognition

[0098] This image recognition module 120 is the core technology of the entire system. It uses a deep learning model to accurately identify and categorize radiotherapy devices in input images, providing key support for subsequent multimedia presentations and knowledge linkage. Considering the challenges of diverse image types, complex shooting conditions, and fine target structures in real-world application scenarios, the system supports multiple image sources and interaction modes to fully address real-world usage needs.

[0099] S2.1: Building an intelligent image recognition task

[0100] Step S2 performs intelligent recognition on three types of input images: real-world device images, flat carrier images, and user-uploaded images. The system uses a neural network combined with improved feature extraction and target localization algorithms to extract key semantic features of radiotherapy equipment in the images and achieve robust recognition from multiple angles and resolutions.

[0101] Equipment entity image recognition: This type of image mainly comes from real-life scenes in hospital exhibition halls, radiotherapy centers, or equipment installation areas. It identifies the entire set of radiotherapy equipment entities in the image, such as proton therapy devices, linear accelerators, CyberKnife, Gamma Knife, etc.

[0102] Planar carrier image recognition: This type of image comes from promotional materials posted in hospitals or popular science venues, such as display boards, brochures, and guide maps. It identifies images of radiotherapy equipment in promotional materials, for example, the CyberKnife device shown in "10 Popular Science Posters."

[0103] User-uploaded image recognition: During a hospital visit, users can use their mobile phones to take pictures of the radiotherapy equipment in their department (such as the equipment displayed outside the waiting area or the equipment in the treatment room). Users can also upload historical photos (such as those taken when visiting the hospital or when accompanied by family members to the hospital).

[0104] S2.2: Setting the core algorithm

[0105] S2.2.1: Image Target Positioning

[0106] Image target localization is a fundamental component of the system's image recognition module. Its primary task is to accurately identify the main area of ​​radiotherapy equipment within diverse backgrounds, eliminating distracting factors such as exhibition hall layout and personnel. Considering the complex shape of radiotherapy equipment and the frequent occlusion or small-scale nature of the target area, the system employs a dynamic anchor frame generation mechanism and a multi-objective optimization strategy that integrates the F1 score and PWIoU loss to significantly improve localization accuracy and model robustness.

[0107] Furthermore, the image recognition module 120 includes:

[0108] The target localization unit 121 adopts a dynamic anchor box generation mechanism and optimizes the localization accuracy by fusing the F1 score and the dynamic difficulty perception loss function;

[0109] To address the lack of adaptability of traditional anchor box presets in object detection, this step uses a multi-objective optimization algorithm to dynamically generate anchor boxes that better fit the dataset, improving positioning accuracy and regression efficiency. This step integrates the F1 score and PWloU loss as optimization objectives to balance detection accuracy and box regression quality.

[0110] Furthermore, the target positioning unit 121 includes:

[0111] The classification target calculation subunit 1211 uses the F1 score as the classification optimization target to balance the precision and recall rate and solve the imbalance problem of positive and negative samples;

[0112] The F1 score comprehensively considers precision and recall. When there is an imbalance between positive and negative samples, it can more realistically reflect the overall performance of the recognition model, taking into account both classification accuracy and coverage.

[0113] The regression target calculation subunit 1212 uses a dynamic difficulty-aware loss function as the regression optimization target, dynamically adjusts the gradient weight of the bounding box regression according to the sample abnormality and the monotonic focusing coefficient, and enhances the learning ability of small objects and fuzzy edge samples;

[0114] PWIoU loss introduces a dynamic difficulty perception mechanism based on traditional IoU, which effectively alleviates the problem of gradient instability in target recognition. " and "monotonic focusing coefficient γ" improve the model's learning ability for difficult samples (such as blurred images and small-scale targets), while suppressing overfitting of simple samples.

[0115] Furthermore, a dynamic difficulty-aware loss function dynamically adjusts the gradient weight of bounding box regression according to the sample abnormality and the monotonic focusing coefficient, including: calculating the abnormality β of the current sample, which is the ratio of the IoU loss value of the current sample to the average IoU loss value; when β>1, it is judged as a difficult sample and the gradient gain is amplified to strengthen the learning of small targets and fuzzy edge samples; when β<1, it is judged as a simple sample and the gradient gain is suppressed to prevent overfitting; the gradient gain of difficult samples is exponentially amplified by the monotonic focusing coefficient γ.

[0116] PWIoU loss The definition is as follows:

[0117]

[0118]

[0119]

[0120]

[0121]

[0122] in, : basic loss item; : Distance factor, dynamically adjusts the loss weight based on the center distance and size of the predicted box and the true box; , suppress the gradient dominance of high-quality anchor boxes and balance positive and negative samples; : The coordinates of the center point of the prediction box; : coordinates of the center point of the real frame; : The width and height of the real box, used to normalize the distance error; γ: Monotonic focusing coefficient, amplifying the gradient contribution of difficult samples; : The IoU loss value of the current sample (reflecting the overlap error between the predicted box and the true box); : Dynamic normalization factor, usually the average value of the IoU loss in the current batch or training process, used to standardize the loss value of the current sample and alleviate the vanishing gradient. α: Exponential adjustment factor, controls the decay rate of the gradient gain (hyperparameter), and controls the rate of amplification / suppression (steepness). β: Measures the "abnormality" of the current sample, non-monotonic adjustment of the gradient gain, and adapts to the dynamic sample distribution. β>1 means that the sample loss is higher than the average level, which is a difficult sample (such as a small target, a blurred target), and the gradient gain needs to be increased. β<1 means that the sample loss is lower than the average level, which is a simple sample (such as a large target, a clear target), and the gradient contribution needs to be reduced. δ: Balance point parameter, used to set the threshold for gradient adjustment (usually set to 1, that is, the average loss is used as the benchmark by default).

[0123] when When , γ=1, the gradient gain remains unchanged, and the corresponding sample is the average difficulty.

[0124] when (difficult samples), γ>1, and the larger β or α, the larger γ, the stronger the gradient amplification effect (exponential), and the model pays more attention to such samples.

[0125] when (simple samples), γ<1, the gradient gain is suppressed to prevent the model from overfitting simple samples.

[0126] Dual-objective fusion mechanism 1213, which weights and superimposes the F1 score and the dynamic difficulty perception loss function to form a multi-objective joint loss function :

[0127]

[0128] Among them, λ is the balance weight (0<λ<1), which is used to adjust the optimization emphasis of classification accuracy and positioning accuracy; represents the classification loss based on the F1 score; represents the regression loss based on improved IoU.

[0129] Gradient backpropagation is performed based on the joint loss function to synchronously update the parameters of the classification network and regression network to achieve collaborative optimization of classification and positioning results;

[0130] Anchor box optimization strategy 1214 uses a nonlinear convergence factor to dynamically adjust the search range, performing global exploration in the early stages of iteration and local refinement and convergence in the later stages;

[0131] Improved convergence factor and dynamic adjustment strategy: Traditional anchor box optimization methods are prone to falling into local optimality due to fixed search strategies and weak local exploration capabilities when facing complex image backgrounds or multi-scale targets, resulting in problems such as anchor boxes failing to cover real targets or repeated positioning. To this end, this paper introduces an improved gray wolf optimization algorithm in the anchor box search process and designs a dynamic nonlinear convergence factor. To enhance its adaptive ability. The convergence factor dynamically adjusts the search range based on a nonlinear function, and the specific form is as follows:

[0132] The traditional grey wolf optimization algorithm is prone to fall into local optimality. Enhanced global search capabilities:

[0133]

[0134] Where ρ is the convergence control factor of the current iteration; t is the current iteration number; T is the maximum number of iterations; μ is the adjustment strength coefficient (the empirical value is 2); e is the base of the natural logarithm, approximately equal to 2.71828, which is used in the exponential function to ensure that the adjustment factor changes in a nonlinear manner.

[0135] The anchor box optimization strategy dynamically adjusts the convergence factor during the training process, and its value transitions nonlinearly from global search (ρ≈2) to local fine-tuning (ρ≈0) with the iteration progress t / T.

[0136] This function is used in the early iteration phase (i.e. Small) brings the convergence control factor ρ close to 2. At this point, the optimization strategy emphasizes global exploration, capturing multiple possible locations through a wide-range search, helping to avoid missing targets or falling into local minima. In the later iterations (i.e., t→T), the convergence control factor ρ gradually approaches 0, and the optimization process shifts from extensive search to refined local optimization, focusing on refining the anchor box boundaries and positions to improve final positioning accuracy.

[0137] This mechanism effectively balances the search breadth and convergence speed in anchor frame training, enabling the system to quickly locate the approximate target area in the early stages of training and achieve fine-tuning of anchor frame parameters in the later stages of training. It is suitable for radiotherapy equipment scenarios where there are occlusions, complex structures, or blurred outlines in the image.

[0138] The anchor box generation mechanism 1215 constrains the predicted offset through the Sigmoid function and dynamically outputs the device bounding box coordinates.

[0139] Dynamic anchor frame generation: The quality of anchor frame generation directly affects the accuracy and recall of the image recognition module. Traditional anchor frame methods rely on fixed size presets, which often lead to problems such as poor matching or redundant candidate frames when facing multi-scale and multi-angle targets. To improve adaptability and efficiency, the system has designed a dynamic anchor frame generation mechanism based on network prediction and parameter offset. It can flexibly adjust the size and position of anchor frames based on different input images and target features. The core formula of this mechanism is as follows:

[0140]

[0141] in, : The center point coordinates of the target frame (the final coordinates after dynamic adjustment), : The width and height of the target frame (the final size after dynamic scaling); : The predicted value of the offset of the center point coordinate output by the neural network, : The predicted value of the width and height scaling ratio output by the neural network, and the four parameters represent the relative adjustment values; : The center point coordinates of the initial anchor frame (preset position); : The width and height of the initial anchor box (preset size); : Sigmoid function, constraining the output to interval to ensure the rationality of the anchor box coordinates and size.

[0142] By linking the anchor frame optimization strategy 1214 with the anchor frame generation mechanism 1215, the anchor frame is continuously fine-tuned and optimized using loss feedback during the training process, achieving adaptive optimization of the entire process from coarse-grained initialization to fine-grained positioning.

[0143] Feature extraction unit 122, which uses a multi-scale feature extraction network with a dense connection structure and residual skip connection mechanism, combined with sub-pixel convolution upsampling to preserve image details;

[0144] S2.2.2: Image feature extraction

[0145] Furthermore, the feature extraction unit 122 includes: a data enhancement unit that performs rotation, scaling, and copy-paste processing on the input image; an encoding-decoding structure that extracts multi-scale semantic features in the downsampling stage and restores spatial resolution through sub-pixel convolution in the upsampling stage; and a residual connection channel that superimposes shallow features with the reconstructed feature map to prevent gradient disappearance.

[0146] After image target localization in step S2.2.1, feature extraction unit 122 further performs feature extraction on the identified target. Image feature extraction aims to extract key visual semantic information from radiotherapy equipment images to support subsequent recognition and matching tasks. To improve the model's adaptability to diverse image inputs, the system first uses classic data augmentation methods such as rotation, scaling, cropping, and copying and pasting to expand the sample size and simulate changes in the equipment under different shooting angles and environmental conditions.

[0147] In the feature extraction network structure, the system uses dense connections and residual skip connections, combined with a multi-level sub-pixel convolution upsampling strategy, to effectively improve the model's ability to restore image details and express deep features. The activation function uses PReLU to enhance the model's ability to model nonlinear structures. The convolution kernel size is unified to 3×3, and the number of channels is configured to (64,16) or (16,16) depending on the network level, balancing extraction depth and computational efficiency. The specific process is described as follows:

[0148]

[0149]

[0150] in, : Feature map of the n+1th layer; : No. The sub-pixel convolution upsampling operation of the layer is used to restore the resolution; : Feature map (residual information) passed in parallel with the current layer; : The reconstructed feature map (or output feature map) of the nth layer, that is, the image features after sub-pixel convolution upsampling in the image feature extraction process , shallow feature map connected with residual The result obtained after addition; C: the original feature map of the input; : Feature map after downsampling of the nth layer; : Downsampling operation of the nth layer (such as convolution, maximum pooling, etc.).

[0151] The above structure implements an "encoding-decoding" processing flow: the encoding stage extracts multi-scale high-level semantic features through downsampling; the decoding stage uses sub-pixel convolution to restore spatial details and avoid information loss; the residual connection effectively prevents gradient vanishing in deep networks while enhancing feature fidelity.

[0152] Ultimately, feature extraction unit 122 accurately extracts key visual features of radiotherapy equipment, such as its outline, functional components, and brand logo, and performs high-dimensional feature matching with pre-annotated standard images in a database. The model supports multi-view recognition (e.g., front and side views of the equipment, and images of the equipment in operation), while also integrating with a 3D model library to compensate for viewing angles, ensuring comprehensive and accurate recognition.

[0153] The result screening unit 123 removes redundant prediction boxes based on the improved clustering non-maximum suppression algorithm through clustering grouping and confidence penalty mechanism, and outputs the final recognition result.

[0154] S2.2.3: Output the final recognition result

[0155] After completing device target localization (step S2.2.1) and feature extraction (step S2.2.2), the model generates multiple predicted boxes for each image. Each box includes the predicted category (e.g., proton therapy device, linear accelerator, CyberKnife, etc.), a confidence score, and coordinate information. Due to the presence of redundant predictions (e.g., multiple overlapping boxes pointing to the same target), this step S2.2.3 introduces an improved clustered non-maximum suppression algorithm. Through clustering, iterative suppression, and a score penalty mechanism, it retains the optimal box and removes redundancy, ensuring accurate and efficient target detection results.

[0156] Furthermore, the result screening unit 123 includes: calculating the overlap relationship between the predicted frames based on the intersection-over-union matrix; an iterative suppression mechanism, which eliminates redundant frames overlapping with high-confidence frames in a round-by-round manner through a mask matrix; and a confidence penalty mechanism, which uses an exponential decay function to reduce the false suppression rate of occluded objects. Specifically:

[0157] First, we construct the IoU (intersection over union) matrix between the prediction boxes and group the candidate boxes with high overlap:

[0158] Where X: the output upper triangular IoU matrix; : Extract the upper triangular matrix, ignoring the diagonal , avoid double counting; Represents the N×N dimensional IoU matrix, the matrix elements Stores the overlapping relationship between all boxes; N: the total number of predicted boxes in a single image; : No. Hedi prediction boxes; : The intersection-over-union ratio of two candidate boxes measures the degree of overlap, IoU = intersection area / union area (range [0,1], the larger the value, the higher the overlap).

[0159] The iterative suppression process is based on the IoU matrix and retains candidate boxes with high confidence and low redundancy through a mask mechanism. The following steps are performed:

[0160]

[0161] Where, m: number of iterations, initial m=1, the maximum number of iterations is ; : The retention mark of the previous iteration ( Indicates reservation, indicates removal); : diagonal mask matrix, composed of Generate, used to retain the current activation frame; : Diagonal matrix generation operation; : Suppression matrix, representing the candidate set that has significant overlap with the retained box; : The maximum intersection-over-union ratio of each column is used to determine whether there is a high degree of overlap; : Take the maximum value operation; Threshold (set to 0.5), if , then suppress the box; is the candidate box retained label vector of the mth iteration.

[0162] The iterative suppression mechanism gradually removes redundant frames through multiple rounds of screening, where the mth round generates a mask matrix based on the retained results of the m-1th round and Update the current round retention flag.

[0163] To avoid retaining interference boxes with low overlap but high confidence, the system introduces an exponential decay mechanism to dynamically penalize the confidence of overlapping boxes:

[0164]

[0165] in, : jth candidate box The original confidence of : Corrected confidence after penalty; : Confidence penalty factor (set to 0.2, empirical value), specifically represents the temperature coefficient that controls the intensity of the penalty and controls the decay rate. The smaller the value, the stricter the penalty. : Candidate boxes in round m and The IoU value between them; Represents the index set of all other candidate boxes i whose IoU with candidate box j is greater than or equal to the threshold ε and is not j itself; : exponential decay function, box j highly overlaps The box i is penalized, the higher the overlap ( The larger the value, the smaller the penalty factor (the smaller the exponential decay function value), resulting in The more it is reduced, the more real targets that are misdetected due to occlusion are retained; Indicates the candidate box All overlapping boxes Perform a multiplicative penalty, and when multiple frames overlap, the penalties are superimposed (for example, the Gamma Knife is overlapped by 2 redundant frames: .

[0166] The specific process of the improved clustering non-maximum suppression algorithm is as follows:

[0167] 1. Input: Prediction box set , the corresponding confidence score , IoU threshold ε (usually 0.5), temperature coefficient τ (usually 0.2).

[0168] 2. Initialize the retained marker vector b = [1, 1, ..., 1] (length N).

[0169] 3. Construct IoU matrix: Calculate the IoU between all predicted boxes and get an N×N matrix M, where , extract its strictly upper triangular part (excluding the diagonal) to obtain the matrix X.

[0170] 4. Iteration suppression:

[0171] Loop m = 1 to N:

[0172] If all elements in b are 0, then exit the loop;

[0173] In the set of boxes currently marked as retained (b is 1), find the confidence The highest candidate box index ;

[0174] For all currently marked reserved j that satisfies b j == 1 and j != k box :calculate (or obtained from the precomputed M ), if iou >= ε, then set b j = 0 (suppress box ), and apply the confidence penalty: Update the box Confidence .

[0175] 5. Output: all b i ==1 candidate box and its (possibly penalized) confidence as the final test result.

[0176] For example, when a linear accelerator is detected multiple times: Initial state: Generate 3 overlapping frames ( ), Round 1: (Highest score) Reserved ,inhibition ;Termination: Output unique box This can solve the problem of repeated detection of radiotherapy equipment. It can also retain the blocked equipment, for example: when the CyberKnife is partially blocked: the traditional NMS directly deletes the low-resolution frame → the target is lost. The present invention adopts this algorithm, through the confidence penalty → after correction Still satisfied .

[0177] This mechanism is particularly suitable for situations where there are occlusions or structurally similar areas in medical device images, and can significantly improve the model's ability to retain real targets.

[0178] The entire process in step S2.2.3 supports parallel matrix calculations. IoU matrix processing uses an upper triangular form to avoid repeated calculations and significantly reduce computational complexity. The measured average inference time is controlled at 7.1ms per image, meeting the performance requirements of real-time interactive scenarios. Ultimately, after cluster suppression and confidence adjustment, the system outputs the optimal radiotherapy device recognition results, including device category, precise location (bounding box coordinates), and confidence score. These recognition results directly drive the subsequent popular science content presentation and multimedia interaction (see step S3 for details).

[0179] The popular science interaction module 130 matches the knowledge base content according to the recognition results, displays the device information in a multimodal form, and adapts the content depth based on the user profile;

[0180] S3. Interaction of popular science content

[0181] After the system recognizes and classifies radiotherapy equipment images, it automatically links relevant knowledge content to deliver a "what you see is what you get" interactive science experience through multimodal presentation and personalized push notifications. This module, centered around multi-dimensional knowledge coverage, diverse communication media, and user-friendly user profiles, aims to comprehensively enhance public awareness of radiotherapy equipment.

[0182] Furthermore, the science popularization interaction module 130 includes:

[0183] Multi-dimensional knowledge display unit 131 provides equipment structure principles, treatment process animations, technology comparison charts and AR scene fusion demonstrations. The AR scene fusion demonstration specifically uses image anchor point alignment technology based on feature point matching (such as ORB, SIFT) or plane detection to stably superimpose a 3D model or dynamic path animation on the identified equipment image or real scene. For proton therapy equipment, the motion trajectory of the beam from the accelerator to the tumor target is visually displayed.

[0184] S3.1: Multi-dimensional knowledge display

[0185] To meet the medical knowledge needs of users at different levels, the system presents radiotherapy equipment-related content in a three-dimensional manner, from basic information to clinical applications, and then to technology comparisons and future trends. Specifically, it includes:

[0186] (1) Basic information display

[0187] After device identification, the system displays its name, function, and structural components. It also provides easy-to-understand explanations of technical principles, such as visually demonstrating how a proton beam precisely focuses on a tumor, helping the public understand its unique characteristics compared to traditional X-rays. It also showcases R&D background and industry information, such as the implementation status of China's first batch of proton radiotherapy systems, to strengthen users' confidence in the technological strength of domestically produced equipment.

[0188] (2) Interpretation of clinical application

[0189] Based on the identified device type, the system automatically matches its applicable disease type (e.g., linear accelerators for lung cancer and nasopharyngeal carcinoma, CyberKnife for small intracranial tumors, etc.). Animated presentations show the complete treatment process, from positioning, planning, treatment, to follow-up, enhancing patient accessibility and engagement. Presenting real clinical cases or doctor-patient interviews enhances the content's credibility and relatability.

[0190] (3) Analysis of technical advantages

[0191] The system uses charts and comparative animations to highlight the performance comparisons of different devices in terms of treatment accuracy, side effects, treatment duration, and the breadth of indications. For example, the IBA Proteus PLUS proton radiotherapy system demonstrates its advantages in dose control for pediatric cancer treatment, emphasizing its controllable, precise, and safe features. The system also supports interactive switching between traditional radiotherapy and novel radiotherapy scenarios to enhance user engagement.

[0192] (4) AR scene fusion display

[0193] Using augmented reality (AR) technology, when a user scans an image with a mobile device, the system automatically overlays a 3D model or dynamic animation onto the original image. For example, after identifying an image of a proton therapy device, an animation demonstrating the proton beam's path from the accelerator, through the beam tube, and onto the tumor is overlaid. The system precisely aligns virtual content with the real-world image based on image anchor points, seamlessly blending the physical and virtual elements for an enhanced immersive experience.

[0194] Rich media output unit 132, supporting video playback, synchronous output of graphics, text and voice, and natural language question and answer interaction;

[0195] S3.2: Rich media content formats

[0196] To adapt to different terminals and usage scenarios, the system uses a combination of multiple media carriers to display content, improving communication effects and user acceptance:

[0197] (1) Video / animation playback

[0198] The system includes built-in short video resources, including "Device Operation Principles," "Treatment Simulations," and "Expert Explanations," which users can click to play. For example, a video explaining how a linear accelerator generates and controls high-energy radiation provides a visual demonstration of the complex physical process.

[0199] (2) Graphics, text and voice fusion output

[0200] The system automatically and synchronously displays graphic and text summaries alongside expert audio presentations, helping users with different reading habits access content. Users can choose between "Patient Perspective Interpretation" and "Doctor Interpretation Mode" to achieve multilingual presentation of the same knowledge point.

[0201] (3) Intelligent interactive Q&A

[0202] Based on an NLP semantic recognition model, the system allows users to ask questions via voice or text (e.g., "Do I need to be hospitalized during CyberKnife treatment?") and returns matching text, image, or video responses. The backend links to authoritative content libraries (such as remote diagnosis and treatment guidelines, patient education literature, and health policy interpretations), enabling intelligent search and precise recommendations.

[0203] The user portrait adaptation unit 133 distinguishes between types of teenagers, patients and medical staff, and dynamically adjusts the content presentation format and depth.

[0204] S3.3 Personalized content adaptation

[0205] To improve user engagement and content matching, the system intelligently adjusts the depth and form of content presentation based on user profiles, achieving the goal of "customer-tailored and precise science popularization":

[0206] (1) Teenagers

[0207] Provide engaging and gamified content interpretation methods, such as comparing radiotherapy equipment to "tumor snipers" or "minimally invasive radiation weapons"; displaying proton therapy devices as "proton rocket launch bases," and using interactive simulation animations to help users understand the process of proton beams aiming and striking tumor targets. Display cartoon-like 3D models and interactive question-and-answer challenges stimulate learning interest and participation.

[0208] (2) Patient users

[0209] Focus on practical issues that patients care about, such as treatment side effects, procedure safety, and precautions before and after treatment. Provide practical information such as real patient cases, postoperative recovery guidelines, and answers to frequently asked questions to enhance doctor-patient trust.

[0210] (3) Medical staff

[0211] Displays the device's technical parameters, user interface, performance indicators, and upgrade paths. Provides the latest clinical research results, guideline interpretations, and scientific research progress, serving as a reference for physicians' education and teaching.

[0212] The deployment integration module 140 supports deployment on online platforms and offline terminals, and triggers recognition and interaction functions through image scanning or uploading.

[0213] S4. Deployment and Integration

[0214] In order to achieve wide application and multi-scenario adaptation of the system, two implementation methods are designed: offline deployment and online platform integration, to achieve application effects that cover all scenarios and reach all people.

[0215] Furthermore, the deployment integration module 140 includes: offline deployment mode, configuring touch terminals and AR devices in the physical exhibition hall, supporting code scanning and identification to link physical exhibits; online integration mode, providing image upload and recognition services through WeChat applets and virtual exhibition halls.

[0216] S4.1: Offline scenario deployment

[0217] Interactive devices such as touchscreen all-in-ones, AR recognition terminals, or electronic guide screens are deployed in hospital radiotherapy centers and health science exhibition halls. These systems integrate with physical display panels, device prototypes, or graphic information panels, supporting code scanning, photo recognition, and voice interaction to display recognition results and scientific information on-site. This integrated display method of image recognition, AR content, and physical devices enhances the realism and interactivity of science education.

[0218] S4.2: Online platform integration

[0219] The system will be embedded in platforms such as the "Popular Science Column" on the hospital's official website, the WeChat official account menu, and the patient service mini-program, providing the public with an accessible access to health education anytime, anywhere. Users can upload images via their mobile phones (e.g., photos of equipment taken at the clinic), and the system will automatically identify and return matching content pages, improving ease of use. A VR virtual exhibition hall version will be developed to accommodate online events such as "National Science and Technology Week" and "Cancer Prevention and Treatment Promotion Week," enabling immersive displays of equipment knowledge and remote science education, effectively expanding the scope and application of science education.

[0220] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the methods.

[0221] It should also be noted that, in the embodiments of the present application, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device comprising the elements.

[0222] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the embodiments of the present application may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown in the embodiments of the present application, but rather will conform to the widest scope consistent with the principles and novel features disclosed in the embodiments of the present application.

Claims

1. An interactive system for popularizing and displaying radiotherapy equipment based on image recognition, characterized in that: include: The data acquisition and construction module is used to collect physical images, 3D models, and technical documentation of radiotherapy equipment, and build an image database containing equipment type labels and a knowledge base containing multi-dimensional popular science content; The image recognition module is used to input the collected images into the deep learning model. Through a neural network combined with an improved feature extraction and target positioning algorithm, it extracts the key semantic features of the radiotherapy equipment in the image. Based on the improved clustering non-maximum suppression algorithm, it retains the optimal frame and removes redundancy to achieve accurate recognition and classification of radiotherapy equipment. The popular science interaction module matches knowledge base content based on recognition results, displays device information in a multimodal manner, and adapts content depth based on user profiles; Deploy integrated modules, support deployment on online platforms and offline terminals, and trigger recognition and interaction functions through image scanning or uploading.

2. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 1 is characterized in that: in, The image recognition module includes: The target positioning unit adopts a dynamic anchor box generation mechanism and optimizes positioning accuracy by fusing the F1 score with a dynamic difficulty perception loss function; The feature extraction unit uses a multi-scale feature extraction network with a dense connection structure and residual skip connection mechanism, combined with sub-pixel convolution upsampling to preserve image details; The result screening unit, based on the improved clustering non-maximum suppression algorithm, removes redundant prediction boxes through clustering grouping and confidence penalty mechanism, and outputs the final recognition result.

3. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 2 is characterized in that: The target positioning unit includes: The classification target calculation subunit uses the F1 score as the classification optimization target to balance precision and recall and solve the imbalance problem of positive and negative samples; The regression target calculation subunit uses a dynamic difficulty-aware loss function as the regression optimization target. It dynamically adjusts the gradient weight of the bounding box regression based on the sample abnormality and the monotonic focusing coefficient to enhance the learning ability of small objects and samples with blurred edges. The dual-objective fusion mechanism combines the F1 score with the dynamic difficulty perception loss function to form a multi-objective joint loss function. Gradient backpropagation is performed based on the joint loss function to synchronously update the parameters of the classification network and regression network, achieving coordinated optimization of classification and positioning results. Anchor frame optimization strategy uses a nonlinear convergence factor to dynamically adjust the search range, performing global exploration in the early stages of iteration and local refinement and convergence in the later stages; The anchor box generation mechanism constrains the predicted offset through the Sigmoid function and dynamically outputs the device bounding box coordinates.

4. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 3 is characterized in that: The dynamic difficulty perception loss function dynamically adjusts the gradient weight of bounding box regression according to the sample abnormality and monotonic focusing coefficient, including Calculate the abnormality β of the current sample, which is the ratio of the IoU loss value of the current sample to the average IoU loss value; When β>1, it is determined to be a difficult sample and the gradient gain is amplified to strengthen the learning of small targets and fuzzy edge samples; When β<1, it is judged as a simple sample and the gradient gain is suppressed to prevent overfitting; The gradient gain of difficult samples is exponentially amplified by the monotonic focusing coefficient γ.

5. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 2 is characterized in that: The feature extraction unit includes: Data augmentation unit, which performs rotation, scaling, and copy-paste processing on the input image; The encoding-decoding structure extracts multi-scale semantic features in the downsampling stage and restores spatial resolution through sub-pixel convolution in the upsampling stage; The residual connection channel superimposes the shallow features with the reconstructed feature map to prevent the gradient from disappearing.

6. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 2 is characterized in that: The result screening unit includes: Calculate the overlap between prediction boxes based on the intersection-over-union matrix; Iterative suppression mechanism, which uses a mask matrix to remove redundant boxes that overlap with high-confidence boxes in rounds; The confidence penalty mechanism uses an exponential decay function to reduce the false suppression rate of occluded targets.

7. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 1 is characterized in that: The science popularization interaction module includes: Multi-dimensional knowledge display unit, providing equipment structure principles, treatment process animation, technology comparison charts and AR scene fusion demonstration; Rich media output unit, supporting video playback, synchronous output of images, text and voice, and natural language question and answer interaction; The user portrait adaptation unit distinguishes between teenagers, patients and medical staff, and dynamically adjusts the content presentation format and depth.

8. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 7 is characterized in that: The AR scene fusion demonstration is specifically as follows: Through image anchor alignment technology based on feature point matching or plane detection, a 3D model or dynamic path animation can be stably superimposed on the recognized device image or real scene; For proton therapy devices, the movement trajectory of the beam from the accelerator to the tumor target is visually displayed.

9. The radiotherapy equipment popular science display interactive system based on image recognition according to claim 1 is characterized in that: The deployment integration module includes: Offline deployment: touch terminals and AR devices are configured in the physical exhibition hall to support QR code scanning and linkage with physical exhibits; Online integration provides image upload and recognition services through WeChat mini-programs and virtual exhibition halls.

10. The radiotherapy equipment popular science display interactive system based on image recognition according to any one of claims 1 to 9, characterized in that: The radiotherapy equipment includes proton therapy equipment, linear accelerator, CyberKnife and Gamma Knife.

Images