Assisted alzheimer's evaluation method and system based on clock drawing test

Abstract

The application discloses an Alzheimer's disease auxiliary evaluation method and system based on a clock drawing test, and relates to the technical field of Alzheimer's disease auxiliary evaluation. Firstly, a unified space reference direction is determined according to the posture of a subject to correct the test image rotation, thereby reducing the influence of posture differences on the space analysis result. On this basis, the clock face region is processed by partitioning, and a lateral aggregation index is constructed through the number relationship in each partition to quantitatively represent the abnormality degree of the digital space distribution. At the same time, the drawing time is introduced into the evaluation process, and the evaluation threshold is dynamically adjusted by comparing with the reference time range, so that the judgment standard can adapt to the drawing behavior characteristics of different subjects. The method takes into account the drawing result characteristics and the drawing process information, improves the stability and consistency of the evaluation result, and is beneficial to improving the objectivity and reliability of the clock drawing test in the auxiliary evaluation of Alzheimer's disease.

Description

Technical Field

[0001] This invention relates to the field of Alzheimer's disease auxiliary assessment technology, specifically to an auxiliary Alzheimer's disease assessment method and system based on the clock drawing test. Background Technology

[0002] Alzheimer's disease is a neurodegenerative disease characterized by progressive cognitive decline, often accompanied by abnormalities in visuospatial processing and executive functions in its early stages. The clock drawing test, as a simple and intuitive neuropsychological assessment method, is widely used as an adjunct to the screening and evaluation of cognitive impairment. This test assesses a subject's spatial cognition and planning abilities by analyzing the structure of the clock face drawn, the distribution of numbers, and the completion status, and has high application value in clinical practice.

[0003] Existing assessment methods based on the clock drawing test primarily rely on the static appearance of the drawing results to determine abnormal number distribution or layout imbalance. However, in practice, different subjects exhibit significant differences in drawing speed, resulting in substantial individual variations in completion time. Without incorporating information related to the drawing process into the assessment, relying solely on fixed judgment criteria to analyze the results makes it difficult to guarantee the stability and consistency of the assessment results across different subjects.

[0004] Furthermore, existing technologies, when automating the analysis of clock drawing test results, typically fail to effectively correlate the drawing process information with the spatial distribution characteristics of the numbers. They lack targeted quantitative processing of the differences in the distribution of numbers in different areas of the clock face, making it difficult to accurately reflect the spatial cognitive abnormalities associated with Alzheimer's disease. As a result, they are still insufficient in terms of the accuracy and adaptability of the auxiliary assessment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an auxiliary Alzheimer's disease assessment method and system based on the clock drawing test.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses an auxiliary Alzheimer's disease assessment method based on a clock drawing test, comprising the following steps:

[0008] Acquire the test image drawn by the subject during the clock drawing test, and simultaneously acquire the drawing time of the test image and the orientation information representing the subject's drawing posture.

[0009] Based on the orientation information, the spatial reference direction of the subject is determined, and the test image is rotated and corrected according to the spatial reference direction to obtain a corrected test image.

[0010] Identify the clock face area and clock numerals from the calibration test image;

[0011] The clock face area is divided into sections by the vertical central axis passing through the center of the circle in the clock face area, and a first set and a second set are formed based on the clock digits in each section.

[0012] Based on the proportional relationship between the number of clock digits in the first set and the second set, a lateral clustering index is generated to characterize the degree of spatial abnormality in the spatial distribution of the clock digits drawn by the subjects.

[0013] The drawing duration is compared with a preset reference duration range, and the threshold adjustment direction is determined based on the comparison result;

[0014] Based on the difference between the drawing duration and the boundary value of the reference duration range, and combined with the quantitative relationship of clock digit identifiers in the first set and the second set, the threshold adjustment range is calculated;

[0015] Based on the threshold adjustment direction and the threshold adjustment amplitude, the preset basic threshold is processed to obtain a dynamic threshold;

[0016] The lateral aggregation index is compared with the dynamic threshold, and an evaluation signal is output based on the comparison result.

[0017] Secondly, this invention discloses an assisted Alzheimer's disease assessment system based on a clock drawing test, comprising:

[0018] The data acquisition module is used to acquire the test image drawn by the subject in the clock drawing test, and simultaneously acquire the drawing time of the test image and the orientation information representing the subject's drawing posture.

[0019] The spatial reference correction module is used to determine the spatial reference direction of the subject based on the orientation information, and to rotate and correct the test image according to the spatial reference direction to obtain a corrected test image.

[0020] The image recognition module is used to identify the clock face area and clock digit markings from the calibration test image;

[0021] The spatial partitioning module is used to partition the clock face area with the vertical central axis of the center of the clock face area as the boundary, and to form a first set and a second set based on the clock digits in each partition.

[0022] The lateralization assessment module is used to generate a lateralization clustering index to characterize the degree of spatial abnormality in the clock digits drawn by the subject, based on the proportional relationship between the number of clock digit identifiers in the first set and the second set.

[0023] The adjustment direction determination module is used to compare the drawing duration with a preset reference duration range and determine the threshold adjustment direction based on the comparison result;

[0024] The adjustment range calculation module is used to calculate the threshold adjustment range based on the difference between the drawing duration and the boundary value of the reference duration range, and in combination with the quantitative relationship of clock digit identifiers in the first set and the second set;

[0025] The dynamic threshold generation module is used to process the preset basic threshold according to the threshold adjustment direction and the threshold adjustment amplitude to obtain the dynamic threshold;

[0026] The evaluation output module is used to compare the lateral aggregation index with the dynamic threshold and output an evaluation signal based on the comparison result.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] 1. By introducing information related to the drawing process while obtaining the clock drawing test results, the evaluation no longer relies solely on the static results after the drawing is completed, but comprehensively considers the temporal characteristics of the drawing behavior, effectively making up for the problem that existing clock drawing test evaluation methods ignore the differences in the drawing process;

[0029] 2. Based on the relationship between the drawing time and the preset reference time range, the threshold is dynamically adjusted in terms of direction and amplitude. In this way, the evaluation criteria can be adaptively adjusted according to the differences in the drawing behavior of the subjects, avoiding the problem that the evaluation results deviate from the actual cognitive state due to the use of a uniform standard when the drawing speed is significantly too fast or too slow, thereby improving the rationality of the evaluation conclusion.

[0030] 3. By dividing the clock face area into zones and constructing a lateral clustering index based on the quantitative relationship of clock digits in different zones, the degree of abnormality in the distribution of digits in the clock drawing test can be quantitatively characterized. Compared with the judgment method that only relies on whether the overall layout is reasonable, this index can highlight the uneven distribution of digits in different areas, which is conducive to revealing the spatial distribution deviation related to cognitive function abnormalities, thereby improving the information utilization efficiency of the clock drawing test results in auxiliary assessment. Attached Figure Description

[0031] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0032] Figure 1 This is a flowchart of the method of the present invention;

[0033] Figure 2This is a data flow diagram of the present invention;

[0034] Figure 3 This is a system architecture diagram of the present invention. Detailed Implementation

[0035] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0036] This invention relates to an auxiliary Alzheimer's disease assessment method based on a clock drawing test, belonging to the field of medical auxiliary assessment and intelligent image analysis technology. Specifically, it is an intelligent assessment method that combines information from the drawing process with spatial distribution characteristics to quantitatively analyze the results of the clock drawing test and output an assessment signal.

[0037] The clock drawing test, a common tool for cognitive function screening, is widely used in the auxiliary assessment of Alzheimer's disease and related cognitive impairments due to its simplicity and low dependence on test subjects. Existing assessment methods often focus on the static analysis of the final drawing result, such as determining whether numbers are missing, the order is incorrect, or the layout is unbalanced. However, in practical applications, different test subjects exhibit significant differences in rhythm, speed, and operating habits during the drawing process. Analyzing the results based solely on fixed judgment rules makes it difficult to account for individual differences, thus limiting the stability and consistency of the assessment results. Furthermore, test subjects may exhibit postural shifts or inconsistent device placement angles during the drawing process, which can easily interfere with spatial distribution analysis and affect the accuracy of the assessment.

[0038] To address the aforementioned problems, the present invention aims to provide a clock drawing test evaluation method that can simultaneously incorporate drawing process information and uniformly correct the image spatial reference. By constructing an analysis mechanism that matches the actual drawing state of the subject, the reliability and discriminativeness of the evaluation results are improved.

[0039] To achieve the above objectives, the present invention employs the following technical means: First, while acquiring the clock drawing test image drawn by the subject, the drawing duration and directional information representing the drawing posture are simultaneously collected; based on the directional information, the spatial reference direction of the subject is determined, and the test image is rotated and corrected to eliminate the influence of posture differences on spatial analysis; in the corrected test image, the clock face area and its clock digit markings are identified, and the clock face area is divided into partitions with the vertical central axis of the clock face center as the boundary, forming corresponding digit sets; based on the proportional relationship of the number of digits in different partitions, a lateral clustering index is generated to reflect the degree of abnormality in the spatial distribution of digits; simultaneously, the subject's drawing duration is compared with a preset reference duration range, and the basic threshold is dynamically adjusted in combination with the digit distribution characteristics to obtain a dynamic threshold that is adapted to the subject's drawing process; finally, by comparing the lateral clustering index with the dynamic threshold, an evaluation signal for auxiliary assessment is output.

[0040] like Figure 1 , Figure 2 As shown, this application discloses an assisted Alzheimer's disease assessment method based on the clock drawing test, including the following steps:

[0041] A1. Obtain the test image drawn by the subject during the clock drawing test, and simultaneously obtain the drawing time of the test image and the orientation information representing the subject's drawing posture.

[0042] This step mainly acquires three sets of data: the test image, the drawing time of the test image, and the orientation information of the drawing posture.

[0043] The test image specifically refers to the clock image drawn by the subject on an electronic device according to the clock drawing test. The clock drawing test requires drawing the outline of the clock, marking all the numbers, and pointing the hands to a specified time. The drawn test image is essentially the external manifestation of the subject's internal cognitive blueprint (the appearance of the clock) and motor executive function working together, and can be used to analyze the subject's tendency to Alzheimer's disease.

[0044] Drawing time specifically refers to the total time a subject spends from the moment they begin to draw until they stop. This time reflects the subject's ability to understand and complete instructions during the test. This parameter characterizes the processing efficiency and resource allocation ability required to complete a complex cognitive-motor integrated task. Significantly abnormal durations (too fast or too slow) are sensitive behavioral markers of prodromal mild cognitive impairment or executive dysfunction, but duration alone cannot distinguish whether it stems from cognitive hesitation, motor slowness, or simply personality caution.

[0045] Orientation information is primarily acquired through external devices, such as inertial measurement units (e.g., gyroscopes, accelerometers) or visual sensors within smart pens, digital drawing tablets, or cameras. This parameter mainly defines the quantitative data of the orientation of the drawing plane (e.g., drawing paper) in three-dimensional space (e.g., pitch and roll angles with reference to the direction of gravity). It directly characterizes the spatial relationship between the subject's personal coordinate system and the absolute coordinate system of the environment (or the vertical direction of gravity) during the drawing task, providing a standard for subsequent corrections.

[0046] A2. Determine the spatial reference direction of the subject based on the orientation information, and rotate and correct the test image according to the spatial reference direction to obtain a corrected test image;

[0047] To calculate which direction is directly above the subject's subjective perception during the drawing process using orientation information, this can be achieved by extracting the horizontal component perpendicular to the direction of gravity from the orientation information. Based on the calculated spatial reference direction, a rigid body rotation transformation is performed on the test image. This rotation transformation is accomplished through coordinate system mapping, converting the test image from a sensor or camera coordinate system to a subject's subjective visual-spatial coordinate system.

[0048] The purpose of this step is primarily to correct the original test image, ensuring that all subsequent geometric analyses of the corrected test image are performed within the spatial framework intended by the subject.

[0049] For example, a subject who rotates the drawing paper 30 degrees clockwise to draw might subjectively perceive the 12 o'clock position as tilted in the original image. After correction, their subjective vertical axis aligns with the vertical axis of the image coordinates, ensuring that the vertical midline defined by the algorithm coincides as closely as possible with the mental midline in the subject's brain used to organize spatial layout. This is an absolute prerequisite for accurately assessing defects such as unilateral spatial neglect. Without this step, any subsequent automated spatial analysis based on the image's vertical / horizontal axes (such as dividing the image into left and right semicircles) will be based on an incorrect reference frame, leading to misjudgments.

[0050] A3. Identify the clock face area and clock numerals from the calibration test image;

[0051] The clock face region can be identified using algorithms such as edge detection and circular Hough transform to separate the closed contour representing the outer edge of the dial from the image and fit its geometric center. This center is the absolute origin for all subsequent spatial calculations in the entire scheme, and its positioning accuracy directly determines the accuracy of all angle and distance measurements.

[0052] Identifying clock digits can be achieved through connected component analysis, morphological operations, and potentially lightweight character recognition techniques to locate and distinguish the individual stroke regions representing the digits "1" to "12". The output consists of a position label (geometric center coordinates) and a semantic label (digital value) for each digit.

[0053] A4. Divide the clock face area into sections using the vertical central axis passing through the center of the circle in the clock face area as the boundary, and form a first set and a second set based on the clock digits in each section.

[0054] The division into zones using a vertical midline passing through the center corresponds to the neuroanatomical and functional principle that the two hemispheres of the human brain control contralateral spatial attention. The vertical midline of the center of the zone simulates the sagittal midline of the human brain. Based on the zone where the numerical identifier falls, it is assigned to either the first set (usually the left semicircle) or the second set (usually the right semicircle).

[0055] Qualitative observations of unilateral spatial neglect in clinical neuropsychological examinations ("the patient drew all the numbers on the right side of the clock") are transformed into computable binary data groupings. This partitioning rule forcibly transforms the abstract problem of clock face layout into a comparison of the distribution characteristics of elements in two independent spaces, creating conditions for subsequent quantitative modeling. It directly focuses the analysis on revealing specific spatial cognitive deficits caused by hemispheric functional asymmetry.

[0056] A5. Based on the proportional relationship between the number of clock digit identifiers in the first set and the second set, generate a lateral clustering index to characterize the degree of spatial abnormality in the spatial distribution of the clock digits drawn by the subjects.

[0057] This step is a quantitative calculation of the output set in step A4, which transforms the two sets of numbers obtained above into a scalar index.

[0058] The most typical ratio is: number of numbers in the second set / number of numbers in the first set. In an ideal, uniform, and complete clockwork design, each semicircle should contain 6 numbers, and this ratio is equal to or close to 1. Using this ratio as the core, or based on it through simple mathematical transformations (such as taking the logarithm to balance the positive and negative skewness, performing normalization, etc.), a continuous and measurable value is directly generated, namely the skewness clustering index.

[0059] The lateral clustering index characterizes the degree of imbalance in the spatial distribution of digits on the left and right sides of a subject's drawn clock. When unilateral spatial neglect exists, subjects tend to ignore one side of the space (e.g., the left side), resulting in a sharp decrease in the number of digits on that side, while the number of digits on the opposite side increases relatively, or even all of them are crowded onto the other side. This causes the ratio to deviate significantly from 1 (if the left side is the first set, the ratio is much greater than 1). Therefore, this index is the most direct and preliminary quantitative capture of the core pathological phenomenon of unilateral spatial neglect.

[0060] A6. Compare the drawing duration with the preset reference duration range, and determine the threshold adjustment direction based on the comparison result;

[0061] The preset reference duration range is established based on norm data from a large-scale healthy population and represents the normal time window for completing this task; the plotting duration of the subjects is compared with the preset reference duration range to determine the direction of adjustment;

[0062] This step transforms purely temporal behavioral data into a decision-oriented signal. It simulates the intuition of clinical experts considering the subject's execution speed during assessment, setting a personalized tone or tendency for subsequent quantitative judgment. It solves the problem that fixed thresholds cannot distinguish the different pathological possibilities implied by rapid and slow drawing errors.

[0063] The logic for determining the adjustment direction can be as follows:

[0064] If the drawing time is shorter than the lower limit of the reference time range, the adjustment direction is determined to be positive; this indicates that the judgment criteria need to be raised (i.e., the dynamic threshold used for subsequent comparisons should be increased) to make the judgment more stringent. Abnormally rapid drawing may reflect impulsivity, a simplified understanding of the task, or a deficiency in executive control, rather than genuine efficiency. A drawing completed in an extremely short time but containing spatial errors may have more pathological significance in terms of its errors.

[0065] If the drawing time exceeds the upper limit of the reference range, the adjustment direction is determined to be negative; this indicates that the judgment criteria need to be appropriately relaxed (i.e., the dynamic threshold needs to be lowered). Abnormal slowness may primarily stem from motor system sluggishness, extreme caution, or nonspecific fatigue, rather than specific spatial cognitive impairment. To avoid misjudging slight poor composition caused by these factors as positive, the threshold needs to be lowered.

[0066] If the drawing time falls within the reference range, the adjustment direction can be defined as zero adjustment, that is, the reference is not changed.

[0067] A7. Based on the difference between the drawing duration and the boundary value of the reference duration range, and combined with the quantitative relationship of clock digit identifiers in the first set and the second set, calculate the threshold adjustment range;

[0068] This step takes two data points as input: the difference between the plotted duration and the reference range boundary value, which reflects the degree of duration anomaly; and the relationship between the number of clock digits in the first set and the second set, which reflects the degree of spatial deviation.

[0069] The two sets of data mentioned above are correlated and processed according to the following logic: when time anomalies (whether too fast or too slow) and spatial skewness occur simultaneously, a larger adjustment amplitude should be generated. For example, when a painting is not only completed extremely quickly but also has severely skewed numbers, a large positive adjustment amplitude will be calculated; conversely, if the painting is slow but the numbers are evenly distributed, a small negative adjustment amplitude may be generated, or even no adjustment may be made.

[0070] The pathological significance of isolated duration abnormalities or isolated digit lateralization may be ambiguous; however, when both coexist, the signal strength indicating cognitive dysfunction is significantly enhanced. Through this fusion calculation, the modulation amplitude becomes a quantitative indicator characterizing the overall abnormal risk intensity or significance of the test results, providing precise data input for ultimately generating a discrimination criterion specifically matched to this test.

[0071] A8. Based on the threshold adjustment direction and the threshold adjustment amplitude, the preset basic threshold is processed to obtain a dynamic threshold;

[0072] The preset baseline threshold is an initial judgment threshold that is effective for the general population and has been determined based on a large amount of clinical data. The baseline threshold is then combined with the adjustment direction determined in step A6 and the adjustment range determined in step A7. The processing method can be mathematical calculation or other data processing methods.

[0073] After this step, the resulting dynamic threshold is no longer a general fixed value, but a discrimination criterion tailored to this subject's specific test.

[0074] For subjects whose behavioral patterns indicate high risk (such as fast and biased), the dynamic threshold becomes higher (more stringent), requiring their lateral convergence index to reach a higher level to trigger a positive result. This reduces the possibility of misjudging a positive result by random error (improving specificity).

[0075] For subjects whose behavioral patterns suggest the presence of confounding factors (such as slowness without deviation), their dynamic threshold will be appropriately lowered to avoid missing cases where mild spatial defects are masked by slow movement (increasing sensitivity).

[0076] Dynamic thresholds serve as a baseline for testing. They can fluctuate based on behavioral data (duration and preliminary spatial characteristics) provided by the subjects that reflect their testing status, thereby making the final judgment more individualized and clinically reasonable.

[0077] A9. Compare the lateral aggregation index with the dynamic threshold, and output an evaluation signal based on the comparison result.

[0078] The lateralization clustering index, which characterizes the severity of spatial lateralization of the subject and is generated in step A5, is numerically compared with the dynamic threshold generated in step A8, which incorporates the subject's behavioral context. Based on the comparison result, a binary or multi-level evaluation signal is output.

[0079] For example, when the lateral aggregation index is greater than the dynamic threshold, a positive signal is output indicating the presence of anomalous spatial neglected features.

[0080] The final decision is not simply to compare the observed values ​​with a norm, but rather to compare them with a personalized standard calibrated for the specific observation context. This mechanism significantly improves the accuracy and robustness of the system when dealing with diverse and heterogeneous real-world clinical subjects, representing a fundamental advancement and innovation compared to all existing technologies that use fixed thresholds or single-dimensional analysis.

[0081] This application further discloses the process for determining the threshold adjustment direction in step A6, specifically including:

[0082] B1. Obtain the temporal coordinate sequence of the handwriting drawn in the test image as temporal stroke sequence data;

[0083] A temporal coordinate sequence (temporal pen sequence data) refers to a sequence of spatial coordinates of handwriting points arranged in chronological order, recorded by an electronic writing tablet with coordinate sampling capabilities or by tracking the pen tip through high frame rate video. The spatial coordinates of each handwriting point include the x-coordinate, y-coordinate, and timestamp of that point in a coordinate system constructed based on the test image; the entire spatial coordinate sequence of handwriting points not only records what was drawn, but also the spatiotemporal trajectory of how it was drawn.

[0084] This data forms the basis for analyzing the drawing process as a dynamic time series. It enables the system to replay and quantify the drawing kinematics of the subjects, making it possible to detect micro-behavioral patterns (such as hesitation, planning, and correction) related to cognitive processing hidden behind the total duration.

[0085] B2. Based on the time-series pen sequence data, identify continuous pause events at the pen stroke generation points during the drawing process, and extract the interval duration of each pause event;

[0086] In the analysis of coordinate sequences, continuous pause events manifest as intervals where the spatial displacement of multiple consecutive points falls below a threshold but the time interval continues to increase. By identifying pauses during the drawing process, a pause event marks the cessation of an active brush stroke. The time difference between the end points of two adjacent pause events is calculated, representing the duration of the continuous drawing phase.

[0087] Pauses are generally considered to be external behavioral representations of increased cognitive load, planning, or memory retrieval. The distribution of retrieval intervals (i.e., the duration of continuous drawing), rather than the pause duration itself, better reflects the rhythm and stability of cognitive resource allocation. For example, stable, planned drawing is characterized by longer and more uniform intervals; while hesitant, frequently interrupted drawing is characterized by short and variable intervals. This step transforms the continuous coordinate flow into a series of discrete-time features representing cognitive rhythm.

[0088] B3. Based on the distribution of the interval duration, calculate the rendering smoothness correction factor to characterize the stability of the cognitive load during the rendering process;

[0089] This step involves statistically modeling the interval duration sequence extracted in step B2 to generate a rendering smoothness correction factor as an indicator for comprehensive evaluation.

[0090] The rendering smoothness correction factor is a scalar value calculated based on the interval duration distribution. Its calculation logic aims to quantify the stationarity and consistency of the rendering process.

[0091] For example, the coefficient of variation of the interval duration (the ratio of the standard deviation to the mean) or its reciprocal, or a negative correlation mapping, can be used as this factor. The smaller the coefficient of variation, the more stable the interval duration and the higher the fluency; conversely, the lower the fluency, the greater the fluctuation in cognitive load or the unstable executive control.

[0092] The drawing fluency correction factor integrates discrete interval duration sequences into a single indicator with a clear psychological interpretation—cognitive-motor execution fluency. As a moderating variable, it corrects initial judgments based solely on drawing duration (macroeconomic efficiency). Its purpose is to differentiate between "fast and stable" and "fast but flustered," and between "slow and steady" and "slow and struggling"—situations with the same total duration but drastically different cognitive load patterns—providing a basis for subsequent decision-making.

[0093] B4. Based on the comparison results between the drawing duration and the preset reference duration range, and in conjunction with the drawing smoothness correction factor, determine the threshold adjustment direction through predefined decision logic;

[0094] First, based on the drawing time (macro-efficiency), the subjects were divided into three macro-states: "too fast," "normal," or "too slow." Then, in the two abnormal states of "too fast" and "too slow," a drawing fluency correction factor (micro-quality) was introduced for secondary validation.

[0095] Specifically as follows:

[0096] If the rendering time is shorter than the normal range (refer to the lower limit of the rendering time range), it is initially judged that there may be impulsive or shallow processing. In this case, a rendering smoothness correction factor is introduced:

[0097] If the fluency is high (the fluency correction factor is higher than the first fluency threshold), it indicates a "skilled and confident speed," and the accompanying spatial errors are more likely to reflect real cognitive deficiencies, so positive adjustment (tightening the criteria) is adopted. If the fluency is low, it indicates a "chaotic and unstable speed," and the errors may partly stem from uncontrolled execution, the pathological significance of which needs to be questioned, so zero adjustment is adopted to avoid over-interpretation.

[0098] If the drawing time exceeds the normal range (refer to the upper limit of the time range), it is initially judged that there may be motor sluggishness or cognitive sluggishness. In this case, a drawing smoothness correction factor is introduced:

[0099] If the fluency is low (the fluency correction factor is below the second fluency threshold), it is confirmed as "difficult and intermittent slow," and its slight spatial errors are likely related to overall motor or severe cognitive impairment rather than specific spatial neglect, so negative adjustment (relaxed criteria) is adopted. If the fluency is high, it indicates "smooth but cautious slow," and its spatial errors need to be taken seriously, so "zero adjustment" is adopted.

[0100] The above process simulates the thought process of experts in differential diagnosis. By introducing fluency as a validator or veto, it enhances the intelligence of behavioral data interpretation. It effectively solves the problem that a single duration indicator cannot distinguish the intrinsic quality of behavior, preventing misjudgments (whether false positives or false negatives) caused by simply drawing fast or slow. Ultimately, it ensures that the determination of the threshold adjustment direction is based on a comprehensive evaluation of the subject's behavioral efficiency (duration) and behavioral quality (fluency), thus making the generated dynamic threshold more accurate and personalized, significantly improving the robustness and clinical applicability of the entire auxiliary recognition system.

[0101] Decision-making logic includes:

[0102] If the drawing duration falls within the reference duration range, the threshold adjustment direction is set to zero adjustment.

[0103] If the drawing time is shorter than the lower limit of the reference time range, then the first sub-logic is executed:

[0104] The rendering smoothness correction factor is compared with the first smoothness threshold. If the rendering smoothness correction factor is higher than the first smoothness threshold, the threshold adjustment direction is determined to be positive adjustment; otherwise, the threshold adjustment direction is determined to be zero adjustment.

[0105] If the drawing duration is longer than the upper limit of the reference duration range, then the second sub-logic is executed:

[0106] The rendering smoothness correction factor is compared with the second smoothness threshold. If the rendering smoothness correction factor is lower than the second smoothness threshold, the threshold adjustment direction is determined to be negative adjustment; otherwise, the threshold adjustment direction is determined to be zero adjustment.

[0107] Regarding the calculation process of the rendering smoothness correction factor disclosed in step B3, this application further discloses:

[0108] The calculation process for the smoothness correction factor includes:

[0109] B31. Structuring and establishing benchmarks for behavioral sequences;

[0110] The intervals between all identified pause events are arranged strictly according to the chronological order in which they occurred during the drawing process, forming an ordered sequence. , Indices representing ordered sequences. Indicates the first A time interval.

[0111] Calculate the arithmetic mean of all elements in the ordered sequence above as the average interval duration. The average interval duration serves as an internal benchmark for subsequent steps to define "short" and "long" intervals in a relative and personalized manner.

[0112] B32. Quantification of overall stability - basic smoothness value;

[0113] For the above ordered sequence Perform statistical analysis to calculate its statistical distribution characteristics to obtain a baseline fluency value. The baseline fluency value can typically be the variance, standard deviation, or coefficient of variation. The coefficient of variation (the ratio of the standard deviation to the mean) is the preferred option because it eliminates the influence of absolute speed and purely measures the relative amplitude of fluctuation.

[0114] The baseline fluency value is defined as negatively correlated with the degree of dispersion. For example, it can be directly taken as the reciprocal of the coefficient of variation, or the coefficient of variation can be mapped to a value using a monotonically decreasing function. Interval. The lower the dispersion (i.e., the more stable the interval duration), the higher the basic fluency value; conversely, the higher the dispersion, the lower the fluency value.

[0115] The baseline fluency score is a macro-level, generalized assessment of the smoothness of a subject's overall drawing process. A high baseline fluency score indicates stable, planned allocation of cognitive resources and motor output; a low score indicates a chaotic and unpredictable rhythm throughout the process, reflecting potential executive function or attention maintenance impairments.

[0116] B33, Detection of specific cognitive instability patterns - Target anomalous pattern scanning;

[0117] Based on the results of the previous two steps, the abnormal patterns of the target are identified;

[0118] The target anomaly pattern is precisely defined as a temporal structure characterized by "dense hesitation followed by prolonged lag". Its determination rule uses four parameters. Achieving quantification:

[0119] Defines the scan window length (number of consecutive pause events) to limit the analysis range.

[0120] Defines the minimum number of short-interval events that must occur within the window. .

[0121] Define the threshold for "short interval". This means that the short intervals must be significantly lower than the subject's own average pace.

[0122] Define the threshold for "long interval". This means that long intervals must be significantly higher than the average pace.

[0123] Slide an array on an ordered sequence containing... A window containing the number of consecutive pause events is checked to see if there are at least [number] pause events. The interval duration is less than A short event, and the next event in the window is an interval (i.e., a long interval event) greater than [a certain value]. Average duration. If the condition is met, the target abnormal pattern is identified.

[0124] The purpose of target anomalous patterns is to capture a specific phenomenon of cognitive instability: a series of short intervals (“intensive hesitation”) may characterize continuous difficulty in concept retrieval, subtle decision-making conflicts, or minor obstacles to motor initiation; while a subsequent unusually long interval (“prolonged lag”) may suggest a significant depletion of cognitive resources, a brief loss of focus on the task objective, or the need for a higher-level reorganization of the plan. This “burst-exhaustion” pattern is a more specific indicator of underlying cognitive dysfunction than overall instability.

[0125] B34. Classify and process the detected and undetected target anomaly patterns, and integrate the results of the previous steps to generate the final evaluation index.

[0126] If the aforementioned abnormal pattern is identified, it is considered that there is structural cognitive instability. At this point:

[0127] Calculating the negative correction: The negative correction is calculated based on two sub-features: the density of short-interval events and the duration of long-interval events. The higher the density and the longer the long interval, the larger the correction.

[0128] Generate the final correction factor: Subtract the negative correction amount from the base smoothness value. Based on the base smoothness value, an additional, targeted deduction is needed from the smoothness evaluation due to the discovery of specific harmful patterns.

[0129] If no abnormal pattern is detected, the lack of fluency in the drawing process is considered to be mainly due to random, unstructured fluctuations. In this case, the baseline fluency value calculated in step B32 is directly used as the final output. This indicates that fluency problems may stem from general inattention or poor motor control, rather than a specific type of discontinuous cognitive planning.

[0130] The process for generating the final correction factor is as follows:

[0131] Suppose that in step B33, a target anomalous pattern is identified at a certain position in the ordered interval duration sequence. For the target anomalous pattern, the following input parameters are defined and obtained:

[0132] In the continuous triggering of this mode Within each pause event window, the actual number of events judged as short intervals is recorded as follows: The density of short-interval events for: , The range of values ​​is The larger the value, the more frequent the hesitation and lag within the window.

[0133] The actual duration of the identified long-interval events is denoted as The average interval time of the entire ordered sequence is denoted as . Then the abnormal proportion for This value is non-negative; a larger value indicates a higher threshold for judging long-interval events relative to long-interval events. The more it exceeds the limit, the deeper the "stuttering" or "exhaustion" becomes.

[0134] The basic fluency value is calculated based on the statistical distribution characteristics of the ordered sequence. :

[0135] ;in The coefficient of variation represents the ordered sequence.

[0136] negative correction amount It is density and extraordinary proportion The function aims to quantify the negative impact of a target anomalous pattern on overall smoothness. The specific calculation method is as follows:

[0137] ;

[0138] and These are the preset weighting coefficients. They are used to adjust the density. and extraordinary proportion The contribution weight in the total penalty amount. These coefficients can be calibrated by analyzing the relationship between clinical data and expert scores, for example, by letting... , This indicates that long-interval anomalies have slightly higher weights.

[0139] Density of short-interval events The contribution function. Since the density itself is normalized, it can be used directly. Alternatively, it can be nonlinearly mapped to emphasize the effects of high density.

[0140] For example, ;

[0141] It is the abnormal proportion of long-interval events. The contribution function. It can be a linear or saturating function, for example:

[0142] , This is set as an upper limit to prevent excessive penalties and distortion caused by extremely long intervals in a single instance.

[0143] Suppose a pattern is identified, , ,but:

[0144] ;

[0145] set up It lasts for 2 seconds. The value is 1.5. If it is 4 seconds, then:

[0146] .

[0147] set up , , , .

[0148] Then negative correction amount for:

[0149] .

[0150] Finally, the smoothness correction factor is drawn. Calculate using the following formula:

[0151] ;

[0152] To ensure the validity of the results, constraints can be imposed on the calculation results, for example:

[0153] ; to prevent negative values ​​from appearing after correction.

[0154] Regarding the process of determining the threshold adjustment range in step A7, this application further discloses the following:

[0155] C1. Absolute quantification of the degree of behavioral abnormality -- raw duration deviation;

[0156] Calculate the absolute difference between the subject's drawing time and the closest boundary value of the preset reference time range.

[0157] For example, if the duration is too short, it is subtracted from the lower limit of the range; if the duration is too long, it is subtracted from the upper limit of the range; if the duration is within the range, the difference is zero. This absolute difference is called the original duration deviation. .

[0158] Original duration deviation It is the first quantification of the severity of deviations in the subject's behavioral efficiency from the normal range, and serves as the strength baseline for all subsequent modulation calculations. The larger the value, the more significant the behavioral abnormality. In principle, the greater the potential need or potential force for threshold adjustment should be.

[0159] C2. Preliminary quantification of spatial lateralization – basic ratio;

[0160] Calculate the ratio of the number of clock digits in the second set to the number in the first set, and use this ratio as the base ratio. :

[0161] , The ratio of the number of clock digits in the second set. The ratio of the number of clock digits in the first set; in an ideal uniform distribution, the fundamental ratio. Approaching 1. When unilateral space is neglected, It will be significantly greater than 1.

[0162] base ratio It is the most direct and stable mathematical representation of the severity of spatial layout eccentricity. It serves as the original input for amplitude factor calculation. However, alone... The value has limitations: it cannot distinguish the potential pathological differences between a painting with an off-center layout completed within a normal timeframe and a painting with the same off-center layout completed within an abnormal timeframe. Therefore, it needs to be compared with the original duration deviation that represents the degree of behavioral abnormality. Conduct a joint assessment.

[0163] C3. Determine the dynamic weighting coefficient based on the original duration deviation;

[0164] Through a function, the dynamic weight coefficients are adjusted. Deviation from original duration The change in the first extreme value With the second extreme value The changes between them are continuous and satisfy the following conditions: The larger, The closer to ; The smaller (closer to the normal range) The closer to First extreme value With the second extreme value All are preset values, and This can be achieved using a saturation function (such as the sigmoid function or the hyperbolic tangent function).

[0165] When behavioral abnormalities are very significant ( When the value is very high (e.g., a large ratio), it means that the temporal dimension has issued a high-risk signal. At this point, increasing the spatial features (baseline ratio) is crucial. The weight of ) in subsequent comprehensive decision-making (i.e. The closer to the larger Because the simultaneous occurrence of significant behavioral abnormalities and significant spatial abnormalities is a stronger indicator of cognitive impairment, the two should corroborate each other and reinforce each other.

[0166] When behavior is close to normal ( When the temporal dimension risk signal is very small, the decision weight of spatial features is weak. The closer to the smaller This prevents minor spatial imbalances from being over-interpreted in the absence of other supporting anomalies.

[0167] This mechanism ensures a strong response when there are dual anomalies in both spatiotemporal features, while remaining cautious when there is a single anomaly, thereby improving the specificity of the judgment.

[0168] C4. Generate a weighted spatial influence intensity index: amplitude factor;

[0169] Use dynamic weighting coefficients relative to the base ratio Perform a weighted transformation to generate the magnitude factor. :

[0170] ;

[0171] After this step, the original base ratio It is converted into an amplitude factor weighted and calibrated based on the degree of behavioral abnormality. .

[0172] when When it is very large, the larger It will be magnified ,make The significant increase reflects a high-risk situation in both time and space;

[0173] when When very small, smaller It will inhibit ,make The relative decrease reflects a situation that may be due to accidental spatial deviation.

[0174] C5. Calculate the final adjustment range using a preset amplitude calculation function;

[0175] In a monotonically increasing mapping relationship, with amplitude factor As input, the output dynamic gain coefficient is used. ;

[0176] The monotonically increasing mapping relationship embeds a preset amplitude threshold. :

[0177] when hour, ;

[0178] when hour, ;

[0179] Final adjustment range for:

[0180] .

[0181] Determine the amplitude factor With amplitude threshold The purpose of this relationship is to determine whether the intensity of spatial risk after behavioral weighting reaches the level of an independent alarm.

[0182] If achieved, then For the original duration deviation This amplifies the risk. This results in risk confirmation and enhancement: when the time sequence anomaly ( Large) and confirmed high-risk spaces ( When large (or large) values ​​are superimposed, the adjustment range is significantly amplified.

[0183] If not achieved, then ,right Maintain or suppress it to prevent over-regulation caused by a single timing anomaly.

[0184] Final adjustment range It is a fusion of the original behavioral anomaly intensity (original duration deviation) Spatial risk intensity (amplitude factor) after behavioral context weighting ), and the final gain (dynamic gain coefficient) based on whether the space risk holds true. The calculation results are used to ensure that the adjustment level precisely matches the overall risk level presented in this test.

[0185] A monotonically increasing mapping relationship can be achieved using piecewise linear functions:

[0186] ;

[0187] in, This is a preset gain adjustment coefficient greater than 0;

[0188] when hour, ;

[0189] when hour, ;

[0190] when hour, .

[0191] The above implementation clearly divides the mapping into two regions:

[0192] Inhibition region ( ) and magnification area ( ).

[0193] In the inhibition region, and Proportional and less than 1, resulting in the final adjustment range Linearly suppressed;

[0194] In the magnified area, The gain adjustment coefficient increases linearly from 1. The value is determined based on clinical experience.

[0195] This application further discloses the process for determining the dynamic weighting coefficients in step C3, including:

[0196] C31. Obtain the historical test dataset of the standard clock drawing test performed by a group of healthy subjects, and statistically analyze the distribution of drawing time for all healthy individuals based on the historical test dataset to form a reference distribution of drawing time.

[0197] Behavioral data from healthy populations provide an objective reference framework for assessing the behavior of individual subjects within a group context. The upper limit of the reference duration range is determined by this distribution. and lower limit It is an objective statistical limit derived from large sample data (e.g., the mean ± 1.96 standard deviations). This ensures that all subsequent comparisons and calculations based on this range have a solid epidemiological and statistical foundation, making the method reproducible and universal.

[0198] C32. Define the center point of weight change: centering deviation. ;

[0199] Centering deviation The calculation formula is:

[0200] ;

[0201] When the original duration deviation Equal to centering deviation This means that the individual's drawing time deviates exactly from the boundary of the normal range. Therefore, the centering bias... It is set as the zero or response center of the subsequent core mathematical function (hyperbolic tangent function). Population statistical characteristics are directly encoded into the dynamic weight calculation model, making the dynamic weight coefficients... The changes take the boundary of the normal range of the group as the natural inflection point, realizing the accurate connection between the group benchmark and individual data.

[0202] C33. Setting dynamic weight coefficients Theoretical boundary: First extreme value Second extreme value ;

[0203] The first extreme value Set as the theoretical upper limit of the base ratio under preset extreme skew conditions;

[0204] The second extreme value Set as the theoretical lower limit of the base ratio under unbiased conditions;

[0205] First extreme value This represents the most severe case of unilateral spatial neglect, such as almost all numbers on the left being ignored (assuming only one number is drawn), while the right side is filled with all numbers (11 numbers). The base ratio is then... The theoretical upper limit might be close to 11. Let's set this value to... This means dynamic weighting coefficients The achievable upper limit is related to the most severe degree of spatial defect that the assessment scale can reflect.

[0206] In an ideal, unbiased state, with 6 numbers on each side, the basic ratio is... The value is 1. Set this value to 1. This means dynamic weighting coefficients The lower limit is related to the perfect spatial equilibrium state.

[0207] By the first extreme value Second extreme value Ratio to base Theoretical extreme value binding, dynamic weight coefficient It is no longer an abstract mathematical variable; its range of variation... It has a clear clinical interpretation: it represents the entire theoretical pathological spectrum from perfect spatial equilibrium to extreme spatial lateralization. This makes the subsequently calculated weight values ​​directly interpretable.

[0208] C34. Achieving smooth and saturated dynamic mapping: a computational model based on the hyperbolic tangent function;

[0209] Dynamic weighting coefficients The calculation formula is:

[0210] ;

[0211] in, These are the preset steepness control parameters.

[0212] Let be the hyperbolic tangent function, a smooth, continuous, monotonically increasing odd function; its range is . And when the function variable is very large, it approaches (Saturation characteristics).

[0213] The smooth and continuous properties ensure the dynamic weighting coefficients. The changes are smooth, without abrupt changes, which is more in line with the biological fact that cognitive states change continuously.

[0214] The monotonically increasing characteristic satisfies The larger, The closer to The logic.

[0215] Saturation characteristics: when Significantly greater than hour, The term approaches 1, making Approaching ;when Significantly smaller than hour, The term approaches , making Approaching This achieves a design that approaches the extreme value, preventing the weights from increasing or decreasing indefinitely and enhancing the robustness of the system.

[0216] The formula can be understood as having two parts:

[0217] It is the center value of the weight (when hour, , (Equal to this central value).

[0218] It is a dynamically adjusted item, whose sign and size are determined by... The decision made Around the central value Variations within the range.

[0219] Steepness control parameters Its function is to control the function from Change to The rate. The larger the value, the steeper the curve, meaning... Slightly more , It will quickly move towards Increased sensitivity to behavioral abnormalities; The smaller the value, the flatter the curve, and the milder the response to behavioral abnormalities.

[0220] Regarding steepness control parameters The value of can be flexibly calibrated according to different clinical scenarios or assessment strategies, or it can be determined in the following ways:

[0221] Calculate the standard deviation of the drawing duration based on the historical test dataset from step C31. ; It quantifies the natural fluctuation range or dispersion of the time required for healthy individuals to complete the clock drawing test.

[0222] when A larger value indicates that the plotting time of healthy individuals already exhibits a wide range of normal fluctuations. In this case, a specific duration deviation in a single subject has relatively low statistical significance, as this deviation may simply fall within the broad normal fluctuation range of healthy individuals. Therefore, a more cautious and moderate response strategy should be adopted, namely, reducing sensitivity and avoiding overreacting to deviations that may be normal variations. This is achieved by setting a smaller threshold. Values ​​are used to achieve weights Follow The rate of change slows down, and the curve becomes flatter.

[0223] when When the range is small: This means that the plotting time of healthy individuals is highly concentrated, and the normal range is narrow. In this case, even a small deviation in plotting time may have high statistical significance and is more likely to exceed the normal range of individual differences. Therefore, a more sensitive and decisive response strategy should be adopted, namely, increasing sensitivity. This is achieved by setting a larger... Values ​​are used to achieve weights Able to It reacts quickly and significantly to changes, resulting in a steeper curve.

[0224] Based on the above analysis, a constant can be set. ,make The value can be:

[0225] ;constant This is a preset positive real number, for example, with a value of 2. Its core function is to... The overall magnitude of the value is calibrated and scaled to make the final calculated value... The value can fall within the reasonable numerical range required by the algorithm, thus ensuring that the hyperbolic tangent function has the desired dynamic range.

[0226] constant The value of is determined by one or a combination of the following methods:

[0227] Theoretical derivation based on expected response characteristics: For example, setting when Deviation A specific multiple (such as 1x) At that time, expectations When a term reaches a certain value (e.g., 0.76, corresponding to...), the value is determined by the number of items in the product. ), and thus inversely deduce value.

[0228] Clinical data-based calibration: Using independent datasets containing healthy controls and confirmed patients, the system optimizes classification performance (e.g., maximizing the area under the ROC curve, AUC) through grid search, gradient descent, and other optimization algorithms to determine the optimal overall discrimination performance. value.

[0229] This application further discloses the process for determining the lateral aggregation index, specifically including:

[0230] E1. Establish a spatial coordinate system and direction quantization benchmark;

[0231] First, determine the geometric center of the clock face area. Next, for each identified clock digit, calculate its geometric center (e.g., the centroid of the handwriting pixel set). Finally, for each digit, calculate the directional metric of its geometric center relative to the clock face center.

[0232] The center of the clock face is the absolute reference point for the entire spatial analysis. The positional information of all numbers is defined by their relative relationship to the center, which eliminates analytical bias caused by the different positions of the clock face in the image.

[0233] Orientation metrics, typically expressed in angles, quantify the absolute orientation of each number on a circle. For example, a number pointing to the "3 o'clock" direction might have an orientation metric of 73°. This provides more refined and continuous spatial location information than simply determining whether a number is on the left or right semicircle. It transforms the visual position of a number into a series of calculable, continuous angular data. This lays an indispensable data foundation for subsequent, more sophisticated spatial distribution statistics, moving beyond simple left-right partitioning.

[0234] E2. Calculate the degree of skewness in the quantity of space: base ratio;

[0235] Calculate the ratio of the number of clock digits in the second set to the number in the first set, and use this ratio as the base ratio. :

[0236] base ratio It is a core initial screening indicator for diagnosing unilateral neglect, with the advantages of simple calculation and clear meaning. However, this value only reflects the imbalance of numbers and cannot characterize how the numbers are distributed within their respective semicircles, so it is supplemented by subsequent steps.

[0237] E3. Evaluating the quality of spatial distribution: Calculation and optimization of dispersion parameters;

[0238] Preprocess all directional metrics within the left and right semicircle sets separately to suppress or eliminate outliers that significantly deviate from the main distribution. For example, a number might be drawn in an extremely far position due to rendering jitter.

[0239] For the cleaned set of orientation metrics, calculate the statistic of its dispersion, i.e., the dispersion parameter. The dispersion parameter is usually expressed as the angular standard deviation.

[0240] Hand-drawing tests may contain accidental, non-cognitive drawing errors (such as slips of the pen or shaky hands). These errors can severely interfere with the judgment of the true spatial distribution pattern. Outlier suppression can improve the signal-to-noise ratio of the data, ensuring that the dispersion parameters calculated subsequently reflect the subject's conscious spatial layout strategy, rather than accidental errors.

[0241] Dispersion parameters (such as angular standard deviation) quantify whether numbers are tightly clustered in a certain area within their respective semicircles or evenly distributed throughout the semicircle. A smaller dispersion means the numbers are clustered together; a larger dispersion means the numbers are scattered.

[0242] E4, Distribution correction factor for generating spatial distribution patterns;

[0243] Discreteness parameters based on the first set Discreteness parameter of the second set The difference between them generates a distribution correction factor. ;

[0244] The typical calculation method is the ratio of the two:

[0245] ;

[0246] In typical left-side spatial neglect, the numbers on the left (neglected side) are sparse and often placed arbitrarily, leading to... Larger (dispersed); crowded numbers on the right (the side of interest), leading to Smaller (clustered distribution). At this time... The value will be much greater than 1.

[0247] Under normal or uniform distribution conditions and similar, The value is close to 1.

[0248] Therefore, the distribution correction factor As an amplifier of anomalous spatial distribution patterns: when a typical pathological pattern of dispersion on one side and aggregation on the other side appears, The value will increase significantly; when the distribution patterns on both sides are similar, The value has little impact on the final result.

[0249] E5. Synthetic comprehensive pathological quantitative index: lateral aggregation index;

[0250] base ratio With distribution correction factor Multiplying them together yields the final lateral aggregation index. .

[0251] Lateral aggregation index It is a composite quantitative indicator that combines the dual characteristics of quantity imbalance and distribution anomaly.

[0252] when and At the same time, when the value is large (i.e., the number is severely lateralized and the distribution shows a typical pathological pattern), the lateral aggregation index is... It will achieve tremendous growth.

[0253] Conversely, if only the quantity is skewed ( Large, but evenly distributed. ), lateral aggregation index The increase is limited; if the distribution pattern is abnormal ( Large, but the numbers are roughly balanced. ), lateral aggregation index It won't be too high either.

[0254] This application further discloses the process for determining the direction measurement value in step E1:

[0255] E11. Establish a unified geometric reference system;

[0256] A Cartesian coordinate system is established with the center of the clock face area determined by the image recognition algorithm as the origin. Typically, the positive x-axis points to the right of the image, and the positive y-axis points to the bottom of the image (consistent with computer image coordinate system conventions).

[0257] The center of the circle serves as an absolute reference, unifying all spatial position calculations to the geometric center of the clock face itself, ensuring the inherent consistency of the analysis. Regardless of where the subject draws the clock on the paper, or what kind of rotational correction the image undergoes, as long as the center of the circle is accurately positioned, all subsequent angle calculations will be based on a stable and unchanging reference point.

[0258] E12, for the geometric center point of any clock digit. Its coordinates are , , representing 1 to 12 clock digits respectively.

[0259] This completes the mapping from image pixel space to abstract mathematical space, which facilitates subsequent quantitative spatial measurement.

[0260] E13. Use the four-quadrant arctangent function to calculate the accurate original azimuth angle;

[0261] For coordinates geometric center point Through formula Calculate its original orientation angle .function It is the arctangent function in the four quadrants.

[0262] Arctangent function in four quadrants Simultaneously receive and Two parameters, which can be determined based on coordinates The sign determines the quadrant in which the point lies, and returns a value within that quadrant. The radian angle value is then converted into an angle value within the range of -180° to 180°. This angle value is calculated with the origin as the vertex, from positive... Rotate the axis counterclockwise to point The angle formed by the ray.

[0263] E14. Generate standardized continuous angle values: circumferential standardization processing;

[0264] Through formula right Standardize it.

[0265] Indicates modulo operation, returns 0. Divide by The remainder.

[0266] Depend on Calculated The range is This means that it is located near the top of the clock face at an angle close to... (For example The point that is close to the angle. (For example The points are numerically very different. Although they are actually adjacent in circular space, if used directly... Calculating the dispersion of a set (such as the standard deviation of an angle) can introduce huge errors due to such artificial numerical jumps.

[0267] Translate and map all angles to Within a continuous interval.

[0268] Specifically, it adds all negative angles. This makes it a positive angle. For example, Become , Become .

[0269] After this processing, the angle value of all points located at the top of the clock face will be... Smooth and continuous within the interval (e.g., from) arrive Then ).

[0270] Standardized This creates a truly continuous circumferential angle variable. This makes subsequent calculations of statistics such as the angular standard deviation of the set mathematically valid and meaningful. The angular standard deviation algorithm can correctly handle the equivalence between 0° and 360°, thus accurately reflecting whether the numbers are clustered in a small sector (small standard deviation) or dispersed over a large range (large standard deviation).

[0271] This application further discloses the process of preprocessing (outlier suppression) all direction metrics within the left and right semicircle sets in step E3, specifically including:

[0272] E31. Determine the difference between the median angle and the circumferential angle;

[0273] For a given sequence of orientation metrics (either a first set or a second set), first calculate its median angle. Then, calculate the circumferential angle difference between each orientation metric in the sequence and that median angle.

[0274] Directional metrics are circular data, and the arithmetic mean is highly sensitive to extreme values ​​(i.e., potential outliers). An extreme value can significantly skew the mean, preventing it from representing the true center of the data. The median, on the other hand, is the value located in the middle after all angles have been sorted. Its mathematical properties make it naturally robust to extreme values. Even with a few significantly deviated angles, the median can stably reflect the central trend of most data, providing a reliable anchor point for subsequent judgments.

[0275] Because the angle data is circular ( and (coinciding), the circumferential characteristics must be considered when calculating the difference. For example, the median angle is... At a certain angle The difference between their straight lines is However, the difference in circumference should be This difference quantifies the degree to which each data point deviates from the center.

[0276] E32. Based on the calculated differences in all circumferential angles, further calculate the statistics of the angular deviation scale.

[0277] Angle deviation scales are robust estimates of the dispersion of angle values. A commonly used and preferred method is to calculate the median absolute deviation of these circumferential angle differences. Median absolute deviation again leverages the robustness of the median; it represents the typical distance a typical data point deviates from the median angle. Compared to the standard deviation, it is almost unaffected by outliers themselves and more accurately reflects the central tendency of the data subject (i.e., the numbers consciously drawn by the subjects). This scale is an objective measure of the normal range of fluctuation within the data itself.

[0278] E33. Determine the dynamic angle tolerance range based on the calculated angle deviation scale and the preset outlier determination coefficient. For example, the tolerance range can be set as the product of the angle deviation scale and the outlier determination coefficient.

[0279] The tolerance range is not a fixed number of degrees (e.g., fixed at 100 degrees). Instead of being a simple outlier, the tolerance range changes dynamically with the inherent dispersion (angular deviation scale) of the data. For a set whose distribution is inherently dispersed (e.g., the left-hand neglect group, where the numbers are already sparse), the angular deviation scale is larger, automatically assigning a wider tolerance range and avoiding the misclassification and removal of those sparse numbers that are part of the pathological pattern as outliers. Conversely, for a set whose distribution is very concentrated, the tolerance range is narrower, effectively identifying and removing genuine random errors.

[0280] The outlier determination coefficient is an adjustable amplification factor used to define how many times the typical dispersion is considered outlier. For example, a coefficient of 3 means that any point deviating from the median angle by more than 3 times the typical dispersion will be considered an outlier, drawing on the rule of thumb in statistics based on standard deviation.

[0281] E34. Mark the direction measurement values ​​in the sequence whose circumferential angle difference exceeds the dynamic angle tolerance range as outliers and remove them from the sequence.

[0282] Noisy points that significantly deviate from the main data pattern and are highly likely to be caused by drawing errors (such as shaky hands, typos, or brief distractions) are removed. The fundamental purpose is to facilitate the subsequent calculation of the standard deviation (the first dispersion parameter). Second dispersion parameter Provide clean data. The standard deviation is extremely sensitive to outliers; an outlier can significantly increase the standard deviation, thus incorrectly exaggerating the dispersion of that side of the distribution and ultimately distorting the distribution correction factor. The calculation was performed first, and robust outlier suppression was applied to ensure that the standard deviation of the subsequent calculation accurately and stably reflected the subject's conscious spatial layout strategy, rather than random error, thereby greatly improving the robustness and reliability of the entire quantitative assessment process.

[0283] Regarding the distribution correction factor mentioned in step E4 The calculation process is further disclosed in this application, specifically including:

[0284] E41. The standard deviation of the orientation metric values ​​of all clock digital identifiers in the first set and the second set after outlier suppression is used as the first dispersion parameter. Second dispersion parameter ;

[0285] The act of hand-drawing itself can introduce random errors, such as a pen slippage or a brief lapse in attention causing a number to be drawn significantly off-target. Such isolated outliers can severely distort judgments of the overall distribution pattern. Suppression techniques (e.g., using truncated means or robust estimates based on the median) can filter out this non-cognitive "noise," ensuring that subsequent analyses are based on the subject's stable, conscious spatial planning.

[0286] After data cleaning, calculating the standard deviation of angle values ​​within the set is the most classic and effective measure of data dispersion. A smaller standard deviation (e.g., ...) Small means that the numbers on that side are tightly clustered within a smaller angular range; a larger standard deviation (such as...) .... A "large" (or "larger") indicates that the numbers on that side are distributed over a wider angular range. This quantitative description goes beyond the binary judgment of whether the numbers are all on that side and enters into a quality assessment of how the numbers are arranged on that side.

[0287] E43, the first dispersion parameter With the second dispersion parameter The ratio of the distribution correction factor .

[0288] This calculation method directly addresses the typical behavioral pattern of unilateral spatial neglect. In this pathological state, patients often neglect one side of the space (such as the left side), resulting in sparse numbers on that side that are often randomly and scattered. Large); at the same time, focusing all attention on the opposite side (such as the right side) leads to numbers being overcrowded and overlapping. (Small). Therefore, the ratio It will result in a value much greater than 1.

[0289] Conversely, in cases of poor health or nonspecific plotting, even if the number of numbers is slightly skewed, the dispersion of the distribution on both sides is usually quite similar, making... It is close to 1. Therefore, As a correction factor, its core function is selective amplification. It can strongly amplify spatial patterns that are highly pathologically specific, such as those that are dispersed on one side and clustered on the other, while responding very weakly to patterns that are uniform overall but slightly eccentric or drawn in a scattered manner.

[0290] like Figure 3 As shown, this application also discloses: an assisted Alzheimer's disease assessment system based on the clock drawing test, comprising:

[0291] The data acquisition module, as the input front end of the system, synchronously collects test images, drawing duration, and orientation information of drawing posture, providing multi-dimensional, time-aligned raw data for evaluation.

[0292] The spatial reference correction module uses orientation information to determine the subject's subjective spatial orientation and performs rotation correction on the image to eliminate the bias in the analysis benchmark caused by posture or paper tilt, thus ensuring the fairness of spatial analysis.

[0293] The image recognition module automatically identifies and locates the clock face outline, center, and all numerals from the corrected image, converting the image into structured data that can be computed.

[0294] The spatial partitioning module divides the clock face into two neuropsychologically significant regions, left and right, based on the vertical central axis passing through the center of the circle, and assigns the numbers to the corresponding sets, thus establishing a framework for subsequent comparative analysis.

[0295] The lateral assessment module calculates the lateral clustering index, which characterizes the degree of spatial layout anomaly, by analyzing the quantitative ratio of numbers in two sets.

[0296] The adjustment direction determination module determines the basic direction of threshold adjustment (tightening, loosening, or maintaining) by comparing the drawing time with the normal range, and introduces the behavioral efficiency dimension.

[0297] The adjustment amplitude calculation module calculates the specific adjustment intensity by combining the degree of time deviation with the relationship between numerical quantities, thereby realizing the linkage risk assessment of behavior and spatial characteristics.

[0298] The dynamic threshold generation module dynamically processes the fixed base threshold based on the aforementioned direction and amplitude, and outputs a dynamic judgment threshold that adapts to the specific context of this test.

[0299] The evaluation output module generates and outputs the final auxiliary evaluation signal by comparing the lateral aggregation index with the dynamic threshold, thus completing the closed loop from data to conclusion.

[0300] The above modules form a complete processing chain, realizing the entire process from multimodal data acquisition, intelligent correction, feature extraction, context-aware dynamic decision-making to personalized result output. The core of this system lies in upgrading static image analysis to intelligent assessment that integrates behavioral context through a dynamic threshold generation mechanism, significantly improving the objectivity, accuracy, and practicality of assisting in the identification of cognitive impairment in real clinical scenarios.

[0301] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for assisting Alzheimer's disease assessment based on the clock drawing test, characterized by: Includes the following steps: Acquire the test image drawn by the subject during the clock drawing test, and simultaneously acquire the drawing time of the test image and the orientation information representing the subject's drawing posture. Based on the orientation information, the spatial reference direction of the subject is determined, and the test image is rotated and corrected according to the spatial reference direction to obtain a corrected test image. Identify the clock face area and clock numerals from the calibration test image; The clock face area is divided into sections by the vertical central axis passing through the center of the circle in the clock face area, and a first set and a second set are formed based on the clock digits in each section. Based on the proportional relationship between the number of clock digits in the first set and the second set, a lateral clustering index is generated to characterize the degree of spatial abnormality in the spatial distribution of the clock digits drawn by the subjects. The drawing duration is compared with a preset reference duration range, and the threshold adjustment direction is determined based on the comparison result; Based on the difference between the drawing duration and the boundary value of the reference duration range, and combined with the quantitative relationship of clock digit identifiers in the first set and the second set, the threshold adjustment range is calculated; Based on the threshold adjustment direction and the threshold adjustment amplitude, the preset basic threshold is processed to obtain a dynamic threshold; The lateral aggregation index is compared with the dynamic threshold, and an evaluation signal is output based on the comparison result.

2. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 1, characterized in that: The process of determining the threshold adjustment direction includes: Obtain the temporal coordinate sequence of the handwriting drawn in the test image as temporal stroke sequence data; Based on the time-series pen sequence data, identify continuous pause events at the pen stroke generation points during the drawing process, and extract the interval duration of each pause event; Based on the distribution of the interval duration, a rendering smoothness correction factor is calculated to characterize the stability of the cognitive load during the rendering process; Based on the comparison results between the drawing duration and the preset reference duration range, and in conjunction with the drawing smoothness correction factor, the threshold adjustment direction is determined through predefined decision logic; The decision-making logic includes: If the drawing duration falls within the reference duration range, the threshold adjustment direction is set to zero adjustment. If the drawing time is shorter than the lower limit of the reference time range, then the first sub-logic is executed: The rendering smoothness correction factor is compared with the first smoothness threshold. If the rendering smoothness correction factor is higher than the first smoothness threshold, the threshold adjustment direction is determined to be positive adjustment; otherwise, the threshold adjustment direction is determined to be zero adjustment. If the drawing duration is longer than the upper limit of the reference duration range, then the second sub-logic is executed: The rendering smoothness correction factor is compared with the second smoothness threshold. If the rendering smoothness correction factor is lower than the second smoothness threshold, the threshold adjustment direction is determined to be negative adjustment; otherwise, the threshold adjustment direction is determined to be zero adjustment.

3. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 2, characterized in that: The calculation process for the rendering smoothness correction factor includes: Arrange the intervals of all pause events in order of occurrence to form an ordered sequence, and calculate the average interval of the ordered sequence; Calculate the statistical distribution characteristics of the ordered sequence to obtain a basic fluency value, wherein the basic fluency value is negatively correlated with the degree of dispersion of the interval duration; Perform a pattern scan on the ordered sequence to identify whether there is a target abnormal pattern; The rule for determining the target anomaly pattern is: in continuous In each pause event, there are at least The interval time is less than the preset proportional coefficient. Short-interval events that are the product of the average interval duration and the average interval duration, and at least The next event after a short interval event is when the interval duration is greater than a preset proportional coefficient. Long-interval events are the product of the average interval duration and the long interval duration; where, , It is an integer greater than 1. , Let be a pre-defined positive real number, and , ; If the target abnormal pattern is identified, a negative correction amount is calculated based on the density of the short-interval events and the duration of the long-interval events, and the result of subtracting the negative correction amount from the base smoothness value is used as the rendering smoothness correction factor. If the target abnormal pattern is not identified, the basic smoothness value will be directly used as the rendering smoothness correction factor.

4. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 1, characterized in that: The process of determining the threshold adjustment range is as follows: Calculate the absolute difference between the drawing duration and the nearest boundary value of the reference duration range, and use it as the original duration deviation. Calculate the ratio of the number of clock digit identifiers in the second set to the number in the first set, and use this ratio as the base ratio; The dynamic weighting coefficient is determined based on the original duration deviation; wherein, the larger the original duration deviation, the closer the dynamic weighting coefficient is to the first extreme value, and the smaller the original duration deviation, the closer the dynamic weighting coefficient is to the second extreme value. Based on the dynamic weighting coefficients, the base ratios are weighted and transformed to generate an amplitude factor that characterizes the intensity of the influence of the digital distribution. The original duration deviation and the amplitude factor are input into a preset amplitude calculation function to calculate the threshold adjustment amplitude; The amplitude calculation function is configured to: use the amplitude factor as an input variable and calculate a dynamic gain coefficient through a predefined monotonically increasing mapping relationship; The threshold adjustment range is obtained by multiplying the dynamic gain coefficient by the original duration deviation. The monotonically increasing mapping relationship is configured such that when the amplitude factor increases, the calculated dynamic gain coefficient also increases, and when the amplitude factor exceeds a preset amplitude threshold, the dynamic gain coefficient is greater than 1; when the amplitude factor is lower than the amplitude threshold, the dynamic gain coefficient is less than or equal to 1.

5. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 4, characterized in that: The process for determining the dynamic weighting coefficients is as follows: Obtain historical test datasets of clock drawing tests conducted by a group of healthy subjects, and statistically analyze the reference distribution of drawing durations; Based on the aforementioned drawing duration reference distribution, determine the upper limit of the reference duration range. and lower limit And determine the centering deviation. : ； The theoretical upper limit of the basic ratio under preset extreme skew conditions is set as the first extreme value. The theoretical lower limit of the basic ratio under unbiased conditions is set as the second extreme value. ; Based on the original duration deviation Calculate dynamic weight coefficients : ； in, These are the preset steepness control parameters.

6. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 1, characterized in that: The process for determining the lateral aggregation index is as follows: The geometric center of each clock digit is determined from the calibration test image, and the orientation metric of each clock digit is obtained based on the orientation of each geometric center relative to the center of the clock face area. Calculate the ratio of the number of clock digit identifiers in the second set to the number in the first set, and use this ratio as the base ratio; Outlier suppression is performed on the orientation metrics in the first set and the second set respectively, and the dispersion parameter of the processed data is calculated. A distribution correction factor is generated based on the difference in the dispersion parameters between the first set and the second set; The lateral clustering index is obtained by multiplying the base ratio by the distribution correction factor.

7. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 6, characterized in that: The process of obtaining the direction metric value includes: Establish a Cartesian coordinate system with the center of the clock face area as the origin; For any clock digital mark's geometric center point Its coordinates are , ; Calculate the point using the first mathematical relationship. original orientation angle : ; where, function It is the arctangent function in the four quadrants; The original direction angle is determined by the second mathematical relationship. The direction metric value is obtained by standardization. : ; where, function For modulo operation, the direction metric value is... The value is located at Within a continuous angular range.

8. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 7, characterized in that: The distribution correction factor The calculation process is as follows: For the first set, after outlier suppression, the standard deviation of the orientation metric values ​​of all clock digital identifiers within the set is calculated and used as the first dispersion parameter. ; For the second set, after outlier suppression, the standard deviation of the orientation metric values ​​of all clock digital identifiers within the set is calculated and used as the second dispersion parameter. ; The first dispersion parameter With the second dispersion parameter The ratio of the distribution correction factor .

9. The method for assisting Alzheimer's disease assessment based on the clock drawing test according to claim 7, characterized in that: The outlier suppression process is as follows: For the direction measurement value sequence in the first set or the second set, calculate the median angle of the direction measurement value sequence and the circumferential angle difference between each angle and the median angle; Based on the circumferential angle difference, calculate the angular deviation scale of the direction metric value sequence; The dynamic angle tolerance range is determined based on the angle deviation scale and the preset outlier determination coefficient. In the sequence of orientation metrics, those whose circumferential angle difference exceeds the dynamic angle tolerance range are marked as outliers. Remove the orientation metric values ​​corresponding to outliers from the orientation metric value sequence.

10. An Alzheimer's disease assessment system based on the clock drawing test, characterized in that: include: The data acquisition module is used to acquire the test image drawn by the subject in the clock drawing test, and simultaneously acquire the drawing time of the test image and the orientation information representing the subject's drawing posture. The spatial reference correction module is used to determine the spatial reference direction of the subject based on the orientation information, and to rotate and correct the test image according to the spatial reference direction to obtain a corrected test image. The image recognition module is used to identify the clock face area and clock digit markings from the calibration test image; The spatial partitioning module is used to partition the clock face area with the vertical central axis of the center of the clock face area as the boundary, and to form a first set and a second set based on the clock digits in each partition. The lateralization assessment module is used to generate a lateralization clustering index to characterize the degree of spatial abnormality in the clock digits drawn by the subject, based on the proportional relationship between the number of clock digit identifiers in the first set and the second set. The adjustment direction determination module is used to compare the drawing duration with a preset reference duration range and determine the threshold adjustment direction based on the comparison result; The adjustment range calculation module is used to calculate the threshold adjustment range based on the difference between the drawing duration and the boundary value of the reference duration range, and in combination with the quantitative relationship of clock digit identifiers in the first set and the second set; The dynamic threshold generation module is used to process the preset basic threshold according to the threshold adjustment direction and the threshold adjustment amplitude to obtain the dynamic threshold; The evaluation output module is used to compare the lateral aggregation index with the dynamic threshold and output an evaluation signal based on the comparison result.

Images