A medical big data-based craniocerebral injury diagnosis and treatment auxiliary decision system

Abstract

The application discloses a kind of brain injury diagnosis and treatment auxiliary decision system based on medical big data, specifically related to computer-aided diagnosis field, for solving the problem of inaccurate structure evolution interpretation in the existing brain injury diagnosis and treatment process;The trajectory construction module obtains the brain injury CT image sequence in the complete diagnosis and treatment cycle, extracts the brain density value to construct the structure evolution trajectory sample library;Through the abnormal identification module, the evolution abnormal feedback signal is identified;Then the conflict triggering module introduces the patient's historical brain function state, to judge whether there is a pattern conflict between structure change and function evaluation;If there is a conflict, the reasoning reconstruction module reconstructs the cross mechanism reasoning chain;The model training module is based on the reconstructed reasoning chain and function state score Training injury mechanism identification model;Finally, the decision output module assists in pathological mechanism judgment according to the identification result, improves the diagnosis and treatment decision accuracy under complex injury condition.

Description

Technical Field

[0001] This invention relates to the field of computer-aided diagnostic technology, and more specifically, to a decision support system for the diagnosis and treatment of craniocerebral injury based on medical big data. Background Technology

[0002] Current methods for assisting in the diagnosis and treatment of traumatic brain injury (TBI) primarily rely on manual interpretation of CT images and empirical assessment of pathological changes. They lack systematic modeling of structural evolution processes, making it difficult to accurately identify structural abnormalities and infer mechanisms in conjunction with functional status during the course of the disease. In clinical practice, TBI presents in complex ways, potentially arising from the interplay of multiple pathological mechanisms. Static comparison of single structural features is insufficient to reveal potential conflicts in complex evolutionary processes, easily leading to biased mechanism identification. Furthermore, the current lack of a mechanism identification system that unifies the modeling of structural evolution paths with brain function scores prevents logical modeling and conflict identification of inconsistencies between structural changes and functional performance, limiting the level of intelligence in decision support. Existing methods only compare static anatomical regions in image sequences, neglecting the temporal evolutionary characteristics of structural density. They also fail to establish trajectory structures for different pathological mechanisms and cannot infer structural migration trends under the cross-action of multiple mechanisms after identifying abnormal evolution. Especially when there is a discrepancy between the direction of structural changes and functional status assessment results, existing solutions lack dynamic reasoning capabilities, making it difficult to reconstruct mechanisms and assist in judgment.

[0003] To address the aforementioned shortcomings, there is an urgent need to construct an intelligent diagnostic and treatment assistance method that is based on structural evolution trajectories and integrates functional scoring trends for anomaly identification, conflict detection, and mechanism reasoning and reconstruction. Summary of the Invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a decision support system for the diagnosis and treatment of craniocerebral injury based on medical big data to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A decision support system for the diagnosis and treatment of traumatic brain injury based on medical big data includes:

[0007] The trajectory construction module is used to acquire CT image sequences of craniocerebral injury within a complete diagnosis and treatment cycle in medical big data, extract brain region density from the images, and construct a sample library of structural evolution trajectory of brain regions based on time series.

[0008] The anomaly recognition module is used to acquire the patient's CT images to construct the structural change path, compare it with the structural evolution trajectory sample library, and identify the evolutionary anomaly feedback signal in the structural trajectory.

[0009] The conflict triggering module extracts historical brain function data of patients based on structural abnormality feedback signals and detects whether there is a pattern conflict between the direction of structural changes and functional assessment.

[0010] The reasoning reconstruction module discards single-mechanism evolution path assumptions and reconstructs reasoning chains based on multi-mechanism intersections when pattern conflicts exist.

[0011] The model training module trains a damage mechanism identification model based on cross-mechanism inference chains and brain functional states.

[0012] The decision output module, based on the output results of the damage mechanism identification model, assists in decision-making regarding the pathological mechanisms of actual patients.

[0013] In a preferred embodiment, the trajectory construction module extracts brain region density from the image and constructs a sample library of structural evolution trajectories of brain regions based on time series data. Specifically, this is achieved by:

[0014] The CT image sequences of traumatic brain injury were sorted in chronological order, and continuous sample groups with image quality meeting the imaging standards were selected.

[0015] For each image, a registration process is performed to extract the density value of the brain region. The density value is the gray value corresponding to the brain tissue region in the CT image.

[0016] Density change sequences are constructed based on density value changes at adjacent time points, and clustered into evolutionary path clusters according to the same mechanism label, forming a structural trajectory sample library with time and mechanism labels;

[0017] For density change sequences that failed to cluster into any mechanism label, sample integration was performed to obtain the historical medical records of patients with corresponding unknown injuries.

[0018] The mechanism label is a classification index identifier for evolutionary path clusters based on medical diagnostic results.

[0019] In a preferred embodiment, the process of identifying the evolutionary anomaly feedback signal in the structural trajectory within the anomaly identification module specifically includes:

[0020] We acquire continuous CT image sequences of patients with unknown injuries during their medical visits, extract brain region density change sequences through image spatial registration, and construct continuous structural change paths in chronological order.

[0021] The structural change path is compared point by point with all trajectory samples in the structural evolution trajectory sample library, and a similarity scoring matrix is ​​constructed using three types of indicators: density change gradient, boundary migration amplitude, and central symmetry offset.

[0022] In the scoring matrix, when the score of any comparison point is lower than the preset similarity threshold, and the direction of the density change gradient is reversed or a symmetric mutation occurs, an evolutionary anomaly feedback signal is output.

[0023] In a preferred embodiment, the conflict triggering module specifically includes detecting whether there is a mode conflict between the direction of structural change and the functional evaluation:

[0024] Historical brain function status assessment scores were extracted from patients with unknown injuries. These assessment scores were comprehensive scores that included standardized neurological function, motor impairment, and cognitive ability.

[0025] After obtaining the evolutionary anomaly feedback signal, the structural change path established in the anomaly identification module is extracted and converted into a density change direction vector expression, which is then compared with the slope direction of the time curve of the historical brain function state assessment score.

[0026] When the direction of density change is inconsistent with the expected direction of brain function change, it is marked as a conflict. If the number of times a conflict occurs within the treatment period exceeds the set number of times, it is considered that there is a pattern conflict.

[0027] In a preferred embodiment, the inference reconstruction module specifically includes reconstructing the inference chain based on multiple mechanisms:

[0028] After identifying pattern conflicts, the structural evolution hypothesis based on a single mechanism evolution trajectory is abandoned. Several trajectory samples with the highest similarity to CT image sequences of patients with unknown injuries are extracted from the structural evolution trajectory sample library. Candidate paths corresponding to different mechanisms are constructed according to the mechanism labels.

[0029] In the candidate path, the trajectory samples are segmented into time stages based on the inflection point of the density change direction in the functional state change area to generate mechanism trajectory stages divided by time.

[0030] Based on the occurrence time of the pattern conflict, segment-level remapping is performed on the trajectory stages in each candidate path and the structural change paths in the patient's CT image, and dynamic linking is performed on the mechanism trajectory stages with the highest mapping similarity.

[0031] The above linking process is executed recursively, and a cross-mechanism inference chain is constructed based on the possible structural combinations of different mechanisms.

[0032] In a preferred embodiment, the mechanism is a pathological mechanism, pathological change process, or medically defined symptom pattern corresponding to a specific type of medical trauma.

[0033] In a preferred embodiment, the model training module, based on the cross-mechanism inference chain and brain functional state, trains the injury mechanism identification model, specifically including:

[0034] The cross-mechanism inference chains corresponding to all patients with unknown injuries are broken down into segments, and a graph-structured inference network is constructed according to mechanism labels and time stages.

[0035] Obtain the brain function status assessment score sequence of all patients with unknown damage during their medical treatment cycle, map the score sequence to the structural change nodes in the inference network according to time stage, and construct a guiding path system with brain function status assessment score as the edge weight factor.

[0036] The training graph structure path recognition model uses a temporal graph-based attention mechanism to calculate the path probability.

[0037] In a preferred embodiment, the decision output module, based on the output results of the damage mechanism identification model, assists in decision-making regarding the actual patient's pathological mechanism, specifically including:

[0038] The continuous CT images of the actual patient during the current medical cycle are input into the image processing flow to extract the density change sequence of the brain region;

[0039] According to the node setting of the graph structure inference network, the density change sequence is mapped to the corresponding structure change node in the graph, and path passage inference is performed based on the edge weight factor on the node connection edge.

[0040] The mechanism labels associated with the inference path with the highest probability of occurrence are statistically output as the results of structural damage mechanism identification.

[0041] The technical effects and advantages of the present invention, a decision support system for the diagnosis and treatment of traumatic brain injury based on medical big data, are as follows:

[0042] By constructing a sample library based on structural density evolution paths and integrating brain function state assessment trends, this approach enables the automatic identification of potential pattern conflicts between structural and functional abnormalities in patients with traumatic brain injury. Upon detecting pattern conflicts, it constructs inference chains based on multi-mechanism cross-pathways, achieving multi-dimensional inference and reconstruction of pathological mechanisms. Compared to existing methods, this approach overcomes the limitations of single-mechanism modeling, exhibiting higher pathological adaptability and structural evolution recognition capabilities. Furthermore, by introducing a graph structure inference network based on mechanism labels and time stages, and combining it with brain function scoring trends for guided path modeling, it significantly improves the ability to model the correspondence between structural changes and functional performance, effectively enhancing the model's ability to identify complex injury mechanisms. The final mechanism identification results not only improve adaptability to actual patient structural change paths but also provide clinical suggestions for mechanism relocation and examination guidance areas, contributing to improved diagnostic accuracy and efficiency, and possessing strong practical value and promising prospects for wider application. Attached Figure Description

[0043] Figure 1This is a schematic diagram of a traumatic brain injury diagnosis and treatment auxiliary decision-making system based on medical big data according to the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] Figure 1 This invention presents a decision support system for the diagnosis and treatment of traumatic brain injury based on medical big data, comprising:

[0047] The trajectory construction module is used to acquire CT image sequences of craniocerebral injury within a complete diagnosis and treatment cycle in medical big data, extract brain region density from the images, and construct a sample library of structural evolution trajectory of brain regions based on time series.

[0048] The anomaly recognition module is used to acquire the patient's CT images to construct the structural change path, compare it with the structural evolution trajectory sample library, and identify the evolutionary anomaly feedback signal in the structural trajectory.

[0049] The conflict triggering module extracts historical brain function data of patients based on structural abnormality feedback signals and detects whether there is a pattern conflict between the direction of structural changes and functional assessment.

[0050] The reasoning reconstruction module discards single-mechanism evolution path assumptions and reconstructs reasoning chains based on multi-mechanism intersections when pattern conflicts exist.

[0051] The model training module trains a damage mechanism identification model based on cross-mechanism inference chains and brain functional states.

[0052] The decision output module, based on the output results of the damage mechanism identification model, assists in decision-making regarding the pathological mechanisms of actual patients.

[0053] The trajectory construction module extracts brain region density from the image and constructs a sample library of structural evolution trajectories of brain regions based on time series.

[0054] A sequence of CT images related to craniocerebral injury for the target patient throughout the entire treatment cycle is obtained through an image retrieval process targeting a medical imaging archive database. This sequence must include all craniocerebral CT examination records from the patient's first visit to their most recent discharge or follow-up examination, typically covering at least two different treatment stages to reflect structural changes. The retrieved images must carry timestamps to ensure accurate capture time records for subsequent time series construction. After image retrieval, an image sample quality screening process is performed. This process screens image samples using standardized image quality control indicators, including but not limited to image resolution (greater than 512×512 pixels), signal-to-noise ratio (greater than 20dB), artifact interference score (not higher than level 1), and brain tissue edge sharpness score (using a gradient magnitude algorithm, greater than a set threshold of 120). If a CT image sample at a certain time point does not meet the above imaging standards, that image is not included in the continuous sample group. By evaluating the above indicators one by one, continuous image sequences with stable image quality and reasonable time intervals are selected within the same patient's treatment cycle, usually ensuring no less than 5 time points and a time interval of no more than 7 days between each group.

[0055] After acquiring consecutive sample sets, image registration is performed on each CT image to ensure spatial consistency for subsequent structural comparisons and evolutionary calculations. Specifically, the CT image with the best imaging quality is first selected as the reference image. Then, rigid registration and affine transformation are performed on all other images to ensure the alignment of brain regions within each image. The registration process uses a standard image resampling method to unify all images to the same spatial scale, and the registration accuracy is evaluated by the structural edge overlap rate to ensure that the structural alignment error is no more than 20 pixels. After registration, the brain parenchyma and ventricular regions are located in the image according to the standard brain atlas coordinate system and segmented using structural region masks. The density value extraction process is performed by region: for each structural region, all pixels within the region are traversed, and the arithmetic mean of its grayscale values ​​is calculated. That is, the density value is the average grayscale value of the pixels covered by the brain tissue region (such as the frontal lobe, temporal lobe, and ventricles) in the CT image, reflecting the tissue density performance of that region at the current time point. For example, if approximately 3500 effective pixels are extracted from a brain parenchyma region, and their average grayscale value is 123.7, this value can be used as the density index of that structural region in the current image. Repeating this process in images at multiple time points can obtain a sequence of changes in the density value of the structural region over time.

[0056] After constructing the density value sequence, quantitative modeling of adjacent density change trends over time is required to form a density change sequence. Specifically, for each structural region, its average density values ​​at consecutive time points are arranged chronologically to form the original density sequence. Then, the density difference between adjacent time points is calculated, and this difference is normalized to eliminate the interference of individual patient differences on the magnitude of density changes. The normalization process uses the maximum density change magnitude of each patient throughout the entire cycle as the standard benchmark, mapping all density differences to the [-1,1] interval, thereby generating a density change vector sequence. This vector sequence represents the change trend of a structural region over time and is used to assess whether a specific pathological mechanism dominates the evolution of brain tissue. Based on the sample trajectories labeled with pathological mechanisms in historical big data, cluster analysis of density change vectors is performed. The clustering method uses a morphological similarity matching algorithm based on density change vectors, combined with dynamic time warping of the time series to calculate the distance metric between samples. If multiple density change sequences are similar in trend morphology, they are clustered into the same evolutionary path cluster and assigned the corresponding mechanism label. For example, if a density sequence trend is most similar to a known path of cerebral contusion and laceration, it is labeled as a "cerebral contusion and laceration" mechanism.

[0057] Density change sequences that could not be categorized into any single mechanism were uniformly integrated and subjected to medical background tracing to ensure the integrity of the overall sample system and to provide a basis for identifying complex or novel pathological mechanisms of multiple injuries. After sample integration, the historical medical records of patients with corresponding unknown injuries were retrieved, including the patient's initial diagnosis, follow-up records, imaging descriptions, and standardized neurological function, motor impairment, and cognitive ability assessments during the current treatment cycle.

[0058] Mechanism labels are index-based identifiers generated after manual medical review and statistical merging, used to define the differences in pathological subtyping between sample clusters. In the management system of the structural trajectory sample library, mechanism labels are not defined based on image features or algorithmic classification standards, but rather on actual clinical medical diagnoses as the primary basis for classification, ensuring that the trajectory clustering results have interpretability and traceability in subsequent clinical auxiliary diagnosis and mechanism reasoning.

[0059] The anomaly identification module identifies the process of evolving anomaly feedback signals in the structural trajectory.

[0060] Continuous CT image sequences were acquired for each patient with unknown injury throughout their entire treatment period. The image acquisition time intervals had to meet the dynamic monitoring requirements within the treatment period, including at least four time points. The images were first uniformly registered to eliminate spatial errors caused by factors such as patient position and scanning angle at different time points. A combination of rigid and flexible registration was used to ensure consistent position and overlapping boundaries of cranial structures in images at different time points. Subsequently, brain structural region segmentation was performed to extract brain parenchyma, ventricles, and related structural regions. The average grayscale value of each region in the image at each time point was calculated as a density expression index, forming a time-series density change data set.

[0061] After the density change sequence is constructed, a continuous structural change path is established in chronological order. This path is defined as a temporal density change trajectory formed by connecting the image time nodes on the horizontal axis and the average density values ​​of various brain structural regions on the vertical axis. As an important indicator for expressing the evolutionary trend of lesions, the structural change path must be complete and continuous. During its construction, any nodes with substandard image quality will be removed to ensure that the sequence truly reflects the structural change process.

[0062] The structural change path is compared point-by-point with all trajectory samples in the structural evolution trajectory sample library. This point-by-point comparison does not refer to pixel-by-pixel comparison at the image pixel level, nor is it an absolute comparison of system timestamps. Instead, it refers to pairing and comparing the structural density values ​​at each time point under a unified treatment cycle alignment benchmark. Due to individual differences in CT image acquisition times among different patients, to ensure temporal consistency in the comparison, the structural evolution paths in all trajectory sample libraries undergo time normalization processing, constructing a normalized time coordinate system based on the treatment cycle proportional time axis. Each structural trajectory is mapped to a closed interval of 0-1 according to its respective time point. The same normalization processing is also performed on structural paths of unknown patients, ensuring that the point-by-point comparison is completed under a unified treatment cycle reference system. Specifically, the standard follow-up period for patients with brain injuries (such as day 0, day 3, day 7, day 14, and day 30 as defined in the clinical pathway for traumatic brain injury) can be used as the baseline time reference point. During sample collection, the starting point of the normalized time axis is the time point when the patient's first cranial CT image shows a parenchymal density that meets the minimum contrast requirement for lesion identification. Subsequent image time points are proportionally mapped according to the position of this starting point within the standard period. For example, if a patient's first CT image obtained on the second day after admission shows an identifiable low-density area of ​​brain parenchyma, this time point is set as the normalization starting point. If the patient completes subsequent CT examinations on days 5, 10, and 16, these time points correspond to normalized coordinates of 0.25, 0.5, and 0.8, respectively, reflecting their relative time progress within the 30-day observation period.

[0063] During the comparison process, for the brain structure density value corresponding to each time point, three key indicators are calculated: density change gradient (i.e., the rate of density change between two adjacent points), boundary migration amplitude (i.e., the spatial distance of the change in the edge position of the structural region), and central symmetry shift (i.e., whether the density distribution of the brain structure is shifted relative to the midline). The changing trends of these three indicators are calculated over time, and a comparison matrix is ​​established with each path in the sample database. Each comparison unit calculates the similarity score of that node on the three indicators, and then merges them into one item in the overall scoring matrix. The scoring range is set from 0 to 1, with a maximum score of 1 representing complete similarity. In the scoring matrix analysis, a similarity threshold of 0.6 is set. If the overall score at any time point is lower than this threshold and is accompanied by any of the following conditions, it is considered an abnormal feedback event trigger condition: (1) The density change gradient direction reverses at this point, that is, the original density decreases and then increases rapidly, or the density suddenly decreases sharply from a stable state. Such changes are common in the later absorption process of cerebral hemorrhage being interrupted, or in the case of rebleeding after surgery; (2) The structural boundary symmetry index changes abruptly, that is, the structural symmetry index decreases by more than 40% between two consecutive points. Such phenomena often reflect the displacement of the brain midline caused by structural displacement or edema expansion. If any of the above abnormalities simultaneously occurs and the scoring threshold is crossed, an evolutionary abnormality feedback signal is immediately output. This signal will be called by subsequent modules as an important basis for judging the conflict between structural abnormal evolution and functional state.

[0064] The conflict triggering module detects whether there is a pattern conflict between the direction of structural change and the functional evaluation.

[0065] The testing timeframe should be set consistent with the entire treatment cycle of patients with unknown injuries, covering the entire assessment period from initial admission to the current state, avoiding any omission of information nodes related to changes in brain function trends. The extracted brain function assessment scores should include three main categories of indicators: standardized neurological function scores, motor disorder scores, and cognitive ability scores. In constructing the standardized neurological function scores, assessment items including EEG analysis, evoked potential assessment, reflex examination, and consciousness level scores are first identified from the patient's previous records and uniformly converted into quantitative scores according to internationally accepted scoring systems (such as the NIHSS or modified Rankin Scale). The extraction of motor disorder scores should focus on gait analysis, muscle tone assessment, limb coordination tests, and other recorded data, selecting data sources that have been recorded multiple times and have continuity. These indicators should be converted into corresponding scores according to established reference standards; for example, muscle strength scores (MRC scoring method) should be converted into a 0–20 point range. The cognitive ability score should be constructed based on dimensions such as the patient's language comprehension, attention, short-term memory, and executive function. Results from standard cognitive assessment scales such as MoCA and MMSE should be extracted. If multiple versions exist, they should be uniformly converted to a common scale, with a standard total score of 30 points, and converted to a unified scoring scale using a linear mapping method. After completing the extraction and quantification of the above three types of sub-scores, a multidimensional comprehensive scoring system is constructed. The comprehensive scoring structure adopts a weighted summation method, with the weight ratio set as follows: neurological function score 50%, motor disorder score 30%, and cognitive ability score 20%. The weighting is based on the degree of dependence of brain function state on neurophysiological integrity, and also refers to the order of brain dysfunction manifestations in previous clinical studies. The comprehensive score is out of 100; the lower the score, the more severe the brain function impairment.

[0066] Upon detecting an evolutionary anomaly feedback signal, the established structural change path in the anomaly identification module is extracted. This path is a continuous curve showing the change of brain tissue density values ​​over time in CT images. In specific implementation, the average gray value of the corresponding region is extracted for each image node in the time series to form a time series density vector. This density vector, primarily based on structural changes, reflects the direction of tissue morphological evolution. Subsequently, the first-order difference of this density time series is calculated to obtain the density change direction vector, where positive values ​​indicate an increase in tissue density and negative values ​​indicate a decrease in density. Correspondingly, the time series data of historical brain function status scores are extracted, and the slope direction of the scores is calculated using the same method to form a functional change direction vector. In this invention, the premise for defining directional conflict is that the trends of the two factors are mutually divergent. That is, when structural density continuously decreases while brain function scores increase or remain stable, or when structural density increases significantly but functional scores show a downward trend, it is determined to be a directional conflict. In real clinical scenarios, if a patient experiences a continuous decrease in density in the prefrontal cortex but their neurological function score increases from 15 to 20, it indicates that structural degeneration has not led to functional loss, and there may be a mismatch in mechanisms. Similarly, if an abnormal increase in density is found in the basal ganglia, while the motor function score decreases from 90 to 60, it indicates that the abnormal increase in density has not brought about functional recovery, but may be due to pathological increases in density such as edema or hemorrhage, which also constitutes a conflict.

[0067] After completing the above directional judgment, all structure-function directional conflict events are statistically analyzed according to time nodes to form a conflict event time series. Pattern conflict is defined as the occurrence frequency of conflict events exceeding a certain set threshold within a treatment cycle. To ensure the clinical reliability of the judgment, the frequency index is set with reference to standard medical clinical cycles, using a 28-day observation window. If the number of conflicts within the cycle is ≥3 (adjustable to 2-5 times to adapt to different treatment scenarios), a pattern conflict flag is triggered. In conflict judgment, each event must meet two conditions: first, the direction of structural density change is significant (change rate exceeds 5%); second, the direction of functional score slope is statistically significant (e.g., score change amplitude exceeds twice the standard error of assessment). All judgments are based on cross-analysis of structural change paths and functional score time series to avoid misjudging conflict status based solely on single score fluctuations or image artifacts. When three consecutive time nodes show directional deviations, or when multiple discontinuous but cumulatively reach a set threshold conflict nodes exist, a pattern conflict is considered established.

[0068] In the inference reconstruction module, an inference chain based on the intersection of multiple mechanisms is reconstructed.

[0069] After the pattern conflict was formally confirmed, the single-mechanism path was no longer used as the dominant structure for explaining the damage mechanism. A reconstruction process based on cross-mechanism inference chains was initiated. The first step was to retrieve several trajectory samples from the structural evolution trajectory sample library that were most similar to the current patient's structural change path. The retrieval method was based on the full-process similarity score of the structural change path. During the comparison process, structural density trend, boundary migration trend, and symmetry preservation were used as joint similarity evaluation indicators. The final score was a weighted sum of the three, with the structural density change trend weighted at 0.5, the boundary migration magnitude weighted at 0.3, and the symmetry shift weighted at 0.2. After ranking the scores, the top five trajectory samples with the highest similarity scores were selected as candidate trajectories. Here, trajectory samples refer to standardized structural change path data in the structural evolution trajectory sample library that has been normalized by time and mechanistic labels. Each trajectory sample is accompanied by a corresponding mechanism label, which is established based on the medical attribution of clinical diagnosis records and clearly reflects the type of potential injury mechanism carried by the trajectory sample, such as "mechanism of edema absorption period after cerebral hemorrhage", "functional compensation mechanism of ischemic brain injury", "early evolution of vascular dementia mechanism", etc. These trajectory samples are classified according to their mechanism labels and independently formed into mechanism candidate pathways.

[0070] After obtaining a set of candidate pathways for different mechanisms, each pathway is structurally divided into time periods. The core basis for this division is the key inflection points in the density change trend within the functional state change area. Taking a specific implementation as an example, firstly, the brain region segment corresponding to the patient's functional state change area is extracted from each candidate pathway, and the density change curve of this region over time is analyzed. In this change curve, the inflection point of density change is identified, that is, the critical point where the direction of the density curve changes from decreasing to increasing or from increasing to decreasing. Each inflection point serves as a boundary for dividing a time period. In clinical settings, common mechanism trajectory stages include: acute stress phase (e.g., a sharp decrease in density), subacute transition phase (a slowdown in density decrease or the appearance of fluctuations), and compensatory adaptation phase (density recovery or stabilization). For example, in a trajectory sample of a hemorrhage absorption mechanism, the density change trend may show a rapid increase in density during the 0-0.3 treatment cycle (hematoma formation), a gradual decrease in density during the 0.3-0.6 cycle (absorption phase), and a stabilization during the 0.6-1.0 cycle (complete absorption phase), which can be used to divide the trajectory into three mechanism trajectory stages. All time-phase segments were completed within a unified treatment cycle reference system and corresponding time axis, ensuring the comparability of phase division results across patients. After phase division, each mechanism path was assigned multiple phase sub-paths, providing basic units for subsequent phase-based structural matching and dynamic linking operations.

[0071] After completing the phase division of the mechanism trajectory, a remapping comparison process is performed between the structural change path and the mechanism trajectory phase to clarify which mechanism and which phase may dominate the explanation of the conflict point. First, the occurrence time of the pattern conflict is projected onto the corresponding time axis of the treatment cycle reference system, and the normalized phase position of that time point is clarified. It is worth noting that "time" here no longer refers to the actual calendar time, but to the time positioning after the CT image time normalization and mechanism trajectory phase division results within the treatment cycle. For example, with the normalized period [0,1] as the axis, if the conflict event occurs at the 0.65 position, it should fall in the "compensatory adaptation period" phase marked in the candidate path. Subsequently, at this normalized time position, the data segment of the corresponding time period in the patient's CT image structural change path is extracted to form the sub-path of actual structural evolution. This structural change sub-path is compared one by one with the corresponding phase paths in all mechanism trajectories, and the structural trend consistency and density inflection point overlap are used for scoring. The mechanism trajectory phase with the highest score is selected, and this mechanism phase is used as the link node in the candidate mechanism. The linking process involves replacing obsolete segments in the original structural evolution path with new mechanism-stage paths, while recording the source and time period of the mechanisms upon which the link depends to ensure the traceability of the inference chain. This process proceeds chronologically, recursively extending forward and backward from the conflict event, continuing to search for matching mechanism trajectories in subsequent stages, and constructing combined paths dominated by multiple mechanisms in alternating periods. The final cross-mechanism inference chain will contain trajectory stages labeled with different mechanisms and cover the CT image evolution path throughout the entire medical cycle, forming a structural inference chain with high interpretability and support for composite mechanisms.

[0072] Furthermore, it is important to clarify that the engineering semantics of the term "mechanism" are not abstract behavioral models or algorithmic control paths, but rather the medical and pathological essence bound to each structural evolutionary trajectory sample. Specifically, each mechanism label represents a disease evolution pattern or symptom process that is recognized by the medical community and has a clear pathological logical chain in the actual diagnosis and treatment process, such as "hematoma absorption mechanism after cerebral contusion and laceration," "ischemia-reperfusion mechanism after cerebral infarction," "self-limiting absorption mechanism of chronic subdural hematoma," or "secondary hydrocephalus mechanism after cerebral hemorrhage." The naming of these mechanisms is based on the clinical trauma type and is constructed in conjunction with the evolutionary trend of anatomical structural changes and the processual evolutionary characteristics of the symptom composition.

[0073] The model training module trains an injury mechanism identification model based on a cross-mechanism inference chain and brain functional state.

[0074] For all constructed cross-mechanism inference chains of patients with unknown injuries, the inference chains are decomposed into standardized segments. Each cross-mechanism inference chain typically contains multiple mechanism trajectory stages, and each mechanism trajectory stage corresponds to a set of continuous structural change nodes. During the decomposition process, the time segmentation criteria used in the inference reconstruction module are used as a basis to ensure consistency between the two modules. The time stage is defined based on the normalized coordinate axis of the treatment cycle, using the time point when the mean structural density in the first CT image reaches a set threshold as the normalization starting point. Subsequent time periods are defined based on the fluctuation cycle of brain function scores and the inflection point changes of structural density in the mechanism trajectory, defining medically representative time intervals such as acute phase, subacute phase, pre-recovery phase, and recovery phase.

[0075] Each mechanism fragment is treated as a node in a graph network. The connection methods of the edges in the graph are set according to the causal connections in the original inference chain, and a mechanism label is attached to each node to identify its pathological mechanism affiliation. To enhance the coherence of the graph structure, the system further merges shared nodes appearing in multiple mechanism trajectory stages of the same patient, retaining their highest-frequency mechanism label and recording the source mechanism trajectory stage information for subsequent backtracking. Finally, a complete cross-patient mechanism graph inference network is constructed from the mechanism fragments of all patients, ensuring that the connections between nodes in the network have directionality and temporal sequence, forming the basic constraints of the edges in the graph.

[0076] The brain function status assessment score sequence for all patients with unknown injuries during their treatment cycle was obtained. The score sequence was synchronously mapped according to the aforementioned time stages and labeled to the corresponding structural change nodes in a graph-structured inference network, forming a coupling relationship between nodes and scores. When constructing the guidance path system, the functional status score of each patient at each stage was mapped to the corresponding structural change node in the graph structure. Each connecting edge was assigned a weight, defined as the slope of the score change between the time stages corresponding to the two structural change nodes before and after the connection. An upward slope in the score represents a trend of functional improvement, while a downward slope represents a trend of deterioration. The edge weight quantifies the guiding strength of each path in the direction of functional evolution.

[0077] Based on the constructed graph-structured inference network and the edge weight system with guided paths, the training process of the graph-structured path recognition model is executed. This model employs a temporal graph-based attention mechanism inference structure; that is, given an input sequence that is a directed graph at different time stages, a multi-head attention mechanism is used to dynamically evaluate the importance and probability of each path in the graph. The training objective is to identify the mechanism combination sequence with the highest probability of path traversal in the graph, under the condition that the functional score evolution trend best matches. The training process uses the standard cross-entropy loss function and combines the difference between node access frequency and score change as optimization factors to improve the model's ability to identify traversal paths in complex combinations of cross-mechanisms. The final trained path recognition model will be used as the base model in subsequent actual patient mechanism recognition inference.

[0078] The decision output module, based on the output results of the damage mechanism identification model, assists in decision-making regarding the pathological mechanisms of actual patients.

[0079] Continuous CT images of actual patients during the current treatment cycle are centrally processed to ensure that all input images meet the prerequisite requirements of spatial registration and image quality standards. The image registration process performs three-dimensional relocalization of brain structures in continuous image frames to remove non-structural changes caused by factors such as patient position and angle deviations. After registration, key brain regions are extracted using a brain region template-based method, including the brain parenchyma, ventricles, and marginal edema zone. The density value of the corresponding region is calculated in each image, expressed as the average gray intensity per voxel in the CT image, reflecting the radioactive decay characteristics of the tissue. Following the timeline of the image sequence, a sequence of brain density values ​​changing over time is constructed, forming a complete density change sequence data structure as the basic expression of the structural change path.

[0080] After establishing the density change sequence, it needs to be mapped to the previously constructed graph-structured inference network. The nodes of the inference network consist of "mechanism label-time stage" pairs, with each node representing the structural evolution of a specific injury mechanism at a certain time stage of the treatment cycle. Therefore, the input of the density change sequence needs to be structured to form a node input feature format recognizable by the graph structure model. Specific operations include: segmenting the density change sequence according to the same time stages as in the inference network, such as the acute phase, subacute phase, pre-recovery phase, and post-recovery phase; within each stage, extracting statistical features of the density change sequence, such as mean change rate, gradient trend, and symmetry offset index, to construct a structural change vector for that time stage. Each structural change vector serves as a feature vector for the input node, corresponding to the node features in the graph structure.

[0081] A graph-based path recognition model is invoked to assess the path traversal probability of the patient's current input node sequence. Since the model incorporates edge weights during training, these weights essentially represent the changing trends of brain function status scores between adjacent nodes in historical samples. Therefore, to avoid bias caused by directly applying historical scores during the actual inference phase, the current patient's brain function status assessment score sequence needs to be compared. By calculating the matching degree between the current patient's score change trend and the historical score trend in the edge weights (e.g., consistency of score slope direction and numerical similarity), the edge weights are dynamically adjusted to form a patient-specific edge weight mapping. The model then performs temporal graph inference calculations on all possible paths based on the adjusted edge weights, dynamically assessing the path traversal probability through an attention mechanism and outputting the traversal score for all paths. The inference path with the highest traversal probability is identified, and all mechanism label information traversed by this path is extracted. Since the mechanism labels of each node in the graph structure are already annotated during construction, the frequency distribution and combination structure of the mechanisms covered by this path can be statistically analyzed. By using the most frequent or dominant mechanism labels in the pathway as the output of structural damage mechanism identification, this approach supports more interpretative mechanistic judgments of the current patient's pathological state and provides mechanistic-level evidence for subsequent treatment planning and functional prediction. The entire process ensures the co-evolutionary logic between input structural information, functional state changes, and mechanism labels, providing highly reliable auxiliary decision-making for multi-mechanism structural identification in complex damage scenarios.

[0082] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0083] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0084] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0085] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0087] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0089] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0090] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0091] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1.A medical big data-based auxiliary decision system for diagnosis and treatment of craniocerebral injury, characterized in that, The method comprises the following steps: a trajectory construction module is used to obtain CT image sequences of brain injury in a complete diagnosis and treatment cycle in medical big data, extract the brain region density of the images, and construct a structural evolution trajectory sample library of the brain region based on time series; an abnormality identification module is used to obtain CT images of a patient to construct a structural change path, compare the structural change path with the structural evolution trajectory sample library, and identify an evolution abnormality feedback signal in the structural trajectory; a conflict triggering module is used to extract historical brain function state data of the patient based on the structural abnormality feedback signal, and detect whether there is a mode conflict between the structural change direction and the function evaluation; a reasoning reconstruction module is used to discard a single mechanism evolution path hypothesis when there is a mode conflict, and reconstruct an inference chain based on a cross-mechanism; a model training module is used to train an injury mechanism identification model based on the cross-mechanism inference chain and the brain function state; a decision output module is used to output a result based on the injury mechanism identification model to assist in decision-making for the actual patient pathological mechanism; the trajectory construction module extracts the brain region density of the images based on time series to construct a structural evolution trajectory sample library of the brain region in the following manner: the CT image sequences of brain injury are sorted in chronological order, and continuous sample groups with image quality meeting imaging standards are screened; the density values of the brain region are extracted by performing registration processing on each image, and the density values are the gray values corresponding to the brain tissue region in the CT images; density change sequences are constructed based on the density value changes at adjacent time points, evolution path clusters are clustered according to the same mechanism labels, and a structural trajectory sample library with time and mechanism labels is formed; sample integration is performed on the density change sequences that fail to be clustered into any mechanism label to obtain historical medical records of unknown injury patients; the mechanism label is a classification index label of the evolution path cluster based on medical diagnosis results; the reasoning reconstruction module reconstructs the inference chain based on the cross-mechanism in the following manner: after the mode conflict is identified, the structural evolution hypothesis based on the single mechanism evolution trajectory is discarded, a plurality of trajectory samples with the highest similarity to the CT image sequences of the unknown injury patient are extracted from the structural evolution trajectory sample library, and candidate paths corresponding to different mechanisms are constructed according to the mechanism labels; in the candidate paths, the trajectory samples are time-stage segmented based on the density change direction inflection points of the function state change area to generate mechanism trajectory stages divided by time; the trajectory stages in each candidate path and the structural change path of the patient CT image are re-mapped at the segment level based on the occurrence time of the mode conflict, and the mechanism trajectory stage with the highest mapping similarity is executed for dynamic linking; the above linking process is recursively executed to construct a cross-mechanism inference chain according to possible structural combinations of different mechanisms; the model training module trains an injury mechanism identification model based on the cross-mechanism inference chain and the brain function state in the following manner: all cross-mechanism inference chains corresponding to unknown injury patients are fragmented, and a graph structure inference network is constructed according to the mechanism labels and the time stages. Obtain the brain function state evaluation score sequence of all unknown injury patients in the treatment period, map the score sequence to the structural change node in the reasoning network according to the time stage, and construct a guided path system with the brain function state evaluation score as the edge weight factor. The graph structure path recognition model is trained, and the model adopts a time series graph-based attention mechanism to infer the structural change path passing probability. 2.The brain injury diagnosis and treatment auxiliary decision system based on medical big data according to claim 1, characterized in that, In the abnormality recognition module, the process of identifying the evolution abnormality feedback signal in the structural trajectory is specifically: Obtain the continuous CT image sequence collected by the unknown injury patient in the treatment period, extract the brain region density change sequence through image space registration, and construct a continuous structural change path according to the time sequence; The structural change path is compared with all trajectory samples in the structural evolution trajectory sample library point by point, and a similarity score matrix is constructed by using three indexes of density change gradient, boundary migration amplitude and center symmetry deviation; In the scoring matrix, when the score of any comparison point is lower than the preset similarity threshold, and the direction of the density change gradient is reversed or the symmetry is mutated, an evolution abnormality feedback signal is output. 3.The brain injury diagnosis and treatment auxiliary decision system based on medical big data according to claim 1, characterized in that, In the conflict triggering module, detecting whether there is a pattern conflict between the structural change direction and the function evaluation specifically includes: Extract the historical brain function state evaluation score of the unknown injury patient, and the evaluation score is a comprehensive score including standardized neurological function, movement disorder and cognitive ability; After obtaining the evolution abnormality feedback signal, the structural change path established in the abnormality recognition module is extracted and converted into a density change direction vector expression, and compared with the time curve slope direction of the historical brain function state evaluation score; When the density change direction is inconsistent with the expected change direction of the brain function state, it is marked as a conflict, and if the number of conflict states in the treatment period is more than the set number of indicators, it is determined that there is a pattern conflict. 4.The brain injury diagnosis and treatment auxiliary decision system based on medical big data according to claim 1, characterized in that, The mechanism is a pathological mechanism, pathological change process or medically defined symptom module corresponding to the specific medical trauma type of the mechanism label. 5.The brain injury diagnosis and treatment auxiliary decision system based on medical big data according to claim 1, characterized in that, The decision output module, based on the output result of the injury mechanism recognition model, assists in decision-making for the actual patient pathological mechanism specifically includes: Input the continuous CT image of the actual patient in the current treatment period into the image processing process, and extract the density change sequence of the brain region; According to the node setting of the graph structure reasoning network, the density change sequence is mapped to the corresponding structural change node in the graph, and path passing reasoning is performed according to the edge weight factor on the node connection edge; The mechanism label associated with the reasoning path with the highest passing probability is statistically output as the structural injury mechanism recognition result.

Images