Clinical prediction diagnosis method based on intelligent analysis of case data of brain disease patient

Abstract

The invention discloses a clinical prediction diagnosis method based on intelligent analysis of case data of a patient with cerebral diseases, and relates to the field of case analysis, and the method comprises the following steps: 1, extracting image case data and physique data of the cerebral diseases from medical records of the patient; 2, preprocessing the image case data and the physique data, and associating the image data with the physique data to generate a data set; step 3, performing characteristic identification of brain volume, head circumference or brain shape parameters based on the physique data in the preprocessed data set, and performing dynamic region segmentation on the corresponding image case data according to the identified physique characteristics to generate a plurality of segmented regions; a dynamic region segmentation method based on physical data is adopted, so that brain segmentation better fits individual physiological features, the accuracy of focus positioning is improved, the dynamic law of disease development is captured, the subjective deviation of diagnosis results is reduced, and objective and direct decision support is provided for doctors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of case analysis, in particular to a clinical prediction diagnosis method based on intelligent analysis of brain disease patient case data. BACKGROUND

[0002] Brain diseases such as stroke, Alzheimer's disease, Parkinson's disease, and brain tumors are the main causes of death and disability worldwide, causing a huge economic and care burden on patient families and society. Many brain diseases have insidious onset and atypical early symptoms; at the same time, there are great differences in disease manifestations, development trajectories, and treatment responses among different patients, making early diagnosis and personalized treatment extremely difficult. With the popularization of medical imaging technology, medical institutions have accumulated a large amount of high-dimensional image data. At the same time, the improvement of electronic health records also provides rich patient physical and clinical data.

[0003] Traditional computer-aided diagnosis systems usually analyze medical images globally, then it ignores the region specificity of lesions, many brain diseases do not uniformly affect the whole brain, but preferentially attack specific brain areas, and patients have individual differences, patients of different ages, genders, and physical conditions have different brain structure baselines, using the same standard to measure everyone will result in low sensitivity or high false positive rate, it is difficult to capture specific change patterns related to disease pathophysiology, often a reliable diagnosis can be made when structural lesions are already very obvious, at this time the best intervention opportunity may have been missed. SUMMARY

[0004] (I) Technical problems to be solved

[0005] In view of the above shortcomings of the prior art, the present application provides a clinical prediction diagnosis method based on intelligent analysis of brain disease patient case data, which can effectively solve the problems of the prior art.

[0006] (II) Technical solutions

[0007] In order to achieve the above purpose, the present application is realized by the following technical solutions:

[0008] The present application discloses a clinical prediction diagnosis method based on intelligent analysis of brain disease patient case data, comprising the following steps:

[0009] Step 1: Extract image case data and physical data of brain diseases from patient medical records; data extraction is automatically completed through a medical database interface or an electronic health record system, data is extracted from medical records through an automated interface, which realizes the standardization and efficiency of data collection, avoids errors and omissions that may be caused by traditional manual entry, ensures the integrity and consistency of the data source, and lays a solid foundation for subsequent establishment of personalized analysis models, and improves the reliability of the diagnosis process from the source;

[0010] Step 2: Preprocess the image case data and physical data, associate the image data with the physical data, and generate a data set; through preprocessing, association and fusion, the information density of the data is greatly enhanced, so that subsequent analysis can consider physiological structure and individual physical characteristics, breaking through the limitations of traditional methods that rely only on image data, and providing data-level protection for precision medicine;

[0011] Step 3: Based on the physical data in the preprocessed data set, identify the characteristics of brain volume, head circumference or brain shape parameters, and according to the identified physical characteristics, dynamically segment the corresponding image case data to generate multiple segmentation regions; based on the physical data of the patient, dynamically adjust the segmentation result to make it more personalized and more suitable for the actual physiological structure of the patient, significantly improving the accuracy of lesion positioning and the relevance of regional feature extraction;

[0012] Step 4: Pre-set several case characteristics including gray matter density, white matter integrity, ventricle size, lesion volume, cortical thickness or hippocampus volume collection indicators; by pre-setting multi-dimensional collection indicators, abstract clinical medical problems are converted into quantifiable technical tasks. Its advantages lie in systematizing and standardizing the target of feature extraction, ensuring that the analysis process covers key pathological features comprehensively, and avoiding the subjectivity and one-sidedness of manual observation;

[0013] Step 5: In each of the multiple segmentation regions, according to the case characteristic attributes associated with the collection indicators, pre-set several regional collection paths; the regional collection paths include spiral paths, zigzag paths, ring paths or grid-based linear paths; by pre-setting multiple structured collection paths, it ensures that systematic and complete data scanning can be performed in each segmentation region, avoiding feature loss that may be caused by random sampling, and providing a guarantee for constructing high-quality trajectory data;

[0014] Step 6: Collecting region data in sequence to form a plurality of trajectory data based on one or more acquisition indicators; converting static and discrete image point data into dynamic and continuous trajectory data, which is more conducive to capturing the continuous change of pathological features in space and provides more abundant and discriminative input information for the risk model;

[0015] Step 7: Inputting each trajectory data under a certain acquisition indicator into the preset risk screening model one by one to output the risk coefficient of the corresponding trajectory data under the case characteristics, and using the preset risk screening model to automatically and quantitatively evaluate the complex trajectory data, which replaces the traditional mode of completely relying on the personal experience of doctors and outputs intuitive risk coefficients to make the diagnosis basis more objective and measurable;

[0016] Step 8: Determining whether the obtained risk coefficient exceeds the safety threshold, and if so, performing an alarm prompt; introducing a safety threshold as an objective judgment criterion and realizing real-time alarm to provide an intelligent early warning system for doctors, which can timely find high-risk cases and significantly improve the clinical response speed to create a valuable time window for early intervention.

[0017] Further, the image case data in step 2 includes magnetic resonance imaging, computed tomography or positron emission tomography images, and the physical data includes age, gender, body mass index, blood pressure and blood lipid level.

[0018] Further, the process of dynamic region segmentation in step 3 includes the following steps:

[0019] Step 31: Identifying brain volume, head circumference or brain shape parameters by analyzing physical data, wherein the brain volume is estimated based on age and gender through a regression model, the head circumference is directly obtained by measuring data, and the brain shape parameters are extracted from image data by principal component analysis to extract feature patterns;

[0020] Step 32: Setting a threshold range based on brain volume, scaling the size of a predefined segmentation template based on head circumference, or adjusting region boundary constraints based on brain shape parameters to adjust the parameters of the image segmentation algorithm;

[0021] Step 33: Using the adjusted parameters, applying the image segmentation algorithm to segment the image case data to generate a plurality of segmentation regions matched with individual physiological characteristics.

[0022] Further, the segmentation regions in step 3 include lobar regions, gray matter regions, white matter regions or lesion-specific regions.

[0023] Further, the preset process of the region collection path in step 5 is:

[0024] According to the spatial distribution characteristics and change patterns of the case characteristics corresponding to the collection indicators, one or more preset path templates are matched for each collection indicator;

[0025] When the collection indicators are associated with structural case characteristic attributes, the matched path templates are systematic coverage templates, including but not limited to spiral paths from the center to the edge of the region, zigzag paths based on a grid;

[0026] When the collection indicators are associated with functional or connectivity case characteristic attributes, the matched path templates are radial paths based on connection points or streamline paths along the direction of nerve fiber bundles.

[0027] Further, the path direction setting process in step 5 is: if the case characteristic attribute has a known diffusion or development trend, the main direction of the collection path is set to be consistent with the trend; the setting process of the path sampling density is: according to the change gradient of the case characteristic in space, the density of the sampling points is adjusted, the sampling density is automatically increased in the region where the characteristic changes dramatically, and the sampling density is reduced in the region where the characteristic is stable.

[0028] Further, the trajectory data construction process in step 6 includes: data point collection, interpolation processing or smoothing processing to generate continuous time series or spatial sequences, and through sequential collection and data processing, the dispersed regional data is integrated into trajectories with context information.

[0029] Further, the construction process of the risk screening model in step 7 is:

[0030] A plurality of training data sets are obtained from a historical case database, each training data set including trajectory data based on collection indicators as input features and corresponding clinical diagnosis results as labels;

[0031] The trajectory data is feature extracted and processed, the risk screening model is fitted using the training data, and the model coefficients are calculated by maximum likelihood estimation or gradient descent algorithm, wherein the model output is a probability value representing the risk probability of the trajectory data corresponding to the case characteristic.

[0032] Further, the expression for calculating the risk coefficient of the risk screening model in step 7 is:

[0033] ;

[0034] In the formula, represents the risk coefficient, indicating the risk probability of the patient having the case characteristic, a bias term representing a model, which is optimized by historical training data, is used to adjust the risk level of the model, a total number of features, a weight coefficient of the i-th feature of the model, a weight coefficient of the i-th feature of the model, which is learned by using maximum likelihood estimation or gradient descent algorithm on historical training data, reflects the influence degree of the corresponding feature on risk, a feature value extracted from trajectory data, a feature value extracted from trajectory data.

[0035] Further, the alarm prompt in step 8 is displayed in the form of a heat map, a curve graph or a three-dimensional model.

[0036] (Three) beneficial effects

[0037] Compared with the known prior art, the technical scheme provided by the present application has the following beneficial effects:

[0038] 1. By referring to the characteristics of patient physical data, dynamic region segmentation is performed, the segmentation scheme is adaptively adjusted according to the specific conditions of each patient, dynamic personalized analysis is realized, trajectory data is obtained by collecting regional data in a corresponding order in a plurality of regions, the habit of visual tracking along a specific anatomical structure or suspected lesion area when a clinician reviews a film is imitated, two-dimensional or three-dimensional static image data is converted into trajectory data with spatial order and clinical significance, the continuity change mode is captured, and the extracted features have important clinical preventive value.

[0039] 2. By analyzing subtle changes in trajectory data and combining a risk screening model, early and occult lesions in brain tissue can be identified before the patient shows obvious clinical symptoms, valuable time is gained for early intervention and treatment, the natural course of the disease is changed, and objective and direct decision support is provided for doctors. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical scheme in the embodiments of the present application or the prior art, the drawings needed in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creating any creative labor.

[0041] Figure 1 is a flowchart of the present application;

[0042] Figure 2 is a flowchart of the dynamic region segmentation process in the present application. DETAILED DESCRIPTION

[0043] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0044] The present application will be further described below in conjunction with the embodiments.

[0045] The clinical prediction diagnosis method for brain disease patient case data based on intelligent analysis in the present embodiment, as shown in FIGS. 1 and 2, comprises the following steps: Figure 1 and Figure 2

[0046] Step 1: Extracting image case data and physical data of brain disease from patient medical records; the data extraction is automatically completed through a medical database interface or an electronic health record system, and through multi-source data collection, it is ensured that the subsequent analysis is based on comprehensive and personalized patient information, and the diagnosis accuracy is improved.

[0047] Step 2: Preprocessing the image case data and physical data, associating the image data with the physical data, and generating a data set; wherein the preprocessing of the image case data includes denoising, contrast enhancement, size standardization or image registration, the preprocessing of the physical data includes missing value filling, outlier detection or standardization processing, through data cleaning and standardization, the noise and variation are eliminated, and the data quality is improved, laying a foundation for subsequent segmentation and feature extraction; the image case data includes magnetic resonance imaging, computed tomography or positron emission tomography images, and the physical data includes age, gender, body mass index, blood pressure and blood lipid level.

[0048] Step 3: Based on the physical data in the preprocessed data set, the characteristics of brain volume, head circumference or brain shape parameters are identified, according to the identified physical characteristics, the corresponding image case data is dynamically segmented to generate a plurality of segmentation regions; through the combination of individual physiological characteristics, the personalized brain segmentation is realized, the region division is matched with the actual brain structure of the patient, and the pertinence of feature extraction is enhanced; the process of dynamic region segmentation is as follows:

[0049] Step 31: Through analyzing the physical data, the brain volume, head circumference or brain shape parameters are identified, wherein the brain volume is estimated based on age and gender through a regression model, the head circumference is directly obtained through measurement data, and the brain shape parameters are extracted from the image data through principal component analysis to obtain feature patterns;

[0050] ​Step 32: Set a threshold range based on brain volume, scale the size of the pre-defined segmentation template based on head circumference, or adjust the regional boundary constraints based on brain shape parameters to adjust the parameters of the image segmentation algorithm;

[0051] Step 33: Using the adjusted parameters, apply the image segmentation algorithm to the image case data for segmentation to generate a plurality of segmentation regions matching individual physiological characteristics; the segmentation regions include cerebral lobe regions, gray matter regions, white matter regions, or lesion-specific regions.

[0052] By comparing the segmentation results with the standard atlas, using similarity indicators such as Dice coefficients for evaluation, and iteratively optimizing the segmentation parameters according to the evaluation results, the segmentation accuracy and consistency are ensured, the personalized brain image segmentation is realized by integrating the physical characteristics into the segmentation process, the segmentation errors caused by individual differences are overcome, and the clinical relevance of regional division and the reliability of subsequent feature extraction are improved.

[0053] Step 4: Pre-set a number of case characteristics including gray matter density, white matter integrity, ventricle size, lesion volume, cortical thickness, or hippocampal volume acquisition indicators; the acquisition indicators are based on clinical guidelines or disease-specific settings, and by defining quantitative feature indicators, the disease characteristics are converted into measurable data, facilitating subsequent trajectory data construction and risk analysis.

[0054] Step 5: In each of the plurality of segmentation regions, pre-set a number of regional acquisition paths according to the case characteristic attributes associated with the acquisition indicators; the regional acquisition paths include spiral paths, zigzag paths, ring paths, or grid-based linear paths, the paths are dynamically generated according to the region shape and size, and through the pre-set structured paths, the systematicness and coverage of data acquisition are ensured, avoiding missing key regions, and improving the representativeness of trajectory data;

[0055] Step 6: According to the corresponding order of the number of regional acquisition paths, continuously acquire regional data to form a number of trajectory data based on one or more acquisition indicators; the construction process of trajectory data includes: data point acquisition, interpolation processing or smoothing processing to generate continuous time series or spatial sequences, through sequential acquisition and data processing, the dispersed regional data is integrated into trajectories with contextual information, capturing feature change patterns for model input.

[0056] Step 7: By pre-setting a risk screening model, input each piece of trajectory data under a certain acquisition indicator into the risk screening model one by one, and output the risk coefficient of the corresponding trajectory data under the case characteristics;

[0057] Step 8: judging whether the obtained risk coefficient exceeds a safety threshold, and if so, performing an alarm prompt; wherein the safety threshold is set based on the probability distribution output by the model and clinical statistical data, the alarm prompt is realized through sound, visual signal or electronic message, the automatic disease risk prediction is realized through the modeling risk assessment, and the timely alarm assists the doctor in early intervention and diagnosis; the alarm prompt integrates the segmentation region, trajectory data, risk coefficient and alarm information in the form of heat map, curve graph or three-dimensional model.

[0058] The embodiment fuses the brain image and physical data of the patient, performs dynamic region segmentation based on individual characteristics, collects data along a preset path to construct trajectory data capable of reflecting the continuous change rule of disease characteristics, and then uses a risk screening model to perform risk assessment on the trajectory data. Compared with the prior art, the core advantage of the method is that it breaks through the limitation of traditional isolated analysis of medical images, realizes dynamic analysis throughout the feature evolution process, and thus can discover the risk of occult lesions that is easily ignored by conventional methods earlier, thereby improving the sensitivity and reliability of clinical prediction and diagnosis.

[0059] On other levels, the embodiment also provides a preset process of a region collection path, specifically:

[0060] According to the spatial distribution characteristics and change mode of the case characteristics corresponding to the collection index, one or more preset path templates are matched for each collection index;

[0061] When the collection index is associated with structural case characteristic attributes, the matched path template is a systematic coverage template, including but not limited to a spiral path from the center of the region to the edge, a zigzag path based on a grid, to realize uniform and comprehensive collection of region morphological characteristics;

[0062] When the collection index is associated with functional or connectivity case characteristic attributes, the matched path template is a radial path based on a connection point or a streamline path along a neural fiber bundle, to track the change trend of the characteristics in the brain network.

[0063] The path direction setting process is: if the case characteristic attributes have a known diffusion or development trend, the main direction of the collection path is set to be consistent with the trend, and for some degenerative diseases, the path direction is preferentially from the known early affected region to the late affected region; the setting process of the path sampling density is: adjusting the density of the sampling points according to the change gradient of the case characteristics in space, automatically increasing the sampling density in the region where the characteristics change dramatically, and reducing the sampling density in the region where the characteristics are stable, to realize the balance between data collection efficiency and accuracy.

[0064] Compared with the prior art, the traditional image analysis method usually adopts a fixed and unified sampling path, such as simple grid scanning or random sampling, which ignores the essential differences in the spatial distribution and change mode of different case characteristics.

[0065] The method designs an optimal acquisition strategy for different characteristic attributes such as structure, function and connectivity through an intelligent path matching mechanism. For structural characteristics, a systematic coverage template is used to ensure complete capture of the morphological characteristics of brain regions. For functional characteristics, a streamline path is used to track neural pathways, effectively reflecting the dynamic changes of the characteristics in the brain network. Through differential processing, the acquired data can truly reflect the spatial characteristics of specific characteristics, improving data quality from the source.

[0066] In terms of path direction setting, the method breaks through the limitations of traditional isotropic sampling. By combining disease development rules, the acquisition direction is consistent with the pathological evolution trend. Through directional acquisition, the detection rate of early lesions is improved, and the disease progression path can be predicted through time series data analysis, providing prospective guidance for clinical intervention.

[0067] Traditional methods often use uniform sampling, which not only wastes resources but also may miss critical information. The method dynamically adjusts the sampling density based on the gradient of characteristic changes, increasing sampling points in areas with rapid characteristic changes and appropriately reducing them in stable areas, achieving an optimal balance between data acquisition efficiency and accuracy. Through intelligent sampling strategy, compared with fixed density sampling, more rich feature information can be obtained under the same data volume, or the data processing volume can be greatly reduced under the same information demand.

[0068] In addition, the method converts professional medical knowledge into executable acquisition schemes through the establishment of a path template library, ensuring the professionalism of the technology and reducing the operation threshold. Through the combination of knowledge-driven and data-driven modes, the drawbacks of traditional methods relying on operator experience are effectively solved, making the analysis results more objective and reproducible.

[0069] The embodiment provides a construction process of a risk screening model, specifically:

[0070] A plurality of sets of training data are obtained from a historical case database, each set of training data including trajectory data based on an acquisition index as input features and corresponding clinical diagnosis results as labels, wherein the labels are binary variables indicating whether the patient has the case characteristics. The selection of training data is based on time span and disease type balance to ensure the representativeness and diversity of the data set.

[0071] The trajectory data is subjected to feature extraction and processing, a risk screening model is fitted using training data, and model coefficients are calculated through maximum likelihood estimation or gradient descent algorithm, wherein the model output is a probability value representing the risk probability of the trajectory data corresponding to the case characteristics; the trajectory data is associated with the disease risk through the statistical learning characteristics of the logistic regression algorithm, and the model construction process focuses on data quality and feature optimization to generate reliable risk coefficients, thereby enhancing the accuracy and interpretability of the prediction diagnosis.

[0072] The expression for calculating the risk coefficient of the risk screening model is:

[0073] ;

[0074] In the formula, represents the risk coefficient, indicating the risk probability of the patient having the case characteristics, and the risk coefficient directly represents the probability of having the disease, which is convenient for clinicians to understand and apply. When R = 0.85, the doctor can directly interpret that the patient has an 85% chance of having the target brain disease. This intuitive probability output facilitates the combination with clinical experience and simplifies the setting process of the safety threshold;

[0075] represents the bias term of the model, which is optimized through historical training data and is used to adjust the baseline risk level of the model, represents the total number of features, represents the weight coefficient of the first feature of the model, represents the weight coefficient of the first feature of the model, which is learned through historical training data using maximum likelihood estimation or gradient descent algorithm, and reflects the influence degree of the corresponding feature on the risk. The weight coefficient directly reflects the influence degree and direction of each feature on the risk. A positive weight indicates that the feature is positively correlated with the risk, and a negative weight indicates a negative correlation. In the prediction of Alzheimer's disease, if the weight of the hippocampal volume feature is negative, the clinician can intuitively understand the positive relationship between hippocampal atrophy and disease risk. Interpretability is crucial for medical diagnosis, as it not only enhances the doctor's confidence in the model results, but also allows for quick tracing of the cause when the model makes an abnormal prediction, providing a reliable basis for diagnosis decisions;

[0076] represents the first feature value extracted from the trajectory data,

[0077] The above formula has low computational complexity in the reasoning stage, enabling real-time risk prediction, which is particularly important for rapid screening in clinical environments. Doctors can obtain risk assessment results within seconds and take timely intervention measures. The low computational complexity allows the model to run stably on standard medical hardware, reducing system deployment costs;

[0078] ​In the case of moderate feature dimension, the formula can maintain stable prediction performance, even in the face of partial data missing or noise interference, the model can still provide relatively reliable evaluation results, which is particularly important for the common data quality problems in the actual medical environment, the multi-dimensional features of the trajectory data are linearly combined and mapped to the probability space through the logistic regression model, an intuitive risk quantization index is provided, and objective risk warning based on data is realized by comparing with the safety threshold, the reliability and efficiency of clinical decision-making are improved.

[0079] In summary, the application simulates the habit of the clinician in the film reading, that is, along the specific anatomical structure or suspected lesion area for visual tracking, converts the two-dimensional or three-dimensional static image data into trajectory data with spatial order and clinical significance, captures the change mode of its continuity, so that the extracted features have important clinical preventive value.

[0080] The dynamic region segmentation method based on physical data breaks through the limitation of the traditional fixed template, makes the brain segmentation more in line with the individual physiological characteristics, and improves the accuracy of lesion positioning, constructs trajectory data through preset acquisition indicators and paths, converts discrete image information into feature sequence with spatio-temporal continuity, effectively captures the dynamic law of disease development, and quantitatively evaluates the trajectory data through the risk screening model, converts the doctor's experience into reusable algorithm model, and greatly reduces the subjective bias of the diagnosis result.

[0081] The application can identify early and occult lesions of brain tissue before the patient shows obvious clinical symptoms by analyzing subtle changes in trajectory data and combining a risk screening model, which wins valuable time for early intervention and treatment, changes the natural course of the disease, provides objective and direct decision support for doctors, realizes early risk warning, triggers an alarm at the lesion sprouting stage by real-time comparison of the quantitative risk coefficient and the safety threshold, and wins valuable time for intervention and treatment, which not only improves the diagnosis efficiency and reduces the workload of doctors, but also provides reliable support for clinical decision-making through data-driven objective analysis.

[0082] The above embodiments are only used to illustrate the technical solutions of the application, but not limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A clinical predictive diagnostic method based on intelligent analysis of patient case data for brain diseases, characterized in that, Includes the following steps: Step 1: Extract image-based case data and physical condition data of brain diseases from the patient's medical records; Step 2: Preprocess the image case data and physical fitness data, associate the image data with the physical fitness data, and generate a dataset; Step 3: Based on the physical characteristics data in the preprocessed dataset, identify the characteristics of brain volume, head circumference or brain shape parameters. According to the identified physical characteristics, perform dynamic region segmentation on the corresponding image case data to generate multiple segmentation regions. Step 4: Pre-set several case characteristics, including gray matter density, white matter integrity, ventricular size, lesion volume, cortical thickness, or hippocampal volume as data collection indicators; Step 5: Within each of the multiple segmented regions, based on the case characteristic attributes associated with the collected indicators, preset several regional collection paths; Step 6: Collect regional data sequentially according to the corresponding order of the collection paths in several areas to form several trajectory data based on one or more collection indicators; Step 7: Using a preset risk screening model, input each trajectory data under a certain collection indicator into the risk screening model one by one, and output the risk coefficient of the corresponding trajectory data under the case characteristics; Step 8: Determine whether the obtained risk coefficient exceeds the safety threshold. If it does, issue an alarm.

2. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The image case data in step 2 includes magnetic resonance imaging, computed tomography, or positron emission tomography images, and the physical data includes age, sex, body mass index, blood pressure, and blood lipid levels.

3. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The dynamic region segmentation process in step 3 includes the following steps: Step 31: Identify brain volume, head circumference, or brain shape parameters by analyzing body composition data; Step 32: Adjust the parameters of the image segmentation algorithm by setting a threshold range based on brain volume, scaling the size of the predefined segmentation template based on head circumference, or adjusting the region boundary constraints based on brain shape parameters; Step 33: Using the adjusted parameters, apply an image segmentation algorithm to segment the image case data to generate multiple segmentation regions that match individual physiological characteristics.

4. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The segmented regions in step 3 include cerebral lobe regions, gray matter regions, white matter regions, or lesion-specific regions.

5. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The process of presetting the area acquisition path in step 5 is as follows: Based on the spatial distribution characteristics and change patterns of the case features corresponding to the collected indicators, one or more preset path templates are matched for each collected indicator. When the collected indicators are associated with the characteristic attributes of structured cases, the matching path template is a systematic coverage template; When the collected indicators are associated with functional or connectivity case characteristics, the matched path template is a radial path based on connection points or a streamlined path along the direction of nerve fiber bundles.

6. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 5, characterized in that, The path direction setting process in step 5 is as follows: if the case feature attributes have a known diffusion or development trend, the main direction of the collection path is set to be consistent with the trend; the path sampling density setting process is as follows: adjust the density of sampling points according to the spatial change gradient of the case features, automatically increase the sampling density in areas where the features change drastically, and decrease the sampling density in areas where the features are stable.

7. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The process of constructing trajectory data in step 6 includes: data point acquisition, interpolation or smoothing to generate continuous temporal or spatial sequences, and integrating scattered regional data into trajectories with contextual information through sequential acquisition and data processing.

8. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The process of constructing the risk screening model in step 7 is as follows: Multiple sets of training data were obtained from the historical case database. Each set of training data included trajectory data based on the collected indicators as input features and the corresponding clinical diagnosis results as labels. The trajectory data is subjected to feature extraction and processing. A risk screening model is fitted using training data. The model coefficients are calculated using maximum likelihood estimation or gradient descent algorithm. The model output is a probability value, which represents the risk probability of the case characteristics corresponding to the trajectory data.

9. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, The expression for calculating the risk coefficient in step 7 using the risk screening model is as follows: ； In the formula, The risk coefficient represents the probability that a patient possesses the characteristics of a specific case. The bias term representing the model. Total number of representative features Representative model number The weight coefficients of each feature, Represents the first extracted from trajectory data Each feature value.

10. The clinical predictive diagnosis method based on intelligent analysis of patient case data for brain diseases according to claim 1, characterized in that, In step 8, the alarm prompts will integrate and display the segmented area, trajectory data, risk coefficient, and alarm information in the form of heat maps, curves, or three-dimensional models.

Images