An ai-based medical image x-ray pulmonary tuberculosis medical information management system

Abstract

The application discloses an AI-based medical image X-ray tuberculosis medical information management system, which comprises an image data analysis module, a digital model construction module, a particle dynamic simulation module, a feature focusing extraction module and an auxiliary evaluation output module. The system analyzes target medical images to obtain anatomical structures, respiratory phase states and active lesion areas as a source of dissemination. Accordingly, a personalized lung model is constructed and model parameters including simulated airways and airflow are generated. Then, the diffusion and deposition of biological particles in the bidirectional airflow are simulated from the source of dissemination, and probability distribution data of the dissemination risk are generated. With the data as spatial attention weights, the image is subjected to feature extraction, focusing on detecting small abnormalities in high-risk areas, and generating small lesion detection results. Finally, the probability distribution data and the detection results are fused to output a visual evaluation report. The application significantly improves the recognition ability of small lesions by simulating the disease dissemination mechanism to guide detection.

Description

Technical Field

[0001] This invention relates to the field of medical image data processing, and in particular to an AI-based medical image X-ray tuberculosis medical information management system. Background Technology

[0002] Tuberculosis (TB) is a major global public health challenge, and early, accurate diagnosis is crucial for controlling disease transmission and improving patient outcomes. Chest X-rays, due to their accessibility, low cost, and high throughput, have become the preferred imaging modality for TB screening and initial assessment.

[0003] In clinical practice, radiologists manually review chest X-rays to look for imaging features that may indicate pulmonary tuberculosis, such as patchy infiltration, cavity formation, and miliary nodules. However, this process is highly dependent on the doctor's experience and focus, and is prone to problems such as subjectivity, low efficiency, and missed or misdiagnosis due to fatigue, especially in high-load screening scenarios.

[0004] To assist doctors, computer-aided detection (CAD) systems based on artificial intelligence (AI) have been developed and applied to chest X-ray analysis. Existing AI assessment systems typically employ deep convolutional neural networks (CNNs) to learn and extract features from massive amounts of labeled images in an end-to-end manner, enabling the automatic identification or classification of pulmonary tuberculosis lesions. These systems have improved the objectivity and efficiency of detection to some extent.

[0005] Existing AI-assisted assessment systems still face significant and fundamental technical bottlenecks when dealing with early, small, and atypical lesions of pulmonary tuberculosis, specifically as follows:

[0006] Existing technologies typically treat chest X-rays as a whole two-dimensional image for pattern recognition, and their detection capabilities rely on the statistical regularities of overt lesion features in the training data. For atypical manifestations such as the numerous small (e.g., miliary nodules), poorly defined, and low-contrast lesions present in early pulmonary tuberculosis, and the early "tree-in-bud" sign formed along the bronchi, existing models are significantly less sensitive. These minute abnormalities occupy very few pixels in the image, and their feature signals are easily lost during the downsampling process of deep neural networks. They are also easily masked or confused by normal anatomical structures (such as ribs and blood vessels), leading to serious missed detections.

[0007] Therefore, there is an urgent need for a new technical solution that can overcome the limitations of existing AI systems in detecting small pulmonary tuberculosis lesions. Summary of the Invention

[0008] In view of the aforementioned deficiencies of the prior art, the technical problem to be solved by the present invention is to provide an AI-based medical imaging X-ray tuberculosis medical information management system, which aims to improve the detection capability of small tuberculosis lesions.

[0009] To achieve the above objectives, this invention provides an AI-based medical imaging X-ray tuberculosis medical information management system. The system includes: an image data analysis module, a digital model construction module, a particle dynamic simulation module, a feature focusing extraction module, and an auxiliary evaluation output module.

[0010] The image data analysis module is used to analyze the target medical image to obtain the anatomical structure information, respiratory phase status, and dissemination source of the lung field region; wherein, the anatomical structure information includes the division information of lung lobes and lung segments, the respiratory phase status is one of the expiratory phase, inspiratory phase, and transition phase, and the dissemination source is the lesion area with preset active imaging characteristics.

[0011] The digital model building module is used to build a personalized lung model based on the anatomical structure information and the respiratory phase state, and generate model parameters including the simulated bronchial tree topology and bidirectional airflow direction field.

[0012] The particle dynamic simulation module is used in the personalized lung model to simulate the diffusion and deposition process of biological particles under the action of bidirectional airflow, with the dissemination source as the initial release point, based on the model parameters, and to generate dissemination risk probability distribution data.

[0013] The feature focusing extraction module is used to extract features from the target medical image using the dissemination risk probability distribution data as spatial attention weights, focusing on extracting minute abnormal features in high-risk areas of the dissemination risk probability distribution data, and generating minute lesion detection results.

[0014] The auxiliary assessment output module is used to integrate the dissemination risk probability distribution data and the microlesion detection results to generate and output auxiliary assessment visualization information; wherein, the auxiliary assessment visualization information includes at least the spatial distribution information of microlesions guided and associated by the dissemination risk probability distribution data.

[0015] Optionally, the image data analysis module determines the respiratory phase state by analyzing the diaphragm position, rib angle, and lung field translucency characteristics in the target medical image.

[0016] Optionally, the particle dynamic simulation module is specifically used for:

[0017] Based on the bidirectional airflow direction field derived from the respiratory phase state, the diffusion and deposition of the bioparticles into the terminal airways under the influence of the airflow dominated by the inhalation phase, and the diffusion and deposition of the bioparticles into the central airways and contralateral lungs under the influence of the airflow dominated by the exhalation phase are simulated respectively. The simulation results of the two phases are then fused to generate the dissemination risk probability distribution data.

[0018] Optionally, the feature focusing extraction module is configured to perform an iterative optimization process, including:

[0019] Preliminary feature extraction of the entire lung region is performed on the target medical image to obtain an initial candidate lesion set;

[0020] Calculate the spatial correlation between the initial candidate lesion set and the current dissemination risk probability distribution data, and adjust the simulation parameters of the particle dynamic simulation module based on the spatial correlation;

[0021] The optimized propagation risk probability distribution data is regenerated using the adjusted simulation parameters, and a second round of focused feature extraction is performed on high-risk areas to generate the final detection results of the microlesions.

[0022] Optionally, the auxiliary assessment visualization information also includes a visualization layer of the dissemination risk probability distribution data, wherein the visualization layer is displayed in the form of a heat map overlaid on the target medical image or the spatial distribution information of the micro lesions.

[0023] Optionally, the system further includes a multi-hypothesis reasoning module for:

[0024] Generate multiple candidate regions for propagation sources;

[0025] For each of the aforementioned candidate propagation source regions, the particle dynamic simulation module is invoked to generate corresponding candidate propagation risk probability distribution data;

[0026] The spatial matching degree of each candidate dissemination risk probability distribution data and the detection results of the small lesions is calculated, and the candidate dissemination source with the highest matching degree and the corresponding candidate dissemination risk probability distribution data are selected to generate or optimize the auxiliary assessment visualization information.

[0027] Optionally, the particle dynamic simulation module is also used to introduce probability distributions into key physical parameters during the simulation process, and generate propagation risk probability distribution data with confidence intervals through multiple Monte Carlo sampling simulations.

[0028] The system also includes a model-evidence verification module, which is used to compare the high-risk areas predicted by the dissemination risk probability distribution data with the actual distribution of the microlesion detection results in real time or after the fact.

[0029] When the comparison difference exceeds the adaptive threshold, the model-evidence verification module triggers the calibration of at least one parameter in the digital model construction module or the particle dynamic simulation module, and iteratively updates the propagation risk probability distribution data.

[0030] Optionally, the image data analysis module is also used to obtain the individual physiological characteristics of the target patient corresponding to the target medical image;

[0031] The digital model building module adaptively adjusts the personalized lung model and its parameters based on the individual's physiological characteristics, so that the dissemination risk probability distribution data generated by the particle dynamic simulation module reflects the impact of the target patient's individual physiological characteristics.

[0032] The beneficial effects of this invention are as follows: 1. This invention provides a prior attention map with clear physiological significance for subsequent detection by generating dissemination risk probability distribution data through the particle dynamic simulation module. This allows the feature focusing extraction module to concentrate computational resources on scanning high-risk areas most likely to detect lesions, based on the navigation of the dissemination risk probability distribution data. This greatly enhances the detection signal for small and blurred targets in complex backgrounds, effectively reducing the false negative rate. 2. Unlike traditional AI that only outputs classification probabilities, this invention integrates disease mechanisms (aerobicity, airway dissemination) into the analysis process. The final output visualization information not only marks the lesion location but also clearly shows the correlation between lesion distribution and individual intrapulmonary oxygen partial pressure gradient and simulated dissemination path. This provides users with a mechanistic explanation, significantly enhancing the clinical credibility and decision support value of the results. 3. This invention accurately estimates the specific respiratory phase of each image (such as end of deep inspiration, quiet expiration) and drives a dynamic simulation containing bidirectional airflow of inhalation and exhalation accordingly. This upgrades the prediction of bacterial particle dissemination pathways from theoretical deduction to dynamic calculations that closely reflect the patient's actual physiological state at the moment of imaging, greatly improving the spatial accuracy and biological rationality of predicting the location of secondary lesions. 4. This invention can integrate the patient's individual physiological characteristics (such as lung morphology and respiratory phase) to construct a dedicated digital twin model for simulation. This allows risk assessment to move beyond population statistical averages and instead focus on the patient's specific airway structure and respiratory dynamics, making the generated dissemination predictions and detection guidance more individualized and theoretically yielding analysis results that more closely reflect the patient's actual condition. 5. By introducing multi-hypothesis reasoning, uncertainty quantification, and a model-evidence verification feedback mechanism, this invention enables the system to handle complex situations such as unknown dissemination sources and inconsistencies between simulation and detection results, and quantitatively identifies the confidence level of the judgment. This reduces the model's over-reliance on a single initial condition, making the system more robust in the face of atypical cases. Its output uncertainty map can also directly guide doctors to conduct focused reviews, forming an efficient human-machine collaboration.

[0033] In summary, this invention addresses the core pain points of existing AI technologies in detecting small lesions of pulmonary tuberculosis by deeply integrating pathological mechanism simulation, personalized modeling, and artificial intelligence detection, thus providing a more accurate, reliable, and usable intelligent auxiliary assessment solution. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the structure of an AI-based medical imaging X-ray tuberculosis medical information management system provided in a specific embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of the workflow of an AI-based medical imaging X-ray tuberculosis medical information management system provided in a specific embodiment of the present invention. Detailed Implementation

[0036] This invention discloses an AI-based medical imaging X-ray tuberculosis medical information management system. Those skilled in the art can refer to the content of this document and appropriately modify the technical details for implementation. It should be particularly noted that all similar substitutions and modifications are obvious to those skilled in the art and are considered to be included in this invention. The apparatus and application of this invention have been described through preferred embodiments. Those skilled in the art can obviously make modifications or appropriate alterations and combinations to the apparatus and application described herein without departing from the content, spirit, and scope of this invention to implement and apply the technology of this invention.

[0037] The applicant's research revealed that existing AI-based methods for assessing pulmonary tuberculosis lack modeling and utilization of the specific pathophysiological mechanisms of tuberculosis. Pulmonary tuberculosis does not occur randomly; its lesion distribution exhibits a clear biological preference (e.g., areas with higher oxygen partial pressure such as the apical-posterior segment of the upper lobe and the dorsal segment of the lower lobe), and its dissemination follows a specific dynamic pathway (primarily through endobronchial spread). Existing AI systems fail to organically integrate this prior knowledge into their model architecture, attempting only to implicitly learn these complex patterns from pixel data. This results in poor interpretability and weak generalization ability when faced with atypical dissemination patterns not adequately covered by training data. Their output typically only indicates the presence, location, or probability score of lesions, failing to provide physicians with mechanistic insights into "why the lesion is located" and "how the disease might progress," thus limiting their auxiliary value in complex clinical decision-making.

[0038] Therefore, embodiments of the present invention provide an AI-based medical information management system for X-ray tuberculosis imaging, such as... Figure 1 As shown, the system includes: an image data analysis module 101, a digital model construction module 102, a particle dynamic simulation module 103, a feature focusing extraction module 104, and an auxiliary evaluation output module 105.

[0039] The image data analysis module 101 is used to analyze the target medical image and obtain the anatomical structure information, respiratory phase status and dissemination source of the lung field region. The anatomical structure information includes the division information of lung lobes and lung segments, the respiratory phase status is one of the expiratory phase, inspiratory phase and transition phase, and the dissemination source is the lesion area with preset active image characteristics.

[0040] The digital model building module 102 is used to build a personalized lung model based on anatomical structure information and respiratory phase state, and generate model parameters including the simulated bronchial tree topology and bidirectional airflow direction field.

[0041] The particle dynamic simulation module 103 is used in a personalized lung model to simulate the diffusion and deposition process of biological particles under the action of bidirectional airflow, with the dissemination source as the initial release point, and generate dissemination risk probability distribution data based on model parameters.

[0042] The feature focusing extraction module 104 is used to extract features from the target medical image by using the dissemination risk probability distribution data as spatial attention weights, focusing on extracting minute abnormal features in high-risk areas of the dissemination risk probability distribution data, and generating minute lesion detection results.

[0043] The auxiliary assessment output module 105 is used to integrate the dissemination risk probability distribution data and the detection results of small lesions to generate and output auxiliary assessment visualization information; wherein, the auxiliary assessment visualization information includes at least the spatial distribution information of small lesions guided and associated by the dissemination risk probability distribution data.

[0044] It is worth mentioning that high-risk areas are regions in the probability distribution data of dissemination risk that have a risk higher than the first threshold, while microlesions refer to lesions smaller than the second threshold.

[0045] It should be noted that:

[0046] The image data analysis module 101 is the information perception front-end of this system. Its core function is to automatically analyze and identify features from the input digitized chest X-ray images. Specifically, it uses a pre-trained deep learning network and image processing algorithms to perform three parallel sub-tasks: First, it accurately segments the lung field and divides the lung lobes and segments according to standard anatomical atlases, outputting anatomical structure information; second, it analyzes the quantitative characteristics of the diaphragm position, rib inclination angle, and lung field translucency in the image to determine the respiratory phase state (end of inspiration, mid-expiration) corresponding to the moment the image is captured; finally, it identifies and outlines lesion areas in the image that meet preset activity characteristics (such as patchy shadows with blurred edges, thin-walled cavities, etc.) and marks them as potential sources of dissemination. The output of this module constitutes the data foundation for all subsequent analyses.

[0047] The digital model construction module 102 is the core of the system's personalized modeling. It receives anatomical information and respiratory phase status from the previous module, and instead of simply storing the data, it constructs a personalized lung model that matches the patient's current physiological state. This process includes: establishing a three-dimensional spatial mapping based on lung lobe segmentation results; deriving or calling the corresponding bidirectional airflow direction field model based on the respiratory phase status; and generating key model parameters that include the simulated bronchial tree topology (such as the direction and bifurcation of major branches). Essentially, this model is a "digital twin" containing patient-specific anatomical and physiological parameters, providing a computational environment for subsequent physical simulations.

[0048] The particulate dynamics simulation module 103 is the mechanism deduction engine of this system, embodying its core innovation. It runs within a personalized lung model constructed upstream. The simulation uses the dissemination source as a virtual initial release point and, based on the bidirectional airflow direction field defined in the model parameters, calculates (simulates) the trajectory, diffusion range, and final deposition process of the biological particles carrying Mycobacterium tuberculosis (abstracted in the model as a swarm of particles with specific aerodynamic characteristics) under the alternating influence of inhaled and exhaled airflows. The physical output of the simulation is a set of dissemination risk probability distribution data, which spatially quantifies the areas where secondary lesions are most likely to occur, as predicted by the model, and their relative probabilities.

[0049] The feature focusing extraction module 104 is the enhanced detector of this system. It changes the traditional AI approach of uniformly scanning the entire image, innovatively transforming the dissemination risk probability distribution data generated by upstream simulation into spatial attention weights. When extracting features from the original target medical image, the system dynamically allocates computational resources based on these weights, performing higher resolution and higher sensitivity analysis on predicted high-risk areas, thereby focusing on extracting easily overlooked minute abnormal features (such as micronodules and subtle tree-bud signs). This "prediction-guided" mechanism makes the final generated microlesion detection results significantly superior to unguided general detection models in terms of recall rate for small targets of active pulmonary tuberculosis.

[0050] The auxiliary assessment output module 105 is the result synthesis and presentation interface of this system. Its core function is to integrate upstream dissemination risk probability distribution data based on mechanism simulation (representing "theoretical prediction") with the detection results of small lesions based on image analysis (representing "empirical findings"). Through spatial alignment and information overlay algorithms, it generates and outputs a visualization of auxiliary assessment information. This information is not a single diagnostic conclusion, but a rich evidence report. Its key component is the spatial distribution information of small lesions guided and correlated by the dissemination risk probability distribution data. This means that the report will clearly show which detected lesions are located in the predicted high-risk area (enhancing their credibility as dissemination sites) and may highlight areas predicted to be high-risk but for which no clear lesions have been detected yet (indicating key areas for follow-up attention), thereby providing doctors with in-depth, mechanistic explanations for decision support.

[0051] In one specific embodiment, such as Figure 2 As shown, the core workflow of the system in this embodiment of the invention includes: after inputting medical images, the image data analysis module 101 parallelly parses anatomical structure information, respiratory phase status, and dissemination source. Based on the former two, a personalized lung model is constructed and model parameters are generated. Simultaneously, using this model and dissemination source as a starting point, particle diffusion is simulated to generate dissemination risk data. Subsequently, the system uses this risk data as weight to focus on extracting microlesions. Finally, the risk data and extraction results are fused to output visualized assessment information. The entire process achieves a closed loop of "mechanism simulation driving precise detection".

[0052] In this specific embodiment, the image data analysis module 101 determines the respiratory phase state by analyzing the position of the diaphragm, the angle of the ribs, and the translucency characteristics of the lung fields in the target medical image.

[0053] It should be noted that this module quantitatively analyzes key imaging features in the target medical image, such as the position of the diaphragm (such as the degree of sinking), the angle of the ribs (such as the tilt), and the translucency of the lung field (reflecting the air content in the lungs). These features are then input into the pre-trained algorithm model, thereby objectively and automatically determining whether the image was captured in the expiratory phase, inspiratory phase, or transition phase. This provides accurate physiological state input for subsequent personalized airflow modeling.

[0054] In this specific embodiment, the particle dynamic simulation module 103 is specifically used for:

[0055] Based on the bidirectional airflow direction field derived from the respiratory phase state, the diffusion and deposition of biological particles into the terminal airways under the influence of airflow dominated by the inhalation phase, and the diffusion and deposition of biological particles into the central airways and contralateral lungs under the influence of airflow dominated by the exhalation phase, were simulated respectively. The simulation results of the two phases were then fused to generate dissemination risk probability distribution data.

[0056] It should be noted that the module first derives the specific parameters of the corresponding bidirectional airflow direction field based on the respiratory phase. Then, it simulates particle deposition distal to the terminal airways under an inhalation-dominant airflow, and particle deposition proximal to the central airways and potentially contralateral lung regions under an exhalation-dominant airflow. Finally, the system weighted and fused the simulation results from the inhalation and exhalation phases to generate comprehensive dissemination risk probability distribution data that better reflects complete respiratory physiology.

[0057] In this specific embodiment, the feature focusing extraction module 104 is configured to perform an iterative optimization process, including:

[0058] Preliminary feature extraction of the entire lung region from the target medical image yields an initial set of candidate lesions.

[0059] Calculate the spatial correlation between the initial candidate lesion set and the current dissemination risk probability distribution data, and adjust the simulation parameters of the particle dynamic simulation module 103 based on the spatial correlation;

[0060] The optimized dissemination risk probability distribution data is regenerated using the adjusted simulation parameters, and a second round of focused feature extraction is performed on high-risk areas to generate the final small lesion detection results.

[0061] It should be noted that this embodiment describes an iterative optimization mechanism for feature focusing extraction. The process begins with preliminary feature extraction of the entire lung region to obtain an initial candidate lesion set. The system then calculates the spatial correlation between this candidate set and the current dissemination risk probability distribution data, and adjusts the simulation parameters (such as particle release intensity or airflow velocity) in the particle dynamic simulation module 103 based on this correlation. Using the adjusted parameters, the system regenerates more optimized risk data and uses this as guidance for a second round of focused feature extraction, thereby obtaining a more accurate final small lesion detection result, forming an enhanced closed loop of "simulation-detection-calibration".

[0062] In this specific embodiment, the auxiliary assessment visualization information also includes a visualization layer of the dissemination risk probability distribution data, which is displayed in the form of a heat map overlaid with the spatial distribution information of the target medical image or small lesions.

[0063] It should be noted that this embodiment limits the specific presentation format of the auxiliary assessment visualization information. The system converts abstract dissemination risk probability distribution data into an intuitive visualization layer, which is typically presented in the form of a heatmap (e.g., red indicates high risk and blue indicates low risk). This heatmap can be overlaid with the original target medical image or the extracted spatial distribution information of small lesions, allowing doctors to clearly see the spatial correspondence between the predicted risk area and the actual lesion, greatly improving the readability and interpretability of the results.

[0064] In this specific embodiment, the system further includes a multi-hypothesis reasoning module, used for:

[0065] Generate multiple candidate regions for propagation sources;

[0066] For each candidate region of the propagation source, the particle dynamic simulation module 103 is invoked to generate the corresponding candidate propagation risk probability distribution data;

[0067] Spatial matching degree is calculated between the probability distribution data of each candidate dissemination risk and the detection results of small lesions. The candidate dissemination source with the highest matching degree and the corresponding candidate dissemination risk probability distribution data are selected to generate or optimize auxiliary assessment visualization information.

[0068] It should be noted that this embodiment introduces a multi-hypothesis reasoning module to improve system robustness. This module does not rely on a single dissemination source determination, but instead generates multiple possible candidate regions for dissemination sources. For each candidate region, the system calls the particle dynamic simulation module 103 to generate corresponding candidate risk data. Subsequently, by calculating the spatial matching degree between each candidate data and the actually detected microlesions, the system selects the combination with the highest matching degree as the most credible explanation and uses it to optimize the final visualization output, thereby reducing the risk of overall misjudgment due to initial source identification bias.

[0069] In this specific embodiment, the particle dynamic simulation module 103 is also used to introduce probability distributions into key physical parameters during the simulation process, and generate propagation risk probability distribution data with confidence intervals through multiple Monte Carlo sampling simulations.

[0070] The system also includes a model-evidence verification module, which is used to compare the predicted high-risk areas of the dissemination risk probability distribution data with the actual distribution of the detection results of small lesions in real time or after the fact.

[0071] When the difference in comparison exceeds the adaptive threshold, the model-evidence verification module triggers the calibration of at least one parameter in the digital model construction module 102 or the particle dynamic simulation module 103, and iteratively updates the propagation risk probability distribution data.

[0072] It should be noted that this embodiment enhances the system's uncertainty quantification and self-calibration capabilities. During simulation, the particle dynamic simulation module 103 introduces probability distributions for key physical parameters and, through multiple Monte Carlo sampling simulations, outputs risk data with confidence intervals, quantifying the uncertainty of the prediction. Simultaneously, the model-evidence verification module continuously compares the predicted high-risk area with the actual lesion distribution. When the difference exceeds a threshold, it automatically triggers the calibration of upstream model parameters and iteratively updates the risk data, enabling the system to self-correct and continuously adapt to complex situations.

[0073] In this specific embodiment, the image data analysis module 101 is also used to acquire the individual physiological characteristics of the target patient corresponding to the target medical image;

[0074] The digital model building module 102 adaptively adjusts the personalized lung model and its parameters according to individual physiological characteristics, so that the dissemination risk probability distribution data generated by the particle dynamic simulation module 103 reflects the influence of the individual physiological characteristics of the target patient.

[0075] It should be noted that this embodiment expands the dimensions of personalized modeling. The system requires the image data analysis module 101 to further acquire or infer the individual physiological characteristics of the target patient (such as age, lung volume derived from body size, chest morphology, etc.). The digital model construction module 102 then adaptively adjusts the constructed personalized lung model and its parameters (such as airway size, regional lung compliance) based on these characteristics. The aim is to ensure that the final generated dissemination risk probability distribution data not only reflects general pathological mechanisms but also reflects the impact of the unique physiological structure of the specific patient, achieving a deeper level of personalized risk assessment.

[0076] In this embodiment of the invention, the dissemination risk probability distribution data generated by the particle dynamic simulation module 103 provides a prior attention map with clear physiological significance for subsequent detection. This enables the feature focusing extraction module 104 to concentrate its computational resources on scanning high-risk areas where lesions are most likely to be found, based on the navigation of the dissemination risk probability distribution data. This greatly enhances the detection signal for small and blurred targets in complex backgrounds and effectively reduces the false negative rate.

[0077] Unlike traditional AI that only outputs classification probabilities, this invention incorporates disease mechanisms (aerobicness, airway dissemination) into the analysis process. The final visualized information not only marks the location of lesions but also clearly demonstrates the correlation between lesion distribution and individual intrapulmonary oxygen partial pressure gradients and simulated dissemination pathways. This provides users with a mechanistic explanation, significantly enhancing the clinical credibility and decision support value of the results.

[0078] This invention accurately estimates the specific respiratory phase (e.g., end of deep inspiration, quiet expiration) of each image and drives a dynamic simulation of bidirectional airflow including inhalation and exhalation. This upgrades the prediction of bacterial particle dissemination paths from theoretical deduction to dynamic calculations that closely match the patient's actual physiological state at the moment of image capture, greatly improving the spatial accuracy and biological rationality of predicting the location of secondary lesions.

[0079] This invention integrates individual patient physiological characteristics (such as lung morphology and respiratory phase) to construct a dedicated digital twin model for simulation. This allows risk assessment to move beyond population statistical averages and instead be based on the patient's specific airway structure and respiratory dynamics, making the generated dissemination predictions and detection guidance more individualized and theoretically providing analytical results that more closely reflect the patient's actual condition.

[0080] This invention, through the introduction of multi-hypothesis reasoning, uncertainty quantification, and a model-evidence verification feedback mechanism, enables the system to handle complex situations such as unclear dissemination sources and inconsistencies between simulation and detection results, and quantitatively identifies the confidence level of the judgment. This reduces the model's over-reliance on a single initial condition, making the system more robust when facing atypical cases. Its output uncertainty map can also directly guide doctors to conduct focused reviews, forming an efficient human-machine collaboration.

[0081] In summary, by deeply integrating pathological mechanism simulation, personalized modeling, and artificial intelligence detection, this invention addresses the core pain points of insufficient sensitivity and poor interpretability of existing AI technologies in detecting small lesions of pulmonary tuberculosis, and provides a more accurate, reliable, and usable intelligent auxiliary assessment solution.

[0082] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0083] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0084] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An AI-based medical image X-ray tuberculosis medical information management system, characterized in that, The system comprises: an image data analysis module, a digital model construction module, a particle dynamic simulation module, a feature focusing extraction module, and an auxiliary evaluation output module, The image data analysis module is configured to analyze a target medical image to obtain anatomical structure information of a lung field region, a breathing phase state, and a dispersion source; the anatomical structure information comprises lung lobe and lung segment division information, the breathing phase state is one of an expiration phase, an inspiration phase, and a transition phase, and the dispersion source is a lesion region of a preset active image feature; The digital model construction module is configured to construct a personalized lung model according to the anatomical structure information and the breathing phase state, and generate model parameters comprising a simulated bronchial tree topological structure and a bidirectional airflow direction field; The particle dynamic simulation module is configured to simulate, in the personalized lung model, a dispersion and deposition process of biological particles under the action of the bidirectional airflow based on the model parameters, taking the dispersion source as an initial release point, to generate dispersion risk probability distribution data; The feature focusing extraction module is configured to perform feature extraction on the target medical image by taking the dispersion risk probability distribution data as spatial attention weights, to focus on extracting micro-abnormal features in a high-risk region in the dispersion risk probability distribution data, and to generate micro-lesion detection results; The auxiliary evaluation output module is configured to fuse the dispersion risk probability distribution data and the micro-lesion detection results to generate and output auxiliary evaluation visual information; the auxiliary evaluation visual information at least comprises micro-lesion spatial distribution information guided and associated by the dispersion risk probability distribution data. 2.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, characterized in that, The image data analysis module determines the breathing phase state by analyzing diaphragm position, rib angle, and lung field transparency features in the target medical image. 3.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, characterized in that, The particle dynamic simulation module is specifically configured to: According to the bidirectional airflow direction field derived based on the breathing phase state, simulate the dispersion and deposition of the biological particles to the distal airway under the action of the inspiration phase dominant airflow and the dispersion and deposition of the biological particles to the central airway and the contralateral lung under the action of the expiration phase dominant airflow, and fuse the two-phase simulation results to generate the dispersion risk probability distribution data. 4.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, characterized in that, The feature focusing extraction module is configured to perform an iterative optimization process, comprising: performing preliminary feature extraction on the target medical image to obtain an initial candidate lesion set; calculating spatial correlation between the initial candidate lesion set and current dispersion risk probability distribution data, and adjusting simulation parameters of the particle dynamic simulation module based on the spatial correlation; re-generate optimized dispersion risk probability distribution data using the adjusted simulation parameters, and perform a second round of focused feature extraction on the high-risk region to generate final micro-lesion detection results. 5.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, characterized in that, The auxiliary evaluation visual information further comprises a visualization layer of the dispersion risk probability distribution data, which is displayed in a heat map form and superimposed with the target medical image or the micro-lesion spatial distribution information. 6.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, wherein The system further comprises a multi-hypothesis reasoning module configured to: generate a plurality of dispersion source candidate regions; For each of the candidate source area, the particle dynamic simulation module is called to generate corresponding candidate dissemination risk probability distribution data; The candidate dissemination risk probability distribution data are matched with the micro-lesion detection results in space, and the candidate source area and the corresponding candidate dissemination risk probability distribution data with the highest matching degree are selected to generate or optimize the auxiliary evaluation visual information. 7.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, characterized in that, The particle dynamic simulation module is further configured to introduce probability distribution to key physical parameters during simulation, and generate dissemination risk probability distribution data with confidence interval through multiple Monte Carlo sampling simulations; The system further comprises a model-evidence verification module configured to compare the high-risk area predicted by the dissemination risk probability distribution data with the actual distribution of the micro-lesion detection results in real time or after the event; When the difference exceeds an adaptive threshold, the model-evidence verification module triggers calibration of at least one parameter in the digital model construction module or the particle dynamic simulation module, and iteratively updates the dissemination risk probability distribution data. 8.The AI-based medical image X-ray tuberculosis medical information management system according to claim 1, wherein, The image data analysis module is further configured to obtain individual physiological characteristics of a target patient corresponding to the target medical image; The digital model construction module adaptively adjusts the individualized lung model and its model parameters according to the individual physiological characteristics, so that the dissemination risk probability distribution data generated by the particle dynamic simulation module reflects the influence of the individual physiological characteristics of the target patient.

Images