A brain disease prediction method and system for medical assessment of the elderly

Abstract

This invention relates to the field of medical information processing technology and discloses a method and system for predicting brain diseases in medical assessments of the elderly. By introducing methods such as discontinuous response boundary recognition, stage division, interruption semantic description, and interactive correlation calculation, the method distinguishes between long pauses caused by external interference and pathological delays in responses caused by abnormalities related to brain diseases within the same processing framework. This not only improves the accuracy and stability of brain disease risk prediction results in discontinuous response scenarios but also generates stage risk results and abnormal contribution results from interactive units. This provides a clearer basis for interpreting results in medical assessments of the elderly and addresses the problem that existing technologies cannot effectively distinguish between exogenous interruptions and pathological cognitive delays in discontinuous response scenarios in the elderly, leading to poor accuracy in brain disease prediction.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a method and system for predicting brain diseases in the medical assessment of the elderly. Background Technology

[0002] With the increasing demand for brain disease screening among the elderly, predictive methods based on digital assessment tasks are gradually being applied in geriatric medical assessment scenarios. These methods typically use data such as the results of answering assessment questions, overall response time, average reaction time, or total score of a single assessment to make a preliminary judgment on the risk of brain diseases in the elderly. It is easy to understand that the brain diseases mentioned here are not limited to a specific type of illness; in practical applications, they can include cognitive impairment-related brain diseases, neurodegenerative brain diseases, diseases related to brain function decline, or other abnormal brain states that require auxiliary predictive assistance from digital medical assessment.

[0003] In existing technologies, one type of method mainly uses total score statistics or average duration statistics for assessment. This type of method is relatively simple to implement, but it does not fully utilize the dynamic behavior of the elderly during the response process. Another type of method attempts to directly input the response sequence collected during the digital assessment into a continuous sequence analysis model for processing to output corresponding risk results. The above methods can achieve a certain degree of auxiliary judgment when the response process is continuous and the behavior is stable. However, in the actual medical assessment of the elderly, a typical special scenario often occurs, namely, the intermittent response scenario.

[0004] The term "discontinuous response scenario" generally refers to situations in older adults during digital medical assessments where they experience prolonged pauses, resumptions after pauses, reverting to previous answers after resumptions, repeatedly switching between multiple assessment questions, and repeatedly modifying the answer to the same question. It's important to note that these behaviors in older adults do not necessarily indicate a brain disease. In one common scenario, older adults may experience prolonged pauses due to unfamiliarity with the terminal device, temporary external interference, short breaks, or difficulty observing the question. In another scenario, older adults may exhibit similar data performance due to decreased attention span, pathological cognitive delay, or abnormal task progression ability. Because both scenarios can superficially manifest as increased time consumption, more pauses, and repeated answer modifications, traditional methods often struggle to effectively distinguish between them.

[0005] Furthermore, existing technologies typically assume that response sequences are stable and continuous in time, or that information is propagated only based on adjacent time sequences. As a result, when an elderly person's response process is interrupted, pauses before returning to the original task, or pauses before transitioning to a new response path, traditional methods often directly splice together data that do not belong to the same cognitive progression chain. This leads to confusion between exogenous interruptions and pathological delays, ultimately resulting in low accuracy and a high misjudgment rate in predicting brain diseases, making it difficult to meet the requirements for predictive stability and interpretability in elderly medical assessment scenarios.

[0006] Therefore, how to effectively distinguish between exogenous interruptions and pathological cognitive delays in the digital medical assessment of the elderly, based on only a small amount of key interactive data, in the special scenario of intermittent responses, and on this basis improve the accuracy and stability of brain disease prediction results, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] This invention provides a method and system for predicting brain diseases in medical assessments of the elderly, at least to address the problem that existing technologies cannot effectively distinguish between exogenous interruptions and pathological cognitive delays in scenarios where the elderly respond intermittently, thus leading to poor accuracy in predicting brain diseases.

[0008] To achieve the above objectives, the present invention provides a method for predicting brain diseases in medical assessment of the elderly, the method comprising the following steps: Obtain the original assessment interaction data during the digital medical assessment process for the elderly, and establish an assessment interaction main chain based on the original assessment interaction data; Based on the main chain of the evaluation interaction, the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause are extracted to establish a stage reconstruction benchmark for describing intermittent answering behavior. Based on the stage reconstruction benchmark, the discontinuous response boundary is identified, the evaluation interaction main chain is divided into stages, and interruption semantic description data corresponding to each discontinuous response stage is generated. Based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interrupted semantic description data, correlation calculations are performed to obtain the brain disease risk probability, stage risk results, and abnormal contribution results of interaction units. The system performs a result determination based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, and outputs a brain disease prediction result for medical assessment of the elderly.

[0009] Optionally, the original assessment interaction data during the digital medical assessment process for the elderly is obtained, and an assessment interaction main chain is established based on the original assessment interaction data, specifically including: Collect the question identifier data, answer start timestamp data, answer end timestamp data, and answer change record data for each assessment question during the digital medical assessment process for the elderly, as the original assessment interaction data; Based on the question identifier data, the answer start timestamp data, the answer end timestamp data, and the answer change record data, an interactive unit corresponding to each evaluation question is constructed; The interaction units are arranged in chronological order according to the timestamp data of the start of the response, so as to form an interaction unit sequence; If the assessment question corresponding to the same question identifier is reopened and the answer is modified, the answering process after reopening is recorded as a new interaction unit, and the new interaction unit has the same question identifier as the previous interaction unit. The sorted sequence of interaction units is recorded as the evaluation interaction main chain, which serves as the data basis for subsequent extraction of discontinuous response behavior features and the execution of brain disease prediction.

[0010] Optionally, based on the main evaluation interaction chain, the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the task continuity after pauses are extracted to establish a stage reconstruction benchmark for describing discontinuous answering behavior, specifically including: Based on the start and end timestamps of the responses for each interaction unit in the main evaluation interaction chain, the duration of the responses for each interaction unit is calculated. Based on the answer end timestamp data and answer start timestamp data of adjacent interactive units in the evaluation interaction main chain, calculate the inter-question pause duration corresponding to the nth interactive unit; Based on the number of answer states within the corresponding interaction unit in the answer change record data, the number of answer changes is extracted; Based on the number of interactive units with the same question identifier that appeared before the current interactive unit in the evaluation interactive main chain, the replay depth is extracted; Based on the identifiers of questions that have not yet formed a stable submission state or have a tendency to be modified again in a short period of time before the pause occurs, an incomplete task set is constructed, and the re-access of the incomplete task set by a preset number of interactive units after the pause is counted to obtain the task continuity after the pause. The duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause are uniformly organized and used as the benchmark for the stage reconstruction.

[0011] Optionally, based on the stage reconstruction benchmark to identify discontinuous response boundaries, the evaluation interaction main chain is divided into stages, specifically including: Based on the aforementioned stage reconstruction benchmark, the distribution of answer switching behavior, cross-question jumping behavior, and reversion behavior is statistically analyzed within the adjacent windows of each interactive unit to obtain the local behavioral instability of each corresponding interactive unit. Based on the local behavior instability and the inter-question pause duration, construct the dynamic boundary threshold corresponding to the nth interaction unit; Based on the inter-question pause duration, the dynamic boundary threshold, and the task continuity after the pause, calculate the corresponding stage boundary marker; Based on the stage boundary markers, each interaction unit in the evaluation interaction main chain is assigned a stage number to obtain the discontinuous response stage label corresponding to each interaction unit.

[0012] Optionally, generate interruption semantic description data corresponding to each stage of the intermittent response, specifically including: Based on the pause duration between questions, the pause duration between questions in each interactive unit is standardized according to the average pause level of the current elderly people in this assessment process, so as to obtain the standardized pause intensity. Based on the standardized pause intensity, the number of answer changes, the review depth, and the task continuity after the pause, calculate the external interruption indication value corresponding to each interaction unit; Based on the standardized pause intensity, the number of answer changes, the review depth, and the task continuity after the pause, calculate the pathological lag indicator value corresponding to each interaction unit; The exogenous interruption indication value and the pathological lag indication value are used as components of the interruption semantic description data.

[0013] Optionally, the interruption semantic description data further includes stage anomaly density, the generation of which specifically includes: After the starting position of each intermittent response stage, a preset number of interaction units are extracted, and the proportion of the preset number of interaction units that revisit the set of incomplete tasks is counted. If the proportion is greater than the preset recovery threshold, the corresponding intermittent response stage is marked as a recovery stage; if the proportion is not greater than the preset recovery threshold, the corresponding intermittent response stage is marked as a transition stage. Aggregate the pathological lag indicator values ​​corresponding to each interaction unit within the same discontinuous response phase to obtain the phase anomaly density corresponding to the discontinuous response phase. Write the stage anomaly density into the interruption semantic description data.

[0014] Optionally, correlation calculations are performed based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interruption semantic description data, specifically including: The response duration, standardized pause intensity, number of answer changes, review depth, task continuity after pause, external interruption indicator value, pathological lag indicator value, and stage abnormality density of each interactive unit are combined into an interactive feature vector. Based on the interaction feature vector and the discontinuous response stage label, a stage continuity coefficient and a cross-stage barrier coefficient are constructed between interaction units; wherein, the stage continuity coefficient is used to characterize the continuity relationship between different interaction units at the stage level and the replay depth level, and the cross-stage barrier coefficient is used to characterize whether different interaction units cross stage boundaries. Based on the stage continuity coefficient, the cross-stage barrier coefficient, the exogenous interruption indicator value, and the pathological lag indicator value, the association weight between the interaction units is calculated; The information of historical interaction units is weighted and aggregated based on the association weights to obtain the association state representation of the current interaction unit. If the external interruption indicator value corresponding to the current interaction unit is greater than the corresponding pathological hysteresis indicator value, and the current interaction unit and the previous interaction unit belong to different intermittent response stages, reduce the inheritance ratio of the preceding associated state to the current interaction unit; otherwise, maintain the continuous transmission of the preceding associated state. A global state representation for predicting brain disease risk is generated based on the associated state representation of all interactive units.

[0015] Optionally, the probability of brain disease risk, stage risk results, and abnormal contribution results of interaction units can be obtained, specifically including: Based on the global state representation and the statistical results of pathological lag indicators, replay depth, number of answer changes and stage abnormality density corresponding to each interactive unit, the probability of brain disease risk is calculated. Calculate the stage risk value for each discontinuous response stage, and generate the stage risk result based on the stage risk value for each discontinuous response stage; The confidence level of this brain disease prediction is calculated based on the degree of dispersion between the stage risk values ​​of each intermittent response stage. Based on the pathological hysteresis indicator value corresponding to each interaction unit and the average intensity associated with other interaction units, the abnormal contribution value corresponding to each interaction unit is calculated to form the abnormal contribution result of the interaction unit.

[0016] Optionally, based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, a result determination is performed, and a brain disease prediction result for medical assessment of the elderly is output, specifically including: The probability of brain disease risk is compared with a preset low-risk threshold and a preset high-risk threshold to obtain the corresponding initial risk level; If the initial risk level is in the medium-to-high risk boundary area, a consistency determination is performed by combining the confidence level and the stage risk results to determine the final risk level. Extract interactive units whose contribution values ​​are higher than a preset contribution threshold from the abnormal contribution results of the interactive units, and generate corresponding abnormal stage description information; The final risk level, the probability of brain disease risk, the stage risk result, and the abnormal stage description information are structured and organized to output the brain disease prediction result. If the final risk level reaches the preset referral threshold, further specialist evaluation prompts will be added to the brain disease prediction results.

[0017] Furthermore, to achieve the above objectives, the present invention also provides a brain disease prediction system for medical assessment of the elderly, comprising: The acquisition module is used to acquire the original assessment interaction data in the process of digital medical assessment for the elderly, and to establish an assessment interaction main chain based on the original assessment interaction data. A module is established to extract the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause based on the main chain of the evaluation interaction, and to establish a stage reconstruction benchmark to describe the discontinuous answering behavior. The generation module is used to identify discontinuous response boundaries based on the stage reconstruction benchmark, divide the evaluation interaction main chain into stages, and generate interruption semantic description data corresponding to each discontinuous response stage. The calculation module is used to perform correlation calculations based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interrupted semantic description data to obtain the brain disease risk probability, stage risk results, and abnormal contribution results of interaction units. The output module is used to perform result determination based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, and output brain disease prediction results for medical assessment of the elderly.

[0018] The beneficial effects of this invention are as follows: It proposes a method and system for predicting brain diseases in medical assessments of the elderly. By introducing methods such as discontinuous response boundary recognition, stage division, interruption semantic description, and interaction correlation calculation, it enables the differentiation and processing of long pauses caused by external interference and pathological delays in responses caused by abnormalities related to brain diseases within the same processing framework. This not only improves the accuracy and stability of brain disease risk prediction results in the specific scenario of discontinuous responses, but also generates stage risk results and abnormal contribution results from interaction units. This provides a clearer basis for interpreting results in medical assessments of the elderly, solving the problem that existing technologies cannot effectively distinguish between exogenous interruptions and pathological cognitive delays in scenarios involving discontinuous responses in the elderly, thus leading to poor accuracy in predicting brain diseases. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a brain disease prediction method for medical assessment of the elderly, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the brain disease prediction system for medical assessment of the elderly, according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] It should be noted that the brain disease prediction method for medical assessment of the elderly proposed in this invention is mainly applicable to preliminary stratification scenarios for the elderly in digital medical assessment settings, such as assisted screening, assisted judgment, risk warning, or before further specialist assessment. The digital medical assessment scenario can be a cognitive assessment process completed through a tablet terminal, touch terminal, computer terminal, all-in-one terminal, or other electronic assessment terminal capable of recording the interaction process.

[0022] It should be further noted that the data processed by this invention is not limited to a certain fixed question type. The assessment questions can be multiple choice questions, matching questions, true / false questions, image recognition questions, memory recall questions, sequence arrangement questions, or other assessment tasks that can form question-level interactive records.

[0023] This invention provides a method for predicting brain diseases in medical assessments of the elderly, referring to... Figure 1 As shown, it includes the following steps: S1: Obtain the original assessment interaction data during the digital medical assessment process for the elderly, and establish an assessment interaction main chain based on the original assessment interaction data.

[0024] Specifically, the process involves collecting question identifier data, start-of-response timestamp data, end-of-response timestamp data, and answer change records for each assessment question during the digital medical assessment of the elderly, as the raw assessment interaction data. Based on the question identifier data, the start-of-response timestamp data, the end-of-response timestamp data, and the answer change records, interaction units corresponding to each assessment question are constructed. The interaction units are arranged chronologically according to the start-of-response timestamp data to form an interaction unit sequence. If an assessment question corresponding to the same question identifier is reopened and the answer is modified, the reopened response process is recorded as a new interaction unit, and this new interaction unit maintains the same question identifier as the previous interaction unit. The sorted sequence of interaction units is recorded as the assessment interaction main chain, serving as the data basis for subsequent extraction of discontinuous response behavior features and the execution of brain disease prediction.

[0025] In this embodiment of the invention, step S1 is mainly used to establish the basic data framework for the entire brain disease prediction process. Specifically, by acquiring and uniformly organizing the question-level interaction records of the elderly during the digital medical assessment process, an assessment interaction main chain upon which all subsequent analysis actions depend can be formed. It should be noted that this invention does not simply record the final answer result or the final total score, but rather retains the time sequence and answer change trajectory during the answering process. This allows subsequent processing to see details such as when the elderly started answering questions, when they ended answering questions, whether they returned to a particular question, and whether they changed their answers multiple times within the same question. It is these details that enable this invention to handle the special scenario of discontinuous answering, which is difficult for traditional methods to handle effectively.

[0026] Specifically, in this embodiment, the system can collect data on the question identifier, start-of-response timestamp, end-of-response timestamp, and answer change records for each assessment question during the digital medical assessment process. The question identifier data identifies which assessment question corresponds to each interaction; the start-of-response and end-of-response timestamps describe the continuous response period for that assessment question; and the answer change records describe the changes made by the elderly person during that continuous response period, including answer selection, modification, rollback, and final confirmation. By synchronously collecting the above data, an original assessment interaction data set can be formed.

[0027] Furthermore, in one executable implementation, a single, continuous answering activity by an elderly person regarding a particular question can be defined as an interaction unit, and the nth interaction unit can be represented as: ; in, This represents the question identifier corresponding to the nth interaction unit. This represents the start timestamp of the response from the nth interaction unit. This represents the timestamp indicating the end of the response for the nth interactive unit. This represents the answer change record data corresponding to the nth interaction unit. It's easy to understand that this representation is intended to bind the question, time, and answer changes into a unified data unit in subsequent processing, so that all subsequent behavioral features can be derived from the same data base.

[0028] In practical applications, elderly individuals do not necessarily complete all questions in a strictly linear manner during digital medical assessments. For example, they might answer question 3 first, then pause due to thinking or other reasons, continue answering questions 4 and 5, and then return to question 3 to revise their answer. To address this, this invention does not treat the two actions on question 3 as the same answering process, but records them as two separate interaction units, while maintaining the same question identifier for both. This approach allows for accurate identification of actions such as reviewing, revising, and jumping between questions in subsequent processing. In other words, the interaction unit established by this invention is not a static record of questions, but a dynamic record of question-level answering events.

[0029] After forming multiple interaction units, these units can be arranged chronologically according to their start timestamps to form an interaction unit sequence. Further, this sequence is designated as the evaluation interaction main chain. This evaluation interaction main chain is used to uniformly describe all question-level answering activities during the elderly person's medical assessment process. It serves as the foundational data structure for subsequently extracting data such as answer duration, pause duration between questions, number of answer changes, review depth, task continuity after pauses, and semantic descriptions of interruptions.

[0030] In one specific implementation, suppose an elderly person participates in a digital cognitive assessment consisting of more than twenty questions, with the system continuously recording their actions in the background. The elderly person answers the first few questions relatively smoothly, but exhibits noticeable pauses, brief exits, re-entries, and repeated modifications to previous questions at several points in the middle. In this case, the present invention organizes the entire assessment process into a main chain of assessment interactions with a clear temporal sequence. This main chain not only preserves the sequential relationship between the questions but also the time gaps between different questions and the fact that the same question is accessed multiple times. Therefore, subsequent steps are no longer limited to a total time or score but rather address a complete and structured answering process.

[0031] S2: Based on the main chain of the evaluation interaction, extract the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause, and establish a stage reconstruction benchmark to describe the discontinuous answering behavior.

[0032] Specifically, based on the answer start timestamp and answer end timestamp data corresponding to each interaction unit in the evaluation interaction main chain, the answer duration corresponding to each interaction unit is calculated; based on the answer end timestamp and answer start timestamp data of adjacent interaction units in the evaluation interaction main chain, the inter-question pause duration corresponding to the nth interaction unit is calculated; based on the number of answer states within the corresponding interaction unit in the answer change record data, the number of answer changes is extracted; based on the number of interaction units with the same question identifier that appeared before the current interaction unit in the evaluation interaction main chain, the replay depth is extracted; based on the question identifiers that have not yet formed a stable submission state or have a tendency to be modified again in a short period of time before the pause occurs, an incomplete task set is constructed, and the revisit status of the incomplete task set by a preset number of interaction units after the pause is counted to obtain the task continuity after the pause; the answer duration, the inter-question pause duration, the number of answer changes, the replay depth, and the task continuity after the pause are uniformly organized as the benchmark for the stage reconstruction.

[0033] In this embodiment of the invention, step S2 is mainly used to further transform the evaluation interaction main chain established in step S1 into a structured behavioral quantity that can reflect the characteristics of discontinuous response behavior. Specifically, step S1 answers which interactions occurred, while step S2 answers what behavioral significance these interactions have in the discontinuous response scenario. To achieve this objective, the present invention is not satisfied with merely counting the total or average response time, but rather performs a detailed analysis of each interaction unit and its adjacent relationships, thereby constructing the key benchmark data required for subsequent stage reconstruction.

[0034] Specifically, this can be achieved by first using the start timestamp of the response from the nth interaction unit. and the time stamp of the end of the answer Calculate the duration of the response from this interactive unit. Its expression is: ; in, This represents the duration of the response in the nth interaction unit. This quantity reflects the length of time an elderly person spends answering a question in this continuous response activity. It should be noted that the duration of the response itself is not directly equivalent to cognitive abnormality, because slow response in the elderly may be related to individual differences, the difficulty of the question, or short-term distraction. Therefore, in this invention, the duration of the response is only one of the basic behavioral quantities, and it needs to be analyzed in conjunction with other behavioral quantities.

[0035] Furthermore, the pause duration between questions can be calculated based on the temporal relationship between adjacent interactive units. Its expression is: ; in, This represents the pause duration before the start of the nth interaction unit and after the end of the previous interaction unit. It's easy to understand that the pause duration between questions is one of the key foundations that distinguishes this invention from traditional total-score assessment. Traditional methods often only focus on the time spent answering a single question, failing to effectively utilize the gaps between questions. However, in intermittent answering scenarios, truly meaningful anomalies often occur precisely in the pauses between questions, the resumption of answers after pauses, and the transitions after pauses. Therefore, this invention specifically extracts the pause duration between questions to characterize the rhythmic changes in elderly individuals across different answering activities.

[0036] Based on this, the present invention also extracts the number of answer changes based on the answer change record data. In one executable implementation, if the nth interaction unit contains... If there are multiple answer states, the number of answer changes can be represented as: ; in, This represents the number of answer states within the nth interaction unit. This represents the number of answer changes in the nth interaction unit. This data is primarily used to describe the degree of hesitation, wavering, or repeated confirmation among older adults within the same question. It should be noted that a high number of answer changes does not necessarily indicate an increased risk of brain disease, as some questions are inherently complex, or older adults may adopt a more cautious approach to answering questions, leading to more modifications. Therefore, this quantity also needs to be interpreted in conjunction with other behavioral metrics.

[0037] Furthermore, the present invention also extracts the playback depth. This is used to reflect how many times the question corresponding to the current interactive unit has been accessed before. Its expression is: ; in, This indicates the replay depth of the question corresponding to the nth interactive unit. This is an indicator function, taking the value 1 when the condition is true and 0 otherwise. It's easy to understand that a greater depth of revisiting indicates that older adults are more likely to switch back and forth between multiple questions and reopen previous ones. This behavior is significant in scenarios involving intermittent answering, as it reflects whether older adults exhibit behavioral patterns such as pausing to correct previous questions or reconfirming earlier content when answering subsequent questions.

[0038] It should be noted that simply measuring the duration of responses, the duration of pauses between questions, the number of answer changes, and the depth of review is insufficient to address the core technical problem addressed by this invention. This is because both extrinsic interruptions and pathological cognitive delays can manifest as longer pauses, increased review times, or more answer modifications. If only these phenomena are statistically analyzed without determining whether the original task chain was returned after the pause, effective differentiation remains difficult. Therefore, this invention further constructs a set of incomplete tasks. And thereby extract the task continuity after the pause. .

[0039] Specifically, before a significant pause occurs, the system can identify questions that have not yet reached a stable submission state, or questions that, although answers have been provided, still show a tendency to be modified again in the short term, based on the question identifiers, answer change records, and submission status of several preceding interaction units. These question identifiers are then grouped into a set of incomplete tasks. Subsequently, after the pause, a preset number L of interaction units are selected, and it is counted whether they revisit the questions in the set of incomplete tasks, thereby obtaining the task continuity after the pause. Its expression is: ; in, This indicates the task continuity after the pause corresponding to the nth interaction unit, and L represents the number of interaction units used to observe the recovery behavior after the pause. This represents the set of unfinished tasks prior to the pause. The significance of this value lies in providing an explicit data metric to describe whether the older adult truly returned to the unfinished task chain after the pause. A higher value indicates better recovery after the pause; a lower value indicates a significant deviation in the response path after the pause.

[0040] It's easy to understand that the task continuity after a pause is a key improvement of this invention compared to traditional methods. In typical continuous answering scenarios, the chronological order itself is the task progression order, and the system doesn't need to specifically determine whether the pause leads back to the original task. However, in scenarios where elderly people answer intermittently, this determination is crucial. By introducing a set of unfinished tasks and the task continuity after a pause, this invention can identify whether the pause merely interrupts the current task or completely disrupts the original cognitive progression chain.

[0041] After completing the above calculations, the present invention will provide the duration. Length of pause between questions Number of times the answer was changed Looking back at the depth and task continuity after pause A unified organization serves as the benchmark for phase reconstruction. This benchmark is not simply a data stacking process, but rather the foundational support for subsequent identification of phase boundaries, generation of interruption semantic description data, and execution risk calculation. In one specific implementation, if an elderly person pauses significantly for a long time before a question, and then quickly returns to that question to make changes, the pause duration between questions may be relatively long, but the task continuity after the pause is also relatively high. Conversely, if the elderly person does not return to the original task after a long pause, but jumps back and forth between multiple questions with multiple answer modifications, not only will the pause duration be longer, but the depth of review and the number of answer changes will also increase, and the task continuity may be lower. Through the above methods, this invention provides clear, continuous, and traceable data basis for subsequent phase division and semantic differentiation.

[0042] S3: Based on the stage reconstruction benchmark, identify the discontinuous response boundary, divide the evaluation interaction main chain into stages, and generate interruption semantic description data corresponding to each discontinuous response stage.

[0043] Specifically, based on the stage reconstruction benchmark, the distribution of answer switching behavior, cross-question jump behavior, and reversal behavior is statistically analyzed within the adjacent windows of each interaction unit to obtain the local behavior instability of each interaction unit; based on the local behavior instability and the inter-question pause duration, a dynamic boundary threshold corresponding to the nth interaction unit is constructed; based on the inter-question pause duration, the dynamic boundary threshold, and the task continuity after the pause, the corresponding stage boundary marker is calculated; based on the stage boundary marker, each interaction unit in the evaluation interaction main chain is assigned a stage number to obtain the intermittent answering stage label corresponding to each interaction unit.

[0044] Furthermore, generating interruption semantic description data corresponding to each stage of intermittent response specifically includes: standardizing the inter-question pause duration based on the average pause level during the current assessment process for the elderly, to obtain a standardized pause intensity; calculating the external interruption indicator value corresponding to each interaction unit based on the standardized pause intensity, the number of answer changes, the review depth, and the task continuity after the pause; calculating the pathological lag indicator value corresponding to each interaction unit based on the standardized pause intensity, the number of answer changes, the review depth, and the task continuity after the pause; and using the external interruption indicator value and the pathological lag indicator value as components of the interruption semantic description data.

[0045] Furthermore, the interruption semantic description data also includes stage anomaly density. The generation of the stage anomaly density specifically includes: extracting a preset number of interaction units after the start position of each intermittent response stage, and calculating the proportion of the preset number of interaction units that revisit the set of incomplete tasks; if the proportion is greater than a preset recovery threshold, marking the corresponding intermittent response stage as a recovery stage; if the proportion is not greater than the preset recovery threshold, marking the corresponding intermittent response stage as a transitional stage; aggregating the pathological lag indicator values ​​corresponding to each interaction unit within the same intermittent response stage to obtain the stage anomaly density corresponding to the intermittent response stage; and writing the stage anomaly density into the interruption semantic description data.

[0046] In this embodiment of the invention, step S3 is mainly used to address a core challenge in the context of intermittent answering: how to identify which pauses are merely temporary interruptions in the normal answering rhythm, and which pauses signify a genuine break in the cognitive progression stage. It should be noted that this invention does not use a fixed duration threshold to mechanically judge long pauses, because the answering rhythms of elderly individuals vary, and the natural answering time also differs between different questions. Simply defining pauses exceeding a certain number of seconds as abnormal can easily lead to two types of problems: one is misjudging normally but relatively slow answers as abnormal, and the other is burying truly pathologically significant stage-based pauses among individual differences. Therefore, this invention employs a dynamic, context-dependent stage boundary determination method in this step.

[0047] Specifically, the distribution of answer switching behavior, cross-question jumping behavior, and correcting behavior can be statistically analyzed within the adjacent windows of each interactive unit, and the local behavioral instability can be constructed based on this. This local behavioral instability is used to reflect whether there are significant fluctuations in the answering pattern within a certain period of time near the nth interaction unit. It is easy to understand that if there are many jumps and repeated modifications to the answer within a certain local window, then the answering in that area is inherently more complex. In this case, a relatively long pause may not necessarily mean that cognitive progress has been truly interrupted. On the other hand, if the overall behavior of a certain local window is relatively stable, but a long pause suddenly occurs, then that pause deserves more attention.

[0048] Based on this, the present invention constructs a dynamic boundary threshold by combining the inter-question pause duration and the local behavioral instability. Its expression is: ; in, This represents the average pause duration between questions within the adjacent windows of the nth interactive unit. This represents the standard deviation of the pause duration between questions within the adjacent windows of the nth interactive unit. This represents the local behavioral instability within the windows adjacent to the nth interaction unit. and This represents the weighting coefficient. In this way, the present invention makes pause boundary determination no longer dependent on a fixed number of seconds, but on the relative degree of anomaly defined by the current local context. This allows boundary recognition to more closely resemble the actual response process of elderly people.

[0049] Furthermore, this invention does not only divide stages based on whether the pause is long enough, but also incorporates the task continuity after the pause obtained in step S2. In one executable implementation, the stage boundary marker can be obtained using the following formula. : ; in, This indicates whether the nth interaction unit constitutes the starting point of a new discontinuous response phase. Indicates the duration of pauses between questions. Indicates dynamic boundary threshold, Indicates the continuity of the task after the pause. This represents the task continuity threshold. It can be seen that this invention only determines a position as a new stage boundary when the current pause is relatively significant and the original task chain is not effectively returned after the pause. The direct benefit of this setting is that it can distinguish situations where the original task can be resumed despite a longer pause from true stage breaks.

[0050] After obtaining the stage boundary markers, each interaction unit in the evaluation interaction main chain can be further numbered to form multiple discontinuous response stages. It should be noted that the stage obtained here is not just a time segmentation concept, but a stage division result with task progression significance and behavioral coherence significance. That is to say, multiple interaction units within the same stage are more likely to belong to the same cognitive progression chain, while different stages are more likely to indicate changes in response strategy, task path, or cognitive state.

[0051] After completing the stage division, this invention further generates interruption semantic description data for each discontinuous answering stage. First, the pause duration between questions can be standardized to obtain standardized pause intensity. The standardized pause intensity is essentially used to eliminate individual rhythm differences among older adults, so that subsequent analysis focuses on how abnormal the pause is relative to the overall rhythm of the individual in this assessment, rather than just focusing on the absolute number of seconds of the pause.

[0052] Furthermore, this invention is based on standardized pause intensity. Number of times the answer was changed Looking back at the depth and task continuity after pause Calculate the external interrupt indication value respectively. and pathological lag indicator value Among them, the external interrupt indication value The expression is: ; in, This indicates that the nth interaction unit is more likely to receive an indication value caused by an external interruption. This represents the Sigmoid function. , , and This represents the weighting parameter. The basic idea behind this formula is that if a pause is significant, but the task continues for a long time after the pause, and the depth of repeated modification and review of the answer is not significant, then the pause is more likely caused by external interference, temporary distraction, or a short break.

[0053] Correspondingly, pathological hysteresis indicator value The expression is: ; in, This indicates that the nth interaction unit is more likely to produce an indicator value caused by pathological cognitive delay. , , and The formula represents the weighting parameter. It shows that when a pause is not only strong but also accompanied by a high number of answer changes, a high replay depth, and a low task continuity after the pause, it is more consistent with the behavioral characteristics exhibited by pathological cognitive delay. It should be noted that this invention does not simply make one indicator value equal to the complementary value of another indicator value, but rather uses two independent calculation channels to express two different interpretation paths. This better adapts to real-world elderly response data with complex boundaries and mixed performance.

[0054] Based on this, the present invention further aggregates the pathological lag indicator values ​​corresponding to each interaction unit within the same discontinuous response phase, thereby obtaining the phase anomaly density. Its expression is: ; in, This represents the set of interactive units included in the k-th intermittent response phase. This indicates the number of interaction units in this stage. This represents the stage anomaly density for the k-th discontinuous response stage. The stage anomaly density describes the concentration of pathological abnormal signals within a given stage. It is easy to understand that an anomaly in a single interaction unit may stem from an occasional difficulty in a question, but if multiple interaction units within the same stage exhibit high pathological lag indicators, then that stage is more likely to have sustained anomaly significance.

[0055] In one specific implementation, if an elderly person experiences a significant pause after a period of continuous answering, but then immediately returns to a question that was not yet submitted stably before the pause and quickly completes the answer, this pause may correspond to a high exogenous interruption indicator value, while the stage anomaly density of that stage will not increase significantly. Conversely, if the elderly person fails to return to the original task chain after a long pause, but frequently changes their answers between multiple questions, resulting in multiple changes and a persistent situation, the pathological lag indicator values ​​of multiple interaction units will be correspondingly increased, thereby increasing the stage anomaly density of that stage. Thus, in step S3, this invention not only achieves the identification of discontinuous answering boundaries, but also further provides an explanation basis at the semantic level of interruption for subsequent brain disease prediction.

[0056] S4: Perform correlation calculations based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interrupted semantic description data to obtain the brain disease risk probability, stage risk results, and abnormal contribution results of interaction units.

[0057] Specifically, the response duration, standardized pause intensity, number of answer changes, review depth, task continuity after pause, external interruption indicator value, pathological delay indicator value, and stage abnormality density corresponding to each interactive unit are combined into an interaction feature vector. Based on the interaction feature vector and the intermittent response stage label, a stage continuity coefficient and a cross-stage barrier coefficient are constructed between interactive units. The stage continuity coefficient characterizes the continuity relationship between different interactive units at the stage level and the review depth level, and the cross-stage barrier coefficient characterizes whether different interactive units cross stage boundaries. Based on the stage continuity coefficient and the cross-stage barrier coefficient... The association weights between interactive units are calculated using the barrier coefficient, the external interruption indicator value, and the pathological lag indicator value. Based on these association weights, the information of historical interactive units is weighted and aggregated to obtain the association state representation of the current interactive unit. If the external interruption indicator value corresponding to the current interactive unit is greater than the corresponding pathological lag indicator value, and the current interactive unit and the previous interactive unit belong to different discontinuous response stages, the inheritance ratio of the preceding association state to the current interactive unit is reduced. In other cases, the continuous transmission of the preceding association state is maintained. A global state representation for predicting brain disease risk is generated based on the association state representations of all interactive units.

[0058] Furthermore, based on the global state representation and the statistical results of pathological lag indicator values, replay depth, number of answer changes, and stage anomaly density corresponding to each interaction unit, the probability of brain disease risk is calculated; a stage risk value is calculated for each intermittent response stage, and a stage risk result is generated based on the stage risk value of each intermittent response stage; the confidence level of the current brain disease prediction is calculated based on the dispersion between the stage risk values ​​of each intermittent response stage; and anomaly contribution values ​​corresponding to each interaction unit are calculated based on the pathological lag indicator values ​​corresponding to each interaction unit and the average strength associated with other interaction units, to form anomaly contribution results for the interaction units.

[0059] In this embodiment of the invention, step S4 is mainly used to uniformly associate and process the various behavioral and semantic quantities obtained in the preceding steps, thereby forming the final risk calculation result. It should be noted that in scenarios where elderly people respond intermittently, simply splicing and accumulating all interaction units in chronological order can easily lead to two types of errors: one is misinterpreting data that is clearly due to exogenous interruptions as disease risk signals, and the other is breaking up truly meaningful pathological behavioral chains. Therefore, in this step, the invention does not directly perform ordinary temporal accumulation, but instead utilizes stage information, semantic information, and interaction relationship information to specifically adjust the information propagation weights between different interaction units.

[0060] Specifically, the duration of each interactive unit's response can be determined first. Standardized pause intensity Number of times the answer was changed Looking back at the depth Task continuity after pause External interruption indication value Pathological lag indicator value and stage anomaly density The data are combined to form an interactive feature vector. The purpose of this setup is to fully incorporate all the key intermediate data obtained in steps S2 and S3 into the subsequent risk calculation process.

[0061] After obtaining the interaction feature vectors of each interaction unit, this invention further constructs a phase continuity coefficient between interaction units based on the intermittent response phase labels and the replay relationship. and cross-stage barrier coefficient The stage continuity coefficient describes the strong coherence between different interaction units at the stage and review relationship levels; the cross-stage barrier coefficient describes whether two interaction units cross the boundary of a discontinuous response stage. It's easy to understand that even if two interaction units are similar in content, if they cross a significant interruption boundary with a high degree of external interruption, the former should not have an excessive influence on the latter. Conversely, if two interaction units are not completely continuous in time but are in the same stage or belong to the same restorative task chain, they should be allowed to maintain a strong connection.

[0062] Based on this, the present invention further calculates the association weights between interactive units. Its expression is: ; in, This represents the association weight between the i-th interaction unit and the j-th interaction unit. This represents the basic correlation value between the i-th interaction unit and the j-th interaction unit, obtained based on the interaction feature vector. Represents the continuity coefficient of the stage. Indicates the cross-stage barrier coefficient. Indicates the external interrupt indication value. Indicates the pathological lag indicator value. , and This represents the weighting parameter. The technical idea embodied in this formula is: based on the basic correlation, increase the association weight of interaction units with strong stage continuity, increase the attention of interaction units with strong pathological lag characteristics, and reduce the propagation impact of interaction units that cross stages and have strong exogenous interruption characteristics.

[0063] Furthermore, in this embodiment, the present invention also controls whether historical states continue to be passed on based on the relative relationship between the external interruption indicator value and the pathological delay indicator value of the current interaction unit. Specifically, when the external interruption indicator value corresponding to the current interaction unit is greater than the corresponding pathological delay indicator value, and the current interaction unit and the previous interaction unit belong to different intermittent response stages, the inheritance ratio of the preceding associated state to the current interaction unit is reduced; in other cases, the continuous transmission of the preceding state is retained. The reason for this setting is that if the current stage is more likely to be caused by external factors, the preceding state should not be excessively carried into the current stage, otherwise it will amplify the external noise; if the current stage is more likely to be a continuation of a pathological delay chain, the preceding state should be retained in order to characterize the persistence and accumulation of abnormal behavior.

[0064] After completing the above association calculations, the association state representation of all interactive units can be obtained, and a global state representation for predicting the risk of brain diseases can be further formed. Subsequently, based on the global state representation and the statistical results of the pathological lag indicator value, replay depth, number of answer changes, and stage abnormality density corresponding to each interactive unit, the present invention calculates the brain disease risk probability p, the expression of which is: ; Where F represents the comprehensive feature vector formed by combining the global state representation with the above statistical results, W represents the mapping parameter, b represents the bias parameter, and p represents the probability of brain disease risk. It should be noted that this probability of risk is not directly determined by a single behavioral quantity, but rather by the comprehensive structured behavioral patterns throughout the entire response process. Therefore, the output results can more comprehensively reflect the degree of abnormality in the elderly person's assessment process.

[0065] Furthermore, this invention also calculates a stage risk value for each discontinuous response stage, and generates a confidence level Conf based on the dispersion between the stage risk values, the expression of which is: ; in, This represents the stage risk value corresponding to the k-th intermittent response stage. This represents the variance of the risk values ​​for each stage of the intermittent response. The confidence level is used to measure the consistency of the brain disease prediction across different stages. If the differences in risk values ​​between stages are small, it indicates that the system's overall judgment on this assessment is relatively consistent, and the confidence level is high; if only a few stages are significantly higher while the rest are lower, the confidence level is relatively low, indicating that the risk assessment relies more on the local abnormality stage.

[0066] Building upon this, the present invention further calculates the abnormal contribution value of each interaction unit based on the pathological lag indicator value corresponding to each interaction unit and the average intensity associated with other interaction units, thus forming an abnormal contribution result for the interaction units. It is easy to understand that in practical use, doctors or assessors often not only care about the overall risk but also want to know which response stages are most abnormal and which pauses and revisions are most noteworthy. By generating abnormal contribution results for interaction units, the present invention can further identify key interaction locations that significantly influence the final conclusion of brain disease risk, thereby improving the interpretability of the results.

[0067] In one specific implementation, if an elderly person experiences only one relatively significant long pause during the assessment, but is able to resume the original task after the pause and exhibits relatively stable subsequent behavior, the external interruption indicator value corresponding to this long pause may be high, its cross-stage propagation effect will be suppressed, and the overall risk probability will not be unreasonably increased. Conversely, if an elderly person exhibits high pause intensity, high answer change, and high review depth in several consecutive stages, and is unable to return to the original task chain after the pause, the pathological lag indicator values ​​and stage abnormality density of multiple stages will increase synchronously. The correlation calculation will retain the persistence of these abnormal behaviors, thereby driving the global state representation towards a high-risk direction. Thus, in step S4, this invention achieves refined processing of whether the abnormality is continuous, whether the abnormality is interpretable, and whether the abnormality should be accumulated.

[0068] S5: Based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, perform result determination and output brain disease prediction results for medical assessment of the elderly.

[0069] Specifically, the probability of brain disease risk is compared with preset low-risk and high-risk thresholds to obtain the corresponding initial risk level; if the initial risk level is in the medium-to-high-risk boundary region, a consistency judgment is performed based on the confidence level and the stage risk result to determine the final risk level; interactive units with contribution values ​​higher than a preset contribution threshold are extracted from the abnormal contribution results of the interactive units, and corresponding abnormal stage description information is generated; the final risk level, the probability of brain disease risk, the stage risk result, and the abnormal stage description information are structured and organized to output the brain disease prediction result; if the final risk level reaches a preset referral threshold, further specialist evaluation prompts are added to the brain disease prediction result.

[0070] In this embodiment of the invention, step S5 is mainly used to further transform the risk calculation results obtained in step S4 into structured output results suitable for medical assessment scenarios for the elderly. It should be noted that in medical assessment scenarios, simply providing a numerical probability is usually insufficient. Practical users typically also want to know the risk level, the phased risk distribution, the location of key abnormalities, and whether further specialist assessment is needed. Therefore, in this step, the invention not only determines the probability of brain disease risk itself but also combines the phased risk results with the abnormal contribution results of the interaction units to form a more operable final output.

[0071] Specifically, the probability of brain disease risk can first be compared with preset low-risk and high-risk thresholds to obtain the corresponding initial risk level. It is easy to understand that the low-risk and high-risk thresholds can be set according to different application scenarios, different assessment items, different institutional requirements, and different characteristics of the elderly population; they are not limited to a fixed value in this invention. In this way, a preliminary risk classification can be made for the assessment results.

[0072] Furthermore, when the initial risk level is in the medium-to-high risk boundary region, this invention does not directly use the risk probability alone as the final judgment result. Instead, it further combines confidence level and stage risk results to perform a consistency judgment. In other words, if the overall risk probability is in the high range, but only one intermittent response stage shows obvious abnormalities while other stages are generally stable, the system can cautiously adjust the final level based on the lower stage consistency. However, if multiple stages show high risk with small differences between stages, it indicates that the risk judgment has better stability and reliability, and a higher level result can be output more clearly in this case.

[0073] Simultaneously, this invention also extracts interactive units whose contribution values ​​exceed a preset contribution threshold from the abnormal contribution results of the interactive units, and forms abnormal stage description information accordingly. The abnormal stage description information may include the question identifier corresponding to the high-contribution interactive unit, the current stage, whether a long pause occurred, whether the task returned to its original state after the pause, whether it was accompanied by a high number of answer changes, and whether there was a deep review. In this way, the output of this invention is no longer a simple black-box risk value, but a structured assessment result with behavioral explanation clues. This setup helps doctors, assessors, or subsequent reviewers quickly locate the most noteworthy abnormalities in the assessment.

[0074] After completing the above processing, this invention unifies the final risk level, the probability of brain disease risk, the stage risk results, and the description information of the abnormal stage, and outputs the brain disease prediction results. If the final risk level reaches a preset referral threshold, further specialist assessment prompts can be added to the output results, such as suggestions for memory clinic screening, neurological examination, or further professional scale retesting.

[0075] In one specific implementation, if an elderly person's brain disease risk probability is 0.79, and multiple intermittent response stages show high stage risk values ​​with high confidence levels, the system can classify the final result as high-risk. The abnormal stage description information will indicate that the main anomalies are concentrated in stages where the original task was not resumed after several pauses, accompanied by frequent revisions. Conversely, if an elderly person's brain disease risk probability is 0.56, but only one local stage is significantly high with low overall confidence, the system can classify it as medium-risk or pending review, and suggest further evaluation to avoid overjudgment due to sporadic local anomalies.

[0076] Reference Figure 2 , Figure 2 This is a schematic diagram of the brain disease prediction system for medical assessment of the elderly, according to an embodiment of the present invention. Figure 2 As shown, in an optional embodiment, the present invention also proposes a brain disease prediction system for medical assessment of the elderly, comprising: The acquisition module 10 is used to acquire the original assessment interaction data in the process of digital medical assessment for the elderly, and to establish an assessment interaction main chain based on the original assessment interaction data. Module 20 is established to extract the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause based on the evaluation interaction main chain, and to establish a stage reconstruction benchmark to describe the discontinuous answering behavior. The generation module 30 is used to identify the discontinuous response boundary based on the stage reconstruction benchmark, divide the evaluation interaction main chain into stages, and generate interruption semantic description data corresponding to each discontinuous response stage. The calculation module 40 is used to perform correlation calculations based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interrupted semantic description data to obtain the brain disease risk probability, stage risk results, and abnormal contribution results of the interaction unit. The output module 50 is used to perform result determination based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, and output brain disease prediction results for medical assessment of the elderly.

[0077] Other embodiments or specific implementations of the brain disease prediction system for medical assessment of the elderly of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0078] It is understood that in the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Nth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0079] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0080] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A brain disease prediction method for medical assessment of the elderly, characterized by, The method includes the following steps: Obtain the original assessment interaction data during the digital medical assessment process for the elderly, and establish an assessment interaction main chain based on the original assessment interaction data; Based on the main chain of the evaluation interaction, the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause are extracted to establish a stage reconstruction benchmark for describing intermittent answering behavior. Based on the stage reconstruction benchmark, the discontinuous response boundary is identified, the evaluation interaction main chain is divided into stages, and interruption semantic description data corresponding to each discontinuous response stage is generated. Based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interrupted semantic description data, correlation calculations are performed to obtain the brain disease risk probability, stage risk results, and abnormal contribution results of interaction units. The system performs a result determination based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, and outputs a brain disease prediction result for medical assessment of the elderly.

2. The method for predicting brain diseases for medical assessment of the elderly as described in claim 1, characterized in that, Acquire raw assessment interaction data during the digital medical assessment process for the elderly, and establish an assessment interaction main chain based on the raw assessment interaction data, specifically including: Collect the question identifier data, answer start timestamp data, answer end timestamp data, and answer change record data for each assessment question during the digital medical assessment process for the elderly, as the original assessment interaction data; Based on the question identifier data, the answer start timestamp data, the answer end timestamp data, and the answer change record data, an interactive unit corresponding to each evaluation question is constructed; The interaction units are arranged in chronological order according to the timestamp data of the start of the response, so as to form an interaction unit sequence; If the assessment question corresponding to the same question identifier is reopened and the answer is modified, the response process after reopening is recorded as a new interaction unit, and this new interaction unit has the same question identifier as the previous interaction unit. The sorted sequence of interaction units is designated as the evaluation interaction main chain, serving as the data basis for subsequent extraction of discontinuous response behavior features and the execution of brain disease prediction.

3. The method for predicting brain diseases for medical assessment of the elderly as described in claim 2, characterized in that, Based on the main chain of the evaluation interaction, the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the task continuity after pauses are extracted to establish a stage reconstruction benchmark for describing discontinuous answering behavior, specifically including: Based on the start and end timestamps of each interaction unit in the evaluation interaction main chain, the duration of each interaction unit is calculated. Based on the answer end timestamp data and answer start timestamp data of adjacent interactive units in the evaluation interaction main chain, calculate the inter-question pause duration corresponding to the nth interactive unit; Based on the number of answer states within the corresponding interaction unit in the answer change record data, the number of answer changes is extracted; Based on the number of interactive units with the same question identifier that appeared before the current interactive unit in the evaluation interaction main chain, the replay depth is extracted. Based on the identifiers of questions that have not yet formed a stable submission state or have a tendency to be modified again in a short period of time before the pause occurs, an incomplete task set is constructed, and the re-access of the incomplete task set by a preset number of interactive units after the pause is counted to obtain the task continuity after the pause. The duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause are uniformly organized and used as the benchmark for the stage reconstruction.

4. The method for predicting brain diseases for medical assessment of the elderly as described in claim 1, characterized in that, Based on the aforementioned stage reconstruction benchmark to identify discontinuous response boundaries, the evaluation interaction main chain is divided into stages, specifically including: Based on the aforementioned stage reconstruction benchmark, the distribution of answer switching behavior, cross-question jumping behavior, and reversion behavior is statistically analyzed within the adjacent windows of each interactive unit to obtain the local behavioral instability of each corresponding interactive unit. Based on the local behavior instability and the inter-question pause duration, construct the dynamic boundary threshold corresponding to the nth interaction unit; Based on the inter-question pause duration, the dynamic boundary threshold, and the task continuity after the pause, calculate the corresponding stage boundary marker; Based on the stage boundary markers, each interaction unit in the evaluation interaction main chain is assigned a stage number to obtain the discontinuous response stage label corresponding to each interaction unit.

5. The method for predicting brain diseases for medical assessment of the elderly as described in claim 4, characterized in that, Generate interruption semantic description data corresponding to each stage of the intermittent response, specifically including: Based on the pause duration between questions, the pause duration between questions in each interactive unit is standardized according to the average pause level of the current elderly people in this assessment process, so as to obtain the standardized pause intensity. Based on the standardized pause intensity, the number of answer changes, the review depth, and the task continuity after the pause, calculate the external interruption indication value corresponding to each interaction unit; Based on the standardized pause intensity, the number of answer changes, the review depth, and the task continuity after the pause, calculate the pathological lag indicator value corresponding to each interaction unit; The exogenous interruption indication value and the pathological lag indication value are used as components of the interruption semantic description data.

6. The method for predicting brain diseases for medical assessment of the elderly as described in claim 5, characterized in that, The interruption semantic description data also includes stage anomaly density, the generation of which specifically includes: After the starting position of each intermittent response stage, a preset number of interaction units are extracted, and the proportion of the preset number of interaction units that revisit the set of incomplete tasks is counted. If the proportion is greater than the preset recovery threshold, the corresponding intermittent response stage is marked as a recovery stage; if the proportion is not greater than the preset recovery threshold, the corresponding intermittent response stage is marked as a transition stage. Aggregate the pathological lag indicator values ​​corresponding to each interaction unit within the same discontinuous response phase to obtain the phase anomaly density corresponding to the discontinuous response phase. Write the stage anomaly density into the interruption semantic description data.

7. The method for predicting brain diseases for medical assessment of the elderly as described in claim 6, characterized in that, Based on the evaluation interaction main chain, the phase reconstruction benchmark, and the interruption semantic description data, correlation calculations are performed, specifically including: The response duration, standardized pause intensity, number of answer changes, review depth, task continuity after pause, external interruption indicator value, pathological lag indicator value, and stage abnormality density of each interactive unit are combined into an interactive feature vector. Based on the interaction feature vector and the discontinuous response stage label, a stage continuity coefficient and a cross-stage barrier coefficient are constructed between interaction units; wherein, the stage continuity coefficient is used to characterize the continuity relationship between different interaction units at the stage level and the replay depth level, and the cross-stage barrier coefficient is used to characterize whether different interaction units cross stage boundaries. Based on the stage continuity coefficient, the cross-stage barrier coefficient, the exogenous interruption indicator value, and the pathological lag indicator value, the association weight between the interaction units is calculated; The information of historical interaction units is weighted and aggregated based on the association weights to obtain the association state representation of the current interaction unit. If the external interruption indicator value corresponding to the current interaction unit is greater than the corresponding pathological hysteresis indicator value, and the current interaction unit and the previous interaction unit belong to different intermittent response stages, reduce the inheritance ratio of the preceding associated state to the current interaction unit; otherwise, maintain the continuous transmission of the preceding associated state. A global state representation for predicting brain disease risk is generated based on the associated state representation of all interactive units.

8. The method for predicting brain diseases for medical assessment of the elderly as described in claim 7, characterized in that, The results obtained include the probability of brain disease risk, stage risk results, and abnormal contribution results of interaction units, specifically including: Based on the global state representation and the statistical results of pathological lag indicators, replay depth, number of answer changes and stage abnormality density corresponding to each interactive unit, the probability of brain disease risk is calculated. Calculate the stage risk value for each discontinuous response stage, and generate the stage risk result based on the stage risk value for each discontinuous response stage; The confidence level of this brain disease prediction is calculated based on the degree of dispersion between the stage risk values ​​of each intermittent response stage. Based on the pathological hysteresis indicator value corresponding to each interaction unit and the average intensity associated with other interaction units, the abnormal contribution value corresponding to each interaction unit is calculated to form the abnormal contribution result of the interaction unit.

9. The method for predicting brain diseases for medical assessment of the elderly as described in claim 1, characterized in that, Based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, a result determination is performed, and a brain disease prediction result for medical assessment of the elderly is output, specifically including: The probability of brain disease risk is compared with a preset low-risk threshold and a preset high-risk threshold to obtain the corresponding initial risk level; If the initial risk level is in the medium-to-high risk boundary area, a consistency determination is performed by combining the confidence level and the stage risk results to determine the final risk level. Extract interactive units whose contribution values ​​are higher than a preset contribution threshold from the abnormal contribution results of the interactive units, and generate corresponding abnormal stage description information; The final risk level, the probability of brain disease risk, the stage risk result, and the abnormal stage description information are structured and organized to output the brain disease prediction result. If the final risk level reaches the preset referral threshold, further specialist evaluation prompts will be added to the brain disease prediction results.

10. A brain disease prediction system for medical assessment of the elderly, characterized in that, The system includes: The acquisition module is used to acquire the original assessment interaction data in the process of digital medical assessment for the elderly, and to establish an assessment interaction main chain based on the original assessment interaction data. A module is established to extract the duration of answering, the duration of pauses between questions, the number of answer changes, the depth of review, and the continuity of the task after the pause based on the main chain of the evaluation interaction, and to establish a stage reconstruction benchmark to describe the discontinuous answering behavior. The generation module is used to identify discontinuous response boundaries based on the stage reconstruction benchmark, divide the evaluation interaction main chain into stages, and generate interruption semantic description data corresponding to each discontinuous response stage. The calculation module is used to perform correlation calculations based on the evaluation interaction main chain, the stage reconstruction benchmark, and the interrupted semantic description data to obtain the brain disease risk probability, stage risk results, and abnormal contribution results of interaction units. The output module is used to perform result determination based on the brain disease risk probability, the stage risk result, and the abnormal contribution result of the interaction unit, and output brain disease prediction results for medical assessment of the elderly.

Images