Stroke patient NIHSS score automatic evaluation system

Abstract

The application relates to the technical field of intelligent medical auxiliary diagnosis and automatic evaluation of nerve functions, in particular to an NIHSS score automatic evaluation system for stroke patients. The system comprises a multi-modal induction and perception unit, which is used as a front-end data inlet and collects patient responses in real time to generate a video stream containing depth and color information and a synchronous audio stream; a dynamic reference calibration unit, which extracts kinematic characteristics to construct an individualized neural reference template; a neural motion frequency spectrum decomposition unit, which generates a frequency spectrum pathological characteristic vector distinguishing muscle weakness and ataxia; a contralateral mirror inhibition analysis unit, which generates compensation and driving confidence representing real neural driving intentions; a cross-modal reflection analysis unit, which generates a sensory pathway integrity index according to the difference in nerve conduction velocity; and a collaborative scoring decision engine, which maps multi-modal characteristics to a standardized NIHSS score. The application effectively eliminates the interference of age and basic physical fitness on the score, and can also provide high-precision comparison references for subsequent abnormality determination of the affected side.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical auxiliary diagnosis and automatic evaluation of neural function, in particular to an NIHSS score automatic evaluation system for stroke patients. BACKGROUND

[0002] With the increasing demand for precise medical treatment of stroke, objective quantitative evaluation of the neural function of patients is particularly important; the NIHSS score is the gold standard for measuring the severity and prognosis of stroke, and its accuracy directly determines the effectiveness of the rehabilitation program;

[0003] At present, clinical bedside examinations are generally performed by medical staff, and qualitative scoring is completed by observing limb movement with the naked eye, performing pain tests by acupuncture, etc.; however, the traditional manual evaluation method mainly relies on the subjective experience of doctors, and it is difficult to maintain consistency among different observers, and it is impossible to capture the subtle tremor characteristics invisible to the human eye to distinguish muscle weakness from ataxia; although existing computer vision auxiliary technologies introduce image analysis, in the actual ward environment, they will fail once the limbs are covered by bedding; in addition, simple visual monitoring cannot distinguish between real neural drive of the affected limb and compensatory movement driven by the healthy limb, and the general medical statistical threshold ignores the individual differences of patients, which easily misjudges the physiological slowness of the elderly as pathological changes; contact-type sensory function tests also have the disadvantages of iatrogenic infection risk and evoking nervousness in patients;

[0004] Therefore, how to realize a non-contact, anti-shielding interference, and accurate identification of compensatory behavior and individual differences of stroke neural function automatic quantitative evaluation has become a problem to be solved in the field.

[0005] The above information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] To solve the above technical problems, the present application discloses an NIHSS score automatic evaluation system for stroke patients, in particular, the technical scheme of the present application is:

[0007] The multi-modal induction and perception unit, as a front-end data entry, is configured to issue synchronous instructions in a preset time sequence through a composite instruction generator, establish a stimulation-feedback closed loop, and real-time collect patient responses to generate video streams and synchronous audio streams containing depth and color information;

[0008] The dynamic benchmark calibration unit, as the cornerstone of pathological judgment, is configured to receive video streams, guide the patient's healthy limbs to complete calibration actions, extract kinematic features to construct an individualized neural benchmark template, thereby defining a dynamic reference system for subsequent abnormality judgment;

[0009] The neuromotor spectrum decomposition unit is configured to receive the video stream of the affected limb and the individualized neural reference template, convert the spatial displacement into a frequency domain signal, and analyze the low-frequency components and specific frequency band peaks based on the threshold in the template to generate a spectrum pathological feature vector that distinguishes between myasthenia and ataxia.

[0010] The contralateral mirror inhibition analysis unit, as a correction module to overcome occlusion and compensation, is configured to lock the healthy limb area when a movement command is issued to the affected side, detect the antagonistic movement of the healthy side and compare it with the resting noise threshold in the template, and generate compensation and drive confidence that characterize the real neural drive intention.

[0011] The cross-modal reflex analysis unit is configured to emit sudden audio-visual stimuli, capture facial micro-expressions and calculate latency, and generate sensory pathway integrity indices based on differences in nerve conduction velocity.

[0012] The collaborative scoring decision engine, acting as the coupling control center, is configured to receive feature vectors, confidence scores, and integrity indices. It uses a hierarchical weighting strategy combined with compensatory correction logic to map multimodal features into standardized NIHSS scores.

[0013] Preferably, the composite instruction generator in the multimodal induction and perception unit is configured to synchronously issue voice instructions and visual induction signals such as flashing red dots on the screen according to a preset timing sequence.

[0014] The multimodal induction and perception unit is further configured to directly distribute and transmit the acquired high-dimensional video stream as the raw signal to be processed to the dynamic reference calibration unit and the neural motion spectrum decomposition unit.

[0015] Preferably, the process of constructing an individualized neural benchmark template by the dynamic benchmark calibration unit includes:

[0016] Extract the maximum velocity threshold of the healthy limb movement as a reference upper limit for assessing muscle strength;

[0017] The motion smoothness coefficient calculated based on the principle of minimizing jerk is extracted as a quantitative indicator reflecting the fineness of the nervous system's control.

[0018] Extract the basic spectral distribution of the healthy limb as a reference for the background noise level;

[0019] The maximum velocity threshold, motion smoothness coefficient, and basic spectral distribution are packaged to generate the individualized neural baseline template, which is then sent to the neural motion spectrum decomposition unit and the contralateral mirror inhibition analysis unit as the zero point for judging abnormalities.

[0020] Preferably, the process of generating spectral pathological feature vectors by the neuromotor spectrum decomposition unit includes a muscle weakness determination step:

[0021] Extracting the energy amplitude of the principal motion component below 1Hz from the frequency domain signal;

[0022] The energy amplitude is compared with the maximum velocity threshold from the individualized neural baseline template;

[0023] When the energy amplitude is significantly lower than a preset ratio of the maximum velocity threshold, the myasthenia gravis feature is marked in the spectral pathological feature vector.

[0024] Preferably, the process of generating spectral pathological feature vectors by the neuromotor spectrum decomposition unit further includes an ataxia determination step:

[0025] Search for energy peaks in the intentional flutter frequency band from 3 Hz to 5 Hz;

[0026] Calculate the ratio of the spectral power density of the energy peak to the average level of background noise determined by the individualized neural benchmark template;

[0027] If the ratio exceeds a preset multiple, then ataxia features are marked in the spectral pathological feature vector.

[0028] Preferably, the process by which the contralateral mirror suppression analysis unit generates compensation and driving confidence includes:

[0029] Calculate the downward pressure of the unaffected limb or the tilt angle of the trunk;

[0030] Determine whether involuntary antagonistic movements are detected that are opposite in direction to the commanded movement and whose amplitude is positively correlated with the expected amplitude of movement on the affected side;

[0031] If the antagonistic movement is detected and its intensity exceeds the resting noise threshold set by the individualized neural baseline template, it is determined that there is a genuine neural driving intention.

[0032] Generate the compensation and driving confidence scores with values ​​ranging from 0 to 1, where higher values ​​represent less compensation components.

[0033] Preferably, the process by which the cross-modal reflex analysis unit generates a sensory pathway integrity index includes:

[0034] Time series analysis was performed on the facial motion units of the eyebrow lowering motion unit and the cheek lifting motion unit;

[0035] The time difference from the moment the stimulus is emitted to the moment when the facial muscles produce a detectable displacement is extracted as the latency.

[0036] The latency period is compared with the preset cortical response threshold and spinal reflex threshold: if the latency period is greater than the cortical response threshold, the corresponding score of 0 is output; if the latency period is between the two, the corresponding score of 1 is output; if the latency period is less than or equal to the spinal reflex threshold or there is no response, the corresponding score of 2 is output.

[0037] Preferably, the collaborative scoring decision engine performs a preliminary screening step of basic motion scoring, specifically including:

[0038] Receive the spectral energy amplitude from the neural motor spectrum decomposition unit;

[0039] Using a pre-defined energy-fraction mapping table, the motor energy of the affected limb is mapped to a basic motor fraction ranging from 0 to 4;

[0040] The mapping table is logically set as follows: if the energy amplitude is greater than or equal to the normal threshold, the basic motion score is 0; if the energy amplitude is close to 0, the basic motion score is 4.

[0041] Preferably, the collaborative scoring decision engine performs a compensatory confidence correction step, specifically including:

[0042] The compensation and driving confidence from the contralateral mirror suppression analysis unit are introduced;

[0043] The corrected motion score is calculated using a logical formula configured as follows: the base motion score is added to the correction term and then rounded down.

[0044] The calculation logic of the correction term is as follows: multiply the preset penalty coefficient by the difference between the compensation and the driving confidence obtained by subtracting the value 1 from the value 1.

[0045] Therefore, when high-intensity compensatory movements are detected on the healthy side, resulting in low confidence, the final score is forced to shift towards a value representing a more severe level of paralysis, in order to eliminate false rehabilitation score bubbles.

[0046] Preferably, the collaborative scoring decision engine is further configured as follows:

[0047] The corrected motion score is weighted and summarized with the sensory pathway integrity index from the cross-modal reflex analysis unit to generate a structured score matrix;

[0048] The output is a standardized NIHSS electronic scoring report containing the specific scores for each sub-item and the total score. The report also includes a compensation confidence warning for items whose scores have changed due to compensation adjustments.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. This invention innovatively utilizes the patient's unaffected limb as a self-reference frame through a dynamic benchmark calibration unit, solving the problem that traditional assessment standards are difficult to adapt to individual differences. The system guides the unaffected side to complete calibration actions, extracting key kinematic parameters such as maximum velocity threshold, motion smoothness, and baseline spectral distribution to construct an individualized neural benchmark template. This design not only establishes a zero point that conforms to the patient's current physiological state, effectively eliminating the interference of age and basic physical condition on the score, but also provides a high-precision comparative reference for subsequent abnormal judgment of the affected side, significantly improving the adaptability and accuracy of the NIHSS score in different patient groups.

[0051] 2. This invention utilizes a neuromotor spectrum decomposition unit, overcoming the limitations of traditional visual observation which can only provide qualitative assessments, and achieving quantitative differentiation between muscle weakness and ataxia. By converting limb spatial displacement into frequency domain signals, the system can accurately capture the energy difference between low-frequency components and the intention tremor frequency band. This technical feature enables the system to penetrate the surface motor impairment and deeply analyze the underlying neural conduction mechanism, generating highly discriminative spectral pathological feature vectors, thereby helping doctors more accurately determine the specific type of damage to the motor control center caused by stroke.

[0052] 3. This invention introduces a linkage mechanism between the contralateral mirror inhibition analysis unit and the collaborative scoring decision engine, effectively solving the common problem of compensatory motor interference in stroke assessment. When issuing motor commands to the affected side, the system locks and monitors the antagonistic movements of the healthy side in real time. By calculating the confidence of compensation and drive, it quantitatively assesses whether the patient is using the healthy side to cover up paralysis on the affected side. Combined with the scoring correction logic of the penalty coefficient, the system can forcibly eliminate false rehabilitation score bubbles caused by compensation, ensuring that the final NIHSS score truly reflects the actual driving ability of the patient's damaged neural pathways.

[0053] 4. This invention achieves a leap from single-motor assessment to holistic sensory-motor circuit assessment by combining a cross-modal reflex analysis unit with multimodal induction technology. The system not only establishes a stimulus-feedback closed loop through sound-light composite commands, but also uses facial micro-expression analysis technology to capture millisecond-level latency differences, thereby quantifying the integrity of sensory pathways. This comprehensive scoring strategy, which combines cortical reaction speed, spinal cord reflex threshold, and limb motor energy, generates a structured scoring matrix, which can provide clinicians with a comprehensive medical report covering motor function, sensory conduction, and neural reaction speed. Attached Figure Description

[0054] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0055] Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0057] Example 1:

[0058] Please see Figure 1 An automated assessment system for NIHSS scores in stroke patients, including:

[0059] The multimodal induction and perception unit, as the front-end data entry point, is configured to issue synchronous commands in a preset sequence through a composite command generator, establish a stimulus-feedback closed loop, and collect patient responses in real time to generate video streams and synchronous audio streams containing depth and color information.

[0060] The dynamic benchmark calibration unit, as the cornerstone of pathological judgment, is configured to receive video streams, guide the patient's healthy limbs to complete calibration actions, extract kinematic features to construct an individualized neural benchmark template, thereby defining a dynamic reference system for subsequent abnormality judgment;

[0061] The neuromotor spectrum decomposition unit is configured to receive the video stream of the affected limb and the individualized neural reference template, convert the spatial displacement into a frequency domain signal, and analyze the low-frequency components and specific frequency band peaks based on the threshold in the template to generate a spectrum pathological feature vector that distinguishes between myasthenia and ataxia.

[0062] The contralateral mirror inhibition analysis unit, as a correction module to overcome occlusion and compensation, is configured to lock the healthy limb area when a movement command is issued to the affected side, detect the antagonistic movement of the healthy side and compare it with the resting noise threshold in the template, and generate compensation and drive confidence that characterize the real neural drive intention.

[0063] The cross-modal reflex analysis unit is configured to emit sudden audio-visual stimuli, capture facial micro-expressions and calculate latency, and generate sensory pathway integrity indices based on differences in nerve conduction velocity.

[0064] The collaborative scoring decision engine, acting as the coupling control center, is configured to receive feature vectors, confidence scores, and integrity indices. It uses a hierarchical weighting strategy combined with compensatory correction logic to map multimodal features into standardized NIHSS scores.

[0065] In this embodiment, the NIHSS score automated assessment system for stroke patients is constructed as an integrated, non-contact neurofunctional quantitative analysis platform. When the system is started, the multimodal induction and perception unit assumes the function of the human-computer interaction interface and is responsible for constructing a precise time-synchronized stimulation and feedback acquisition loop. The acquired raw data stream is then imported into the dynamic benchmark calibration unit, which establishes benchmark data based on the patient's own unaffected limb function and generates an individualized neurological benchmark template.

[0066] The individualized neural baseline template is a dataset containing the kinematic limit parameters of a specific patient. Its core function is to provide a dynamically adjusted comparison zero point for subsequent pathological analysis, thereby avoiding the risk of misjudgment caused by using the average value of a general population. On this basis, the neuromotor spectrum decomposition unit performs frequency domain analysis on the weak movements of the affected limb to generate a spectrum pathological feature vector. This vector clearly distinguishes between insufficient neural drive and coordination disorder through quantitative means.

[0067] In the parallel processing mechanism, the contralateral mirror inhibition analysis unit monitors involuntary coordinated movements of the healthy limb and calculates compensation and drive confidence. This parameter is specifically used to infer whether the patient has a genuine motor intention when the affected limb is occluded or inactive, based on the principle of cross-reflexes in the nervous system. For the assessment of sensory function, the cross-modal reflex analysis unit generates a sensory pathway integrity index by capturing the instantaneous response of facial micromuscles to sudden stimuli. All generated independent feature data are aggregated into the collaborative scoring decision engine, which performs logical weighting and correction operations on the multi-source data and outputs standardized scoring results.

[0068] By implementing the above system architecture, a shift from qualitative assessment relying on subjective observation to quantitative assessment based on data-driven methods has been achieved. The combination of the contralateral mirror inhibition analysis unit and the collaborative scoring decision engine has overcome the technical bottleneck of existing computer vision technology failing in the environment of bedding occlusion, and eliminated camouflaged compensatory actions through the neural reflex mechanism, thereby improving the robustness and accuracy of stroke assessment in complex ward environments.

[0069] Example 2:

[0070] The composite instruction generator in the multimodal induction and perception unit is configured to synchronously issue voice instructions and visual induction signals such as flashing red dots on the screen according to a preset timing sequence.

[0071] The multimodal induction and perception unit is also configured to directly distribute and transmit the acquired high-dimensional video stream as the raw signal to be processed to the dynamic reference calibration unit and the neuromotor spectrum decomposition unit.

[0072] In this embodiment, the design focus of the multimodal induction and perception unit is to ensure strict spatiotemporal alignment between command output and data input; the composite command generator integrated inside the unit adopts a dual-channel sensory stimulation strategy to address the cognitive or attentional impairments that may accompany stroke patients.

[0073] In terms of execution logic, the generator synchronously triggers clear voice commands and presents high-contrast visual guidance signals at specific coordinate positions on the display terminal based on millisecond-level timestamps; this unit incorporates a lightweight human pose estimation model, and this embodiment preferably adopts a model based on... backbone network Architecture or Framework to ensure real-time inference speed on mobile devices is no less than The configuration is to perform convolution operations on each frame of the video stream to extract the two-dimensional pixel coordinates of key points, including the shoulder, elbow, wrist, and finger joints.

[0074] Combining depth and distance information, a coordinate transformation algorithm is used to map two-dimensional pixel coordinates to three-dimensional spatial coordinates, thereby constructing a time-varying sequence of three-dimensional skeletal keypoints. To ensure the fidelity of subsequent signal processing, this unit does not perform lossy compression, but instead distributes the original high-dimensional video stream containing the aforementioned three-dimensional skeletal keypoint sequence data in parallel to the dynamic benchmark calibration unit and the neuromotor spectrum decomposition unit via a high-speed data bus. To ensure that the system can capture millisecond-level facial micro-expression latency and high-frequency tremor features, the video acquisition module in the multimodal induction and perception unit preferably uses a frame rate of not less than [a certain value]. A high-speed industrial camera, with a shutter speed set faster than... To prevent motion blur from affecting feature extraction accuracy;

[0075] By executing the processing logic of the multimodal induction and sensing unit, the system ensures zero-delay correspondence between stimulus emission and motion capture on the time axis, solving the technical problem of misdiagnosis as neurological deficit due to delayed patient response caused by unclear instruction transmission, and providing a pure and synchronous raw data source for subsequent high-precision pathological analysis.

[0076] Example 3:

[0077] The process of constructing an individualized neural benchmark template using a dynamic benchmark calibration unit includes:

[0078] Extract the maximum velocity threshold of the healthy limb movement as a reference upper limit for assessing muscle strength;

[0079] The motion smoothness coefficient calculated based on the principle of minimizing jerk is extracted as a quantitative indicator reflecting the fineness of the nervous system's control.

[0080] Extract the basic spectral distribution of the healthy limb as a reference for the background noise level;

[0081] The maximum velocity threshold, motion smoothness coefficient, and basic spectral distribution are packaged to generate an individualized neural baseline template, which is then sent to the neural motion spectrum decomposition unit and the contralateral mirror inhibition analysis unit as the zero point for judging abnormalities.

[0082] In this embodiment, the dynamic baseline calibration unit performs the task of establishing a patient-specific baseline, aiming to eliminate the influence of individual differences on universal scoring standards. During the system initialization phase, this unit guides the patient to complete a set of standardized linear reciprocating movements using their asymptomatic unaffected limb. The system extracts and calculates the following core parameters from the collected trajectory data:

[0083] The system calculates the peak instantaneous velocity of the unaffected limb throughout the entire movement cycle and sets it as the maximum velocity threshold, which represents the theoretical upper limit of muscle explosive power under the patient's current physiological state.

[0084] The system calculates the motion smoothness coefficient using the principle of minimizing jerk. It performs a third derivative of the displacement signal to obtain the jerk curve and calculates the integral value of the area covered by the curve. The system then uses inverse proportional mapping logic to calculate the smoothness coefficient, specifically by dividing a preset normalization constant by the sum of this integral value and a minimum regularization smoothing factor. Values This is to prevent computational overflow caused by the integral value being zero due to complete stillness.

[0085] The upper and lower limits of normalization are taken from a preset statistical database of healthy individuals; in one specific implementation, the normalization constant is set to... This value is based on pre-collected data. For example, the average value of the integral of acceleration during standard movements in healthy adults is used to determine the outcome; if an external database cannot be connected, the system defaults to using the average smoothness integral of the first three successfully calibrated movements on the unaffected side. The system uses a multiple as a normalization benchmark; it performs a fast Fourier transform on the small displacements of the unaffected limbs during the static holding phase to extract their basic spectral distribution, thereby determining the patient's unique physiological tremor background noise level.

[0086] This unit encapsulates the maximum velocity threshold, motion smoothness coefficient, and basic spectral distribution data to generate an individualized neural baseline template, which is then passed as a dynamic parameter to subsequent analysis units.

[0087] By implementing the construction process of dynamic benchmark calibration units, an innovative self-control mechanism is introduced. Compared with the use of fixed medical statistical thresholds, this method can automatically adapt to the benchmark level of the elderly or frail, avoiding misdiagnosis of muscle weakness caused by stroke due to the patient's own slow movement, and improving the specificity of the assessment.

[0088] Example 4:

[0089] The process of generating spectral pathological feature vectors from the neuromotor spectral decomposition unit includes the muscle weakness assessment steps:

[0090] Extracting signals below the frequency domain from the frequency domain signal The energy amplitude of the principal component of motion;

[0091] The energy amplitude is compared with the maximum velocity threshold from the individualized neural baseline template;

[0092] When the energy amplitude is significantly lower than a preset proportion of the maximum velocity threshold, the muscle weakness feature is marked in the spectral pathological feature vector;

[0093] In this embodiment, the neuromotor spectrum decomposition unit quantifies the loss of limb driving ability through frequency domain feature analysis; after receiving the time-domain displacement signal of the affected limb, the unit converts it into a frequency-domain signal using short-time Fourier transform; during the muscle weakness determination step, the system uses a bandpass filter to specifically isolate and extract frequencies lower than [the specified frequency range]. The system calculates the total energy amplitude within the low-frequency signal component of the frequency band. Simultaneously, it calls the maximum velocity threshold in the individualized neural baseline template and, based on Passevar's theorem that the square integral of the time-domain signal equals the energy integral in the frequency domain, performs the calculation steps for the theoretical energy upper limit: calculates the square value of the maximum velocity threshold, multiplies the square value by a preset time window length coefficient, and then multiplies it by a waveform correction factor with a value between 0.3 and 0.6 to simulate the average power loss under non-constant speed motion, thereby calculating the theoretical energy upper limit with the same dimensions in the frequency domain.

[0094] The logic determination module calculates the ratio of the low-frequency energy amplitude on the affected side to the theoretical upper limit of energy; the logic determination module calculates the ratio of the low-frequency energy amplitude on the affected side to the maximum velocity threshold on the healthy side. If the ratio is lower than the preset muscle attenuation threshold, it indicates that the affected limb lacks the neural driving force to produce effective displacement; based on this determination, the system writes the muscle weakness label and the corresponding attenuation level into the generated spectral pathological feature vector.

[0095] By performing the muscle weakness assessment steps, effective active motor components can be separated from weak tremors, accurately identifying the loss of drive caused by damage to the corticospinal tract, and avoiding the possibility of misjudging involuntary spasms as effective muscle strength.

[0096] Example 5:

[0097] The process of generating spectral pathological feature vectors by the neuromotor spectral decomposition unit also includes an ataxia assessment step:

[0098] exist to Search for peak energy within the intentional tremor frequency band;

[0099] Calculate the ratio of the spectral power density of the energy peak to the average level of background noise determined by the individualized neural baseline template;

[0100] If the ratio exceeds a preset multiple, then the ataxia feature is marked in the spectral pathological feature vector;

[0101] In this embodiment, to distinguish between weakness and ataxia, the neuromotor spectrum decomposition unit further performs an ataxia determination step; this step, based on the characteristics of intention tremor caused by cerebellar injury, locks... to The system performs signal scanning in a specific frequency band, which is the typical frequency range of intention tremor caused by cerebellar lesions recognized in clinical medical statistics; the system searches within this frequency band for significant and prominent energy peaks, i.e. local maxima on the power spectral density curve;

[0102] To verify the statistical significance of the peak, the system calculates the ratio of the power density of the energy peak to the average level of background noise recorded in the individualized neural baseline template, i.e., the signal-to-noise ratio. If the calculated signal-to-noise ratio exceeds a preset abnormal multiple, it indicates that the tremor is not random noise, but a neural oscillation with a specific pathological frequency. Based on this, the system determines that there is a coordination disorder and adds an ataxia label to the spectral pathological feature vector.

[0103] By implementing the ataxia assessment steps, the problem of distinguishing between tremors caused by myasthenia and tremors caused by cerebellar ataxia by visual inspection in traditional clinical observation has been solved. Through quantitative analysis of the frequency domain signal-to-noise ratio, clinicians are provided with key differential criterion for differentiating the location of lesions.

[0104] Example 6:

[0105] The process of generating compensation and driving confidence in the contralateral mirror suppression analysis unit includes:

[0106] Calculate the downward pressure of the unaffected limb or the tilt angle of the trunk;

[0107] Determine whether involuntary antagonistic movements are detected that are opposite in direction to the commanded movement and whose amplitude is positively correlated with the expected amplitude of movement on the affected side;

[0108] If antagonistic movement is detected and its intensity exceeds the resting noise threshold set by the individualized neural baseline template, it is determined that there is a genuine neural driving intention.

[0109] The range of generated values ​​is within to The confidence level of compensation and driving forces between them, where a higher value indicates less compensation component;

[0110] In this embodiment, the contralateral mirror inhibition analysis unit utilizes the Hoover sign principle in neurology, which states that when attempting to move the affected limb, the healthy limb not only does not relax but also produces an involuntary, antagonistic downward pressing motion. When the system issues a command to lift the affected limb, the unit does not rely on the visual image of the affected side but instead locks onto the healthy limb and trunk area for dynamic analysis. In the processing logic, the unit calculates the downward pressure amplitude vector of the healthy limb or the compensatory tilt angle of the trunk in real time.

[0111] The system's core algorithm determines whether significant antagonistic movement is detected based on the following criteria: the direction of the movement is opposite to the direction of the action required on the affected side, and its intensity exceeds the resting noise threshold set in the individualized neural baseline template. The resting noise threshold is obtained by: reading the basic spectral distribution data from the individualized neural baseline template, calculating the integral of the frequency domain signal power spectral density based on the energy conservation principle embodied in Passevar's theorem, and taking the square root of the integral result to obtain the root mean square amplitude value in the time domain, which serves as the comparison benchmark; that is, the resting noise threshold. satisfy:

[0112]

[0113] in The power spectral density of the unaffected limb in a resting state. Pick , Pick Correspondingly, the intensity of competitive sports Defined as the root mean square value of the acceleration of the healthy limb during the movement command from the affected side; if it satisfies ,in, The signal-to-noise ratio coefficient is preferred. And the direction vector of motion Direction of command to the affected side The included angle satisfy If so, it is determined that a valid adversarial movement has been detected;

[0114] If the above conditions are met, the system determines that the patient's brain did indeed issue real neural drive commands, even though there may be no visible displacement on the affected side; based on this, the system generates values ​​within the range of... to The system assesses the confidence level between compensation and drive. If antagonistic movements that conform to physiological laws are detected without trunk tilt, the confidence level tends to 1, indicating that the neural drive is real. If large-scale trunk swaying or other compensatory behaviors are detected, the system judges them as compensatory interference regardless of whether there is an antagonistic response, and adjusts the confidence level value to 0, indicating that the movement intention is not pure.

[0115] The system uses the following linear decay model to calculate the compensation and driving confidence. Ensure that its value is strictly limited to Within the interval:

[0116]

[0117] in, The real-time detected torso tilt angle Let the maximum permissible physiological tilt angle be set as follows: ; The intensity of the exercise is the same as the intensity of the exercise against the unaffected side. Weighting of compensatory punishment for the torso , To drive the confirmation weights, take ; The slope factor of the Sigmoid function;

[0118] The logical meaning of this formula is: the initial confidence level is When trunk tilt is detected, a corresponding weight is deducted, i.e., the first penalty term. If insufficient healthy-side antagonistic drive is not detected, i.e. Close to or less than This causes the Sigmoid function value to approach 0. When the value is lower, the corresponding weight is deducted, i.e., the second penalty term; the combined effect of both results in a lower confidence level when there is compensation or a lack of driving intent. Tend to ;

[0119] By executing the processing logic of the contralateral mirror inhibition analysis unit, this invention innovatively solves the problem of bedding occlusion, which causes all existing vision-based technologies to fail. It indirectly reveals the neurological state of the affected side by utilizing the mirror response of the healthy side, ensuring the continuity and reliability of assessment under non-contact conditions.

[0120] Example 7:

[0121] The process of generating sensory pathway integrity indices using a cross-modal reflex analysis unit includes:

[0122] Time series analysis was performed on the facial motion units of the eyebrow lowering motion unit and the cheek lifting motion unit;

[0123] The time difference from the moment the stimulus is emitted to the moment when the facial muscles produce a detectable displacement is extracted as the latency.

[0124] The latency period is compared with preset cortical response thresholds and spinal reflex thresholds: if the latency period is greater than the cortical response threshold, the corresponding value is output. The output is a score indicator; if the incubation period is between the two, the corresponding output is... The output is a score indicator; if the latency period is less than or equal to the spinal reflex threshold or there is no response, the corresponding output is... The indicators of score;

[0125] In this embodiment, the cross-modal reflectance analysis unit utilizes high-frame-rate reflectance capture technology to replace physical needle prick pain testing. Upon receiving a sudden high-decibel sound or strong light stimulus, the unit activates a high-speed facial tracking algorithm, focusing on monitoring the activation state of the eyebrow lowering action unit and the cheek lifting action unit under the startle reflex. The system accurately measures the time from the moment the stimulus signal is triggered. When the facial muscles first exceed the preset displacement threshold The time difference between these two periods is defined as the incubation period.

[0126] The system logically compares the latency with two physiological thresholds based on nerve conduction velocity: the cortical response threshold and the spinal reflex threshold; if the energy amplitude is greater than or equal to the normal threshold... , Defined as the maximum energy amplitude of the healthy side in an individualized neural baseline template. Then the basic motion score is If the energy amplitude approaches This refers to the upper limit of the sensor's noise floor energy, which is usually taken as the average value of the static background noise. If the basic motor score is [times], then the basic motor score is [number]. If the latency is greater than the cortical response threshold, it indicates that the pain transmission pathway is intact and accompanied by a cognitive response, and an NIHSS score is output. If the latency period is between the two, it indicates the presence of a reflex but cognitive delay, and the output... If the latency period is extremely short or infinite, it is judged as a severe loss of sensation, and output... point;

[0127] By executing the generation process of cross-modal reflex analysis units, zero-contact sensory function assessment is achieved, which has a decisive advantage in infectious disease wards or telemedicine scenarios, completely eliminating the risk of iatrogenic infection and patient fear caused by traditional physical acupuncture.

[0128] Example 8:

[0129] The collaborative scoring decision engine performs the initial screening steps for basic motion scoring, specifically including:

[0130] Receive spectral energy amplitude from the neuromotor spectral decomposition unit;

[0131] Using a pre-defined energy-fraction mapping table, the motor energy of the affected limb is mapped to a range within... to The basic motion score between;

[0132] The mapping table is logically set as follows: if the energy amplitude is greater than or equal to the normal threshold, then the basic motion score is... If the energy amplitude approaches Then the basic motion score is ;

[0133] In this embodiment, the collaborative scoring decision engine initiates the basic motor score screening step, which aims to discretize the continuously changing physical energy values ​​into clinical grades that meet the NIHSS criteria; the engine receives the spectral energy amplitude of the affected limb output by the neuromotor spectrum decomposition unit as an input variable; the engine queries the internally preset energy-score mapping table;

[0134] The mapping table is constructed based on the statistical distribution of large-scale clinical data: when the input energy amplitude is greater than or equal to a preset normal threshold, the system determines that muscle strength is normal and outputs a baseline motor score. When the energy amplitude is extremely low, approaching the sensor's noise floor level, it is determined to be completely paralyzed, and the basic motion score is output. For energy values ​​between the two, the system uses a grading judgment logic that combines vertical displacement components to map them to 1 point, 2 points, or 3 points. Specifically, the system extracts the vertical displacement component of the movement trajectory of the affected limb. If the vertical displacement component is greater than the preset anti-gravity threshold, it is determined that the limb has anti-gravity ability, and the score is mapped to 1 point or 2 points.

[0135] Specific mapping logic functions as follows:

[0136]

[0137] in, For total frequency domain energy, This represents the maximum displacement in the vertical direction. The normal energy threshold is based on an individualized template; Let the threshold for determining anti-gravity displacement be set as follows: ; The upper limit of the sensor's noise floor energy is defined as follows: If the vertical displacement component is less than the anti-gravity threshold, but the horizontal displacement energy is greater than the noise floor level, it is determined that the sensor does not have anti-gravity capability and is mapped to 3 points.

[0138] By performing the basic motor skills assessment initial screening step, the system establishes an objective and quantitative initial screening standard, eliminating the subjective differences in understanding among different doctors regarding the vague description of whether a limb can resist gravity, and providing a solid computational foundation for subsequent fine-tuning.

[0139] Example 9:

[0140] The collaborative scoring decision engine performs a compensatory confidence correction step, which specifically includes: introducing compensatory and driving confidence from the contralateral mirror inhibition analysis unit; and calculating using logical formulas.

[0141] The corrected exercise score uses the following logical formula: add the base exercise score to the correction item and round down.

[0142] The calculation logic for the correction term is as follows: multiply the preset penalty coefficient by the difference between the compensation and the driving confidence obtained by subtracting the value 1 from the value 1.

[0143] Therefore, when high-intensity compensatory movements are detected on the healthy side, resulting in low confidence, the final score is forced to shift towards a value representing a more severe level of paralysis, in order to eliminate false rehabilitation score bubbles.

[0144] In this embodiment, the collaborative scoring decision engine performs a crucial compensatory confidence correction step to identify and eliminate artificially inflated scores obtained through skillful compensatory actions; this step introduces compensatory and driving confidence from the contralateral mirror suppression analysis unit. The system uses a correction algorithm to calculate the final score, and its core logic lies in constructing a penalty item based on the amount of missing confidence.

[0145] The specific calculation logic is as follows: the system calculates the value. Subtract confidence level The difference represents the degree of compensation; this difference is multiplied by a preset penalty coefficient. The system obtains the correction term value; it then adds the correction term to the aforementioned basic motion score and rounds the sum to the nearest integer to obtain the final score. The physical meaning of this algorithm is that when high-intensity compensatory motion is detected on the healthy side, the correction term value increases sharply, forcing the final score to shift to a higher value.

[0146] By performing a compensatory confidence correction step, a cheating firewall is built at the algorithm level, which can effectively eliminate the false limb displacement caused by the patient's trunk swinging, and ensure that the final output NIHSS score truly reflects the remaining function of the corticospinal tract on the affected side, rather than the patient's compensatory skills.

[0147] Example 10:

[0148] The collaborative scoring decision engine is also configured as follows:

[0149] The corrected motion score is weighted and summarized with the sensory pathway integrity index from the cross-modal reflex analysis unit to generate a structured score matrix;

[0150] The output includes a standardized NIHSS electronic scoring report containing the specific scores for each sub-item and the total score, and a compensation confidence warning is marked in the report for items whose scores have changed due to compensation correction.

[0151] In this embodiment, after completing the calculation and correction of all individual indicators, the collaborative scoring decision engine performs the multi-dimensional data aggregation task of the terminal. The engine integrates the compensated motion score, the ataxia characteristics obtained from frequency domain analysis, and the sensory pathway integrity index in a structured manner to construct a structured scoring matrix.

[0152] The system renders and outputs a standardized NIHSS electronic scoring report; the report not only displays the scores of each sub-item and the total score in numerical form, but also has an intelligent risk warning function: for items whose scores have changed after the compensation confidence correction step, the system will prominently mark the compensation confidence warning icon and the specific confidence value in the corresponding position of the report.

[0153] By executing the output configuration of the collaborative scoring decision engine, the system not only provides doctors with a final diagnostic result, but also more transparently displays the derivation process behind the scoring and potential data risks, helping doctors quickly identify patients' compensatory habits and thus develop more targeted rehabilitation training programs.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automated assessment system for NIHSS scores in stroke patients, characterized in that, include: The multimodal induction and perception unit, as the front-end data entry point, is configured to issue synchronous commands in a preset sequence through a composite command generator, establish a stimulus-feedback closed loop, and collect patient responses in real time to generate video streams and synchronous audio streams containing depth and color information. The dynamic benchmark calibration unit, as the cornerstone of pathological judgment, is configured to receive video streams, guide the patient's healthy limbs to complete calibration actions, extract kinematic features to construct an individualized neural benchmark template, thereby defining a dynamic reference system for subsequent abnormality judgment; The neuromotor spectrum decomposition unit is configured to receive the video stream of the affected limb and the individualized neural reference template, convert the spatial displacement into a frequency domain signal, and analyze the low-frequency components and specific frequency band peaks based on the threshold in the template to generate a spectrum pathological feature vector that distinguishes between myasthenia and ataxia. The contralateral mirror inhibition analysis unit, as a correction module to overcome occlusion and compensation, is configured to lock the healthy limb area when a movement command is issued to the affected side, detect the antagonistic movement of the healthy side and compare it with the resting noise threshold in the template, and generate compensation and drive confidence that characterize the real neural drive intention. The cross-modal reflex analysis unit is configured to emit sudden audio-visual stimuli, capture facial micro-expressions and calculate latency, and generate sensory pathway integrity indices based on differences in nerve conduction velocity. The collaborative scoring decision engine, acting as the coupling control center, is configured to receive feature vectors, confidence scores, and integrity indices. It uses a hierarchical weighting strategy combined with compensatory correction logic to map multimodal features into standardized NIHSS scores. The composite instruction generator in the multimodal induction and perception unit is configured to synchronously issue voice instructions and visual induction signals such as the flashing of red dots on the screen according to a preset timing sequence. The multimodal induction and perception unit is further configured to directly distribute and transmit the acquired high-dimensional video stream as the raw signal to be processed to the dynamic reference calibration unit and the neural motion spectrum decomposition unit. The process by which the dynamic benchmark calibration unit constructs an individualized neural benchmark template includes: Extract the maximum velocity threshold of the healthy limb movement as a reference upper limit for assessing muscle strength; The motion smoothness coefficient calculated based on the principle of minimizing jerk is extracted as a quantitative indicator reflecting the fineness of the nervous system's control. Extract the basic spectral distribution of the healthy limb as a reference for the background noise level; The maximum velocity threshold, motion smoothness coefficient, and basic spectral distribution are packaged to generate the individualized neural baseline template, and then sent to the neural motion spectrum decomposition unit and the contralateral mirror inhibition analysis unit as the zero point for judging abnormalities. The process by which the neuromotor spectrum decomposition unit generates spectral pathological feature vectors includes a muscle weakness determination step: Extracting the energy amplitude of the principal motion component below 1Hz from the frequency domain signal; The energy amplitude is compared with the maximum velocity threshold from the individualized neural baseline template; When the energy amplitude is significantly lower than a preset ratio of the maximum velocity threshold, the myasthenia gravis feature is marked in the spectral pathological feature vector; The process of generating spectral pathological feature vectors by the neuromotor spectrum decomposition unit also includes an ataxia determination step: Search for energy peaks in the intentional flutter frequency band from 3 Hz to 5 Hz; Calculate the ratio of the spectral power density of the energy peak to the average level of background noise determined by the individualized neural benchmark template; If the ratio exceeds a preset multiple, then ataxia features are marked in the spectral pathological feature vector; The process by which the contralateral mirror suppression analysis unit generates compensation and driving confidence scores includes: Calculate the downward pressure of the unaffected limb or the tilt angle of the trunk; Determine whether involuntary antagonistic movements are detected that are opposite in direction to the commanded movement and whose amplitude is positively correlated with the expected amplitude of movement on the affected side; If the antagonistic movement is detected and its intensity exceeds the resting noise threshold set by the individualized neural baseline template, it is determined that there is a genuine neural driving intention. Generate the compensation and driving confidence scores with values ​​ranging from 0 to 1, where higher values ​​represent less compensation components; The process by which the cross-modal reflex analysis unit generates a sensory pathway integrity index includes: Time series analysis was performed on the facial motion units of the eyebrow lowering motion unit and the cheek lifting motion unit; The time difference from the moment the stimulus is emitted to the moment when the facial muscles produce a detectable displacement is extracted as the latency. The latency period is compared with the preset cortical response threshold and spinal reflex threshold: if the latency period is greater than the cortical response threshold, the corresponding index of 0 points is output; if the latency period is between the two, the corresponding index of 1 point is output; if the latency period is less than or equal to the spinal reflex threshold or there is no response, the corresponding index of 2 points is output. The collaborative scoring decision engine performs a preliminary screening step for basic motion scoring, specifically including: Receive the spectral energy amplitude from the neural motor spectrum decomposition unit; Using a pre-defined energy-fraction mapping table, the motor energy of the affected limb is mapped to a basic motor fraction ranging from 0 to 4; The mapping table is logically set as follows: if the energy amplitude is greater than or equal to the normal threshold, the basic motion score is 0; if the energy amplitude is close to 0, the basic motion score is 4.

2. The automated NIHSS scoring assessment system for stroke patients according to claim 1, characterized in that, The collaborative scoring decision engine performs a compensatory confidence correction step, specifically including: The compensation and driving confidence from the contralateral mirror suppression analysis unit are introduced; The corrected motion score is calculated using a logical formula configured as follows: the base motion score is added to the correction term and then rounded down. The calculation logic of the correction term is as follows: multiply the preset penalty coefficient by the difference between the compensation and the driving confidence obtained by subtracting the value 1 from the value 1. Therefore, when high-intensity compensatory movements are detected on the healthy side, resulting in low confidence, the final score is forced to shift towards a value representing a more severe level of paralysis, in order to eliminate false rehabilitation score bubbles.

3. The automated NIHSS scoring assessment system for stroke patients according to claim 2, characterized in that, The collaborative scoring decision engine is also configured as follows: The corrected motion score is weighted and summarized with the sensory pathway integrity index from the cross-modal reflex analysis unit to generate a structured score matrix; The output is a standardized NIHSS electronic scoring report containing the specific scores for each sub-item and the total score. The report also includes a compensation confidence warning for items whose scores have changed due to compensation adjustments.

Images