NIHSS scoring diagnosis system based on artificial intelligence and system establishment method thereof

Abstract

The invention relates to an NIHSS score diagnosis system based on artificial intelligence and a system establishment method thereof, and belongs to the technical field of artificial intelligence. The method comprises the following steps: S1, acquiring voice, face images, limb movement and eyeball movement data of a patient through a testing device; s2, analyzing the voice response by using a voice recognition module, and matching the voice response with a standard answer; s3, extracting facial expressions, pupil sizes, limb movement amplitudes and ataxia coordination features through an image processing module; s4, mapping the feature data into each item score based on an NIHSS scoring rule, and calculating a total score; and S5, outputting a scoring result and a corresponding nerve function defect severity level. Quantitative measurement of various examination results is improved, standardization of NIHSS scores is realized, subjectivity of doctor operation is reduced, workload of doctors is reduced, more accurate treatment decisions are further provided for patients, and accurate development of clinical research is promoted.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology and relates to an artificial intelligence-based NIHSS scoring diagnosis system and a method for establishing the system. Background Art

[0002] Stroke, a major disease with extremely high rates of disability and mortality worldwide, places a heavy burden on patients, their families, and society. Accurately assessing the extent of neurological deficits in stroke patients is crucial in the clinical diagnosis and treatment of stroke. This not only supports accurate diagnosis and appropriate treatment planning, but also plays a key role in predicting prognosis. The National Institutes of Health Stroke Scale (NIHSS) score emerged in this context and has become a widely recognized and widely used tool for assessing the extent of neurological deficits in stroke patients. The NIHSS score covers 15 areas, including consciousness, movement, and sensation, with a total score of 42, with higher scores indicating more severe neurological damage. In clinical practice, the NIHSS score plays a multifaceted role. First, it provides physicians with an objective and quantitative basis for assessing the condition of patients with cerebral infarction, helping them to fully understand their neurological status. Based on the NIHSS score, physicians can more accurately predict a patient's prognosis, providing strong support for personalized treatment decisions. Secondly, the NIHSS score is standardized, enabling a unified and standardized assessment of a patient's neurological function. This allows doctors to regularly monitor treatment outcomes, adjust treatment plans promptly, and improve the effectiveness and specificity of treatment. Furthermore, due to its standardized and objective nature, the NIHSS score is widely used in clinical research, providing a unified standard for data comparison and results analysis across studies.

[0003] Currently, the NIHSS is typically assessed by trained physicians. Based on the patient's actual responses and performance, physicians strictly follow the scoring criteria to assign a score to each item. The total score is then summed up, with higher scores indicating more severe neurological impairment. In practice, physicians use the NIHSS to assess patients with cerebral infarction on-site, offering the advantages of speed and convenience, enabling them to quickly obtain information on the patient's neurological function.

[0004] However, with the widespread use of the NIHSS score in clinical practice, its shortcomings have gradually become apparent. Throughout the diagnosis and treatment of cerebral infarction, the patient's neurological function needs to be assessed multiple times due to the dynamic changes in the condition. This means that doctors need to devote a lot of time and energy to completing the NIHSS score, which invisibly increases the doctor's workload and may cause fatigue and negligence in the busy clinical work, thereby affecting the accuracy of the score. More importantly, there is a certain degree of subjectivity in the scoring of some items in the NIHSS score sheet. Different doctors have different understandings and grasps of the NIHSS score sheet. When faced with the same presentation of the same patient, they may give different scores based on their personal experience and judgment. The scoring differences caused by this subjectivity not only affect the accurate judgment of the patient's condition, but may also mislead treatment decisions and prognosis predictions, and have an adverse effect on the patient's treatment.

[0005] Therefore, to ensure the accuracy and reliability of the NIHSS score and improve the diagnosis and treatment of stroke patients, standardization of the NIHSS score is particularly important. Only by establishing more stringent and standardized scoring criteria and operating procedures, and strengthening the training and assessment of physicians, can we minimize the subjectivity and errors of the score and enable the NIHSS score to better serve the clinical diagnosis, treatment, and research of stroke. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide an artificial intelligence-based NIHSS scoring diagnosis system and a system establishment method thereof, so as to achieve standardization of NIHSS scores and provide patients with more accurate treatment decisions.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] An artificial intelligence-based NIHSS scoring and diagnosis system, the system comprising a testing device, a data acquisition module, a data analysis unit, a standardized scoring algorithm module, and an output module;

[0009] The testing device is used to collect physiological data of the patient, including: a facial and limb movement detection camera for capturing the patient's facial expressions and limb movements; a pupil detection camera and an infrared light source for tracking eye movements and pupil size changes; a microphone and a voice interaction unit for collecting the patient's voice response; and a visual field detection display for presenting a dynamic grid to assess the patient's visual field range.

[0010] The data acquisition module includes: a voice acquisition unit that acquires the patient's voice response to standardized questions through a microphone; an image acquisition unit that acquires facial, limb and eye movement images through a multi-angle camera;

[0011] The data analysis unit includes: a speech recognition module that calls an open source API to convert speech into text and compares it with preset answers; an image processing module that uses the YOLO, U-Net and MediaPipe frameworks to identify facial key points, pupil segmentation, body posture and co-motion trajectory;

[0012] The standardized scoring algorithm module quantifies the data analysis results into scores for each item according to the NIHSS scoring rules and generates a total score;

[0013] The output module is used to display the scoring results and the corresponding neurological impairment severity grades.

[0014] Furthermore, in the test device, the pupil detection camera uses a CMOS camera with a resolution of 1920×1080 or above and a frame rate ≥30fps, and is equipped with a lens with a focal length of 5-10mm and an aperture of F2.0-F2.8, and the infrared light source wavelength is 850nm or 940nm.

[0015] Furthermore, in the image processing module, facial expression recognition uses a fine-grained deep learning model to judge the degree of facial paralysis through the geometric features of the facial features (eye closure, mouth corner displacement); limb movement analysis uses MediaPipe to extract 21 hand key points and 31 body posture key points to calculate joint angle changes and movement amplitudes.

[0016] Furthermore, in the standardized scoring algorithm module, for the level of consciousness item, the patient's eye opening and command response are tested through a decibel gradient voice awakening test, and the score is automatically scored in combination with the voice recognition results and facial movement detection data; for the gaze item, the level of movement abnormality is determined based on the comparison between the eye movement trajectory and the preset normal range.

[0017] A method for establishing an artificial intelligence-based NIHSS scoring diagnostic system, the method comprising the following steps:

[0018] S1. Collect the patient's voice, facial image, limb movement and eye movement data through the testing device;

[0019] S2. parse the voice response using the speech recognition module and match it with the standard answer;

[0020] S3, extracting facial expressions, pupil size, limb movement amplitude and coordination features through image processing module;

[0021] S4. Based on the NIHSS scoring rules, the characteristic data is mapped into scores for each item and the total score is calculated;

[0022] S5. Output the scoring results and the corresponding neurological impairment severity grade.

[0023] Furthermore, in step S3, the limb movement amplitude is calculated by the change of joint angles in continuous video frames, and the difference between the initial posture and the maximum lifting posture is compared; the coordination is evaluated by the dynamic error of the distance between the finger and the nose tip in the finger-nose test.

[0024] Furthermore, in step S4, the scoring of the visual field item is based on the patient's visible range of the dynamic red grid, and the degree of visual field loss is determined by the preset normal visual field angles (60° nasal, 100° temporal, 60° above, and 75° below).

[0025] The beneficial effects of the present invention are:

[0026] The present invention provides a rapid, artificial intelligence-based, human-computer interaction NIHSS automatic diagnostic system, which improves the quantitative measurement of various examination results, realizes the standardization of NIHSS scores, reduces the subjectivity of doctors' operations, reduces doctors' workload, further provides patients with more accurate treatment decisions, and promotes the precise development of clinical research.

[0027] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0029] Figure 1 It is a diagram of the overall structure of the system of the present invention;

[0030] Figure 2 This is the overall logic block diagram of the system of the present invention;

[0031] Figure 3 This is the overall structure diagram of the test device;

[0032] Figure 4 A schematic diagram of a checkerboard back panel;

[0033] Figure 5 This is a schematic diagram of the head fixation auxiliary device;

[0034] Figure 6 This is a schematic diagram of the visual field measurement image;

[0035] Figure 7 Schematic diagram of eye movement detection;

[0036] Figure 8A reference line for monocular eye movement testing. DETAILED DESCRIPTION

[0037] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0038] Figure 1 This is a diagram of the overall structure of the system of the present invention. Figure 2 The overall logic block diagram of the system of the present invention is shown in the figure. As shown in the figure, the present invention adopts an artificial intelligence-based human-computer interaction NIHSS scoring and diagnosis system, which includes:

[0039] Input module, used to input data collected by the task;

[0040] A feature extraction module is used to process the data obtained by the input module, perform feature extraction and scoring;

[0041] The data collected by the task includes the following data:

[0042] Data 1: Level of consciousness (1a, 1b, 1c): The machine asks questions, the subject answers, and the speech sequence is collected;

[0043] Data 2: Gaze and eye movements (2): Record the subject's horizontal eye movements according to the machine's requirements and collect eye movement data;

[0044] Data 3: Visual field (3): The machine detects the subject's visual field and collects visual field defect data;

[0045] Data 4: Facial strength (4): Record the subjects’ facial muscle movements according to the machine’s requirements and collect facial muscle movement data;

[0046] Data 5: Upper limb movement (5a, 5b): Record the subject's upper limb movement according to the machine's requirements and collect upper limb movement data;

[0047] Data 6: Lower limb movement (6a, 6b): Record the subject's movement of raising both lower limbs according to the machine's requirements and collect lower limb movement data;

[0048] Data 7: Ataxia (7): Record the subjects' ataxia movements according to the machine's requirements and collect ataxia movement data;

[0049] Data 8: Sensation (8): The machine symmetrically pricks and touches the patient's head, face, and limbs to collect sensory data;

[0050] Data 9: Language (9): Subjects took naming and reading tests, and speech sequences were collected;

[0051] Data 10: Articulation disorder (10): The subject repeats the words spoken by the machine and collects the speech sequence;

[0052] Data 11: Ignore (11): Record the subject's test and response according to the machine's requirements, and collect ignore data.

[0053] The software system and human-computer interface of the present invention include an image / speech acquisition unit, an analysis and detection module, and a NIHSS assessment unit. The NIHSS testing device includes the hardware and auxiliary devices required to complete the automated patient testing. In this embodiment, the system consists of the testing device, acquisition module, data analysis unit, and standardized scoring algorithm. The specific implementation plan for each component is as follows:

[0054] NIHSS testing apparatus:

[0055] The overall composition of the test device is as follows Figure 3 As shown, the facial and limb movement detection camera and the pupil detection camera both use industrial cameras, and use LED light sources and infrared light sources respectively.

[0056] 1) Camera: Choose a high-resolution, high-frame-rate camera to ensure clear, detailed eye images. For example, a CMOS camera with a resolution of 1920×1080 or higher and a frame rate of 30fps or higher can be used to capture eye images under various lighting conditions and dynamic scenes.

[0057] 2) Optical lens: Select a lens with an appropriate focal length, aperture, and field of view based on the actual application scenario and measurement distance. For close-up eye measurements, a lens with a focal length of 5-10mm and an aperture of F2.0-F2.8 can be used to obtain clear close-up images of the eye.

[0058] 3) Light source: Infrared light sources can achieve better pupil image contrast without causing discomfort to the subject, such as infrared LED lights with a wavelength of 850nm or 940nm. For limb and facial movement detection, a ring-shaped LED light source is used to be installed around the camera.

[0059] 4) Choose a clip-on wireless microphone, which the patient carries.

[0060] 5) Choose an integrated display device with integrated speakers.

[0061] The server loads the software system and runs the human-computer interface to complete the collection of images, audio, etc., as well as evaluation and analysis.

[0062] Facial movement and limb movement testing device:

[0063] In this embodiment, the facial movement and limb movement testing device is a 5cm×5cm checkerboard backboard with a patient standing or sitting. Figure 4This is a schematic diagram of a checkerboard backplane, through which facial images and limb movement images of patients according to doctor's instructions can be collected.

[0064] Eye movement detection device:

[0065] In this embodiment, the Figure 5 The patient head fixation assist device shown helps keep the patient's head fixed, allowing only eye movement. The bracket is integrated with a pupil detection camera and infrared light source. The camera captures images of pupil changes.

[0066] On the other hand, perimetry images such as Figure 6 A grid image, each 3mm x 3mm, is displayed on a 180° curved display, approximately 33cm from the eyes. Visual field measurement: A voice prompt prompts the patient to continuously focus both eyes on a yellow grid in the center of their visual field. Red grids appear 90° upward, downward, left, and right from the center, and the patient's focus is recorded. The distribution of the red grids recorded by each eye can be used to assess the patient's binocular visual field.

[0067] Eye movement detection: Figure 7 As shown, a voice prompt prompts the patient to track the red grid with both eyes. The red grid moves on the grid image, gradually moving upward, downward, left, and right from the center point within a 90° area. Images of eye movements within these areas are recorded, and the maximum distance of eye movement in each direction is selected as the eye movement data for subsequent image analysis.

[0068] Acquisition module:

[0069] 1) Human-computer dialogue

[0070] The patient answers the corresponding questions according to the doctor's questions. The patient's voice is collected through the microphone, and the data is transmitted through the wireless communication interface and stored in the server.

[0071] 2) Face / eye / body images

[0072] The doctor issues different instructions according to the test process, and the patient performs the doctor's requirements according to the instructions. The doctor uses the acquisition operation buttons in the software system to implement corresponding image acquisition and store them on the server.

[0073] Data Analysis Unit:

[0074] 1) Human-computer dialogue

[0075] Patients answer the corresponding questions based on the doctor's questions. The patient's voice is collected through the microphone, and the voice recognition API (developed by calling open source model interfaces such as Baidu) outputs the corresponding text, which is compared with the information entered by the patient to automatically identify and determine whether the answer is correct.

[0076] 2) Facial image processing

[0077] For the facial images of patients under different doctor's instructions, the face ROI area is identified through target detection algorithms such as YOLO, and the facial features positioning algorithm (open source model) is called to obtain the key point positioning features of the facial features. Then, the aspect ratio geometric features of the eyes and mouth can be used to identify the closed state of the eyes and mouth.

[0078] More complex actions such as sticking out the tongue, frowning, and showing teeth can be detected by adding corresponding fine-grained data sets to train the YOLO target detection algorithm.

[0079] 3) Eye movements

[0080] Eye movement range information can be measured using eye movement images. Pupil size is determined by segmenting pupil area pixels using the pupil image segmentation model U-Net. Furthermore, pupil diameter is measured based on calibration information. Figure 8 The reference line for monocular eye movement examination, the measurement and judgment standards are as follows:

[0081] Range of eye movement: When the normal eyeball turns inward, the inner edge of the pupil can reach the straight line connecting the upper and lower lacrimal puncta, and when it turns outward, the cornea can reach the outer canthus. The upward and downward movement range is about 5mm.

[0082] Inward rotation: When the eyeball rotates inward, the inner edge of the pupil can reach the line connecting the upper and lower tear points of the inner and outer canthus.

[0083] Abduction: When the eyeball rotates outward, the outer edge of the cornea can reach the outer canthus.

[0084] Upward rotation: When the eyeball rotates upward, the lower edge of the cornea can reach the horizontal reference line (the line connecting the inner canthus and the outer canthus).

[0085] Downward rotation: When the eyeball rotates downward, the upper edge of the cornea can reach the horizontal reference line or slightly higher.

[0086] 4) Limb movement image processing

[0087] Limb movement uses Google's Mediapipe open source framework, which can extract 21 feature points from the hands and 31 key features from the body posture.

[0088] For the recognition of patient hand movements, images of normal people's fist-clenching postures and patients' fist-clenching postures of different degrees are collected. The 21 key points that can be identified by the framework are used to form a data set for training hand movement classification and recognition models (for example, constructing a KNN classifier).

[0089] When collecting images of patients performing limb-lifting movements, they should be asked to stand or sit sideways to facilitate observation and measurement of the extent of limb lift. During observation, continuous video of the patient's movements is recorded. The extent of lift can be determined by tracking the changes in the angles between the upper and lower limbs and the torso in the images above.

[0090] To increase its measurement reliability, the specific angle value can be calculated by the change value calculated by the posture estimation of the two feature points, the initial image taken by the system before the patient raises the upper and lower limbs, and the image when the upper and lower limbs are raised to the highest point, and the maximum lifting angle of the upper and lower limbs can be calculated through feature matching.

[0091] Posture tracking can simultaneously measure values ​​such as the speed of lifting, which can be used to more accurately assess the patient's condition.

[0092] 5) Image processing of limb coordination movements

[0093] Google's Mediapipe open source framework is used to obtain the coordinates of the key points of the patient's nose and hand posture, and then the smoothness of the patient's finger motion trajectory and the final distance and position relationship between the finger and nose are calculated.

[0094] Standardized scoring algorithm

[0095] In combination with the conventional NIHSS scoring method, the present invention proposes a new quantitative scoring method to improve the accuracy and repeatability of the assessment. The specific scoring is as follows:

[0096] 1. Level of consciousness (1a): The patient's name is called out in different decibels, and the patient is scored based on whether they open their eyes and complete the level of consciousness questions (1b) and level of consciousness instructions (1c). Facial images are used to identify the state of eyes open and closed, and the duration of time.

[0097] 0 points: The patient's name is called by a low-decibel voice (below 50 points), the patient can open his eyes, and can correctly complete the level of consciousness questions (1b) and level of consciousness instructions (1c).

[0098] 1 point: When the patient's name is called by a low-decibel voice (below 50 points), the patient opens his eyes and can correctly complete the level of consciousness questions (1b) and level of consciousness instructions (1c). After the questions are stopped, the patient closes his eyes immediately (closes his eyes for less than 3 seconds after answering).

[0099] 2 points: The patient's name is called by a high-decibel voice (60 decibels or above). The patient opens his eyes and closes his eyes after partially completing the level of consciousness question (1b) and the level of consciousness instruction (1c) (eyes closed for <3 seconds after the answer is completed).

[0100] 3 points: The patient's name is called by a low-to-high-decibel voice, but the patient does not open his eyes and is unable to complete the level of consciousness questions (1b) and level of consciousness instructions (1c).

[0101] 2. Questions about the level of consciousness (1b): The voice asks the patient questions such as "What year is it?" and "Where is this place?" The human-computer dialogue analyzes the accuracy of the patient's answers.

[0102] 0 points: Both questions were answered correctly.

[0103] 1 point: Only one question was answered correctly.

[0104] 2 points: Neither question can be answered correctly.

[0105] 3. Level of consciousness instructions (1c): Voice prompts the patient to complete the following two actions: "open eyes, close eyes" and "clench fist, open hand". Facial images recognize the state of the eyes and mouth, and images recognize hand movements to assess whether the patient can complete the instructed actions.

[0106] 0 points: Both actions are correct.

[0107] 1 point: One of the actions is correct.

[0108] 2 points: Both actions are incorrect.

[0109] 4. Gaze (2): The patient stares at the screen and tracks a moving red grid with both eyes. The red grid appears in eight directions within the eye movement range: up and down, left and right, outside and up, outside and down, inside and up, and inside and down. The patient is asked to track the red grid with both eyes. The eye movement is evaluated based on the degree of overlap between the pupil movement trajectory of both eyes and the trajectory of the red grid.

[0110] 0 points: The trajectories completely overlap.

[0111] 1 point: The trajectories do not completely overlap.

[0112] 2 points: No pupil movement in both eyes.

[0113] 5. Visual Field (3): The patient looks at the yellow square in the center of the screen. Red grids appear from the inside out in eight directions: up and down, left and right, outside and up, outside and down, inside and up, and inside and down. The number of times the patient can see the red grids and their distribution are recorded one by one. The binocular visual field is evaluated based on the distribution of the red grids visible to the patient in each direction.

[0114] 0 points: The visual field of both eyes is intact (the red grid within the normal visual field is visible. The nasal side is 60°, the temporal side is 100°, the upper side is 60°, and the lower side is 75°).

[0115] 1 point: Part of the visual field of one or both eyes is lost (the visual range of one or both eyes cannot fully reach the normal visual field range mentioned above).

[0116] 2 points: Hemianopia occurs in both eyes or one eye is completely blind (one side is affected by both eyes at the same time, such as the red grid cannot be seen in the nasal, temporal, upper, or lower visual fields, or one eye cannot see the red grid at all).

[0117] 3 points: Both eyes are completely blind (both eyes cannot see the red grid at all).

[0118] 6. Facial paralysis (4): A facial capture camera is used to prompt the patient to smile, frown, close eyes, puff out cheeks, or whistle. Images of each movement are captured and the facial images are used to analyze the movement of the facial expression muscles on both sides and the degree of bilateral symmetry.

[0119] 0 points: normal facial expression muscle movement (bilateral forehead wrinkles, nasolabial folds, and corners of the mouth are completely symmetrical).

[0120] 1 point: The upper facial muscles on one side have normal movement, while the lower facial muscles have poorer movement than the other side (the forehead lines are symmetrical on both sides, but the nasolabial grooves and corners of the mouth are asymmetrical on both sides).

[0121] 2 points: The upper facial muscles on one side have normal movement, while the lower facial muscles cannot move (the forehead lines on both sides are symmetrical, and the nasolabial groove and corner of the mouth on one side have no movement).

[0122] 3 points: The facial muscles on one side cannot move at all (no movement of the forehead wrinkles, nasolabial folds and corners of the mouth on one side).

[0123] 7. Upper limb movement (5): Using a limb movement detection camera, the patient lies flat on their back. Voice prompts the patient to straighten and raise both upper limbs at a 45° angle to the bed surface and hold for 10 seconds without lowering them. Images of the upper limb movements are captured, and the angle between the end point of the upper limb raising and the supine surface and the duration of the movement are identified. Scoring is performed on the left and right upper limbs separately.

[0124] 0 points: Lift the body up to 45° and maintain for 10 seconds.

[0125] 1 point: Lifting angle 45°, holding time <10 seconds.

[0126] 2 points: lifting angle <45°, holding time <10 seconds.

[0127] 3 points: Unable to lift, but movement is visible.

[0128] 4 points: No movement of the upper limbs.

[0129] 9 points: The examination cannot be performed due to amputation or joint fusion.

[0130] 8. Lower limb movement (6): A limb movement detection camera is used. The patient lies flat on their back. Voice prompts the patient to straighten and raise both lower limbs at a 30° angle to the bed surface and hold them for 5 seconds without lowering. Movement images of both lower limbs are collected. The images identify the angle between the end point of the raising of both lower limbs and the supine surface, the duration, and the speed of descent. The left and right lower limbs are scored separately.

[0131] 0 points: lifting angle 30°, maintained for ≥5 seconds.

[0132] 1 point: Lift up to 30°, maintain for 5 seconds, then drop, with a slow descent speed (descent speed < acceleration due to gravity).

[0133] 2 points: Lifting angle 30°, rapid descent in less than 5 seconds (descent speed = acceleration due to gravity).

[0134] 3 points: The elevation angle is less than 30°, and movement is visible.

[0135] 4 points: No movement of the lower limbs.

[0136] 9 points: The examination cannot be performed due to amputation or joint fusion.

[0137] 9. Limb ataxia (7): Using a limb motion detection camera, the patient is prompted by voice to complete the finger-to-nose movement. Movement images of both upper limbs are collected separately. The images are used to identify the smoothness of the finger movement trajectory, the distance and positional relationship between the finger end point and the nose, and to assess the coordination of the upper limbs: flexibility, incoordination, tremor, and accuracy.

[0138] 0 points: Coordinated limb movements (finger movement trajectory is smooth, and the end point of the finger completely coincides with the nose).

[0139] 1 point: Uncoordinated movement of one limb (the movement trajectory of one finger is curved, or the end point of the finger does not completely coincide with the nose).

[0140] 2 points: Uncoordinated limb movements (the movement trajectories of the fingers on both sides are curved, or the end points of the fingers do not completely coincide with the nose).

[0141] 10. Sensation (8): The doctor uses large needles to symmetrically stimulate the patient's face, upper limbs, and lower limbs on both sides, and evaluates the patient's sensation based on the location, nature, and intensity of the stimulation perceived by the patient.

[0142] 0 points: Able to accurately perceive the location, nature and intensity of the stimulus.

[0143] 1 point: decreased pain sensation, but tactile sensation is present.

[0144] 2 points: loss of pain and touch sensation.

[0145] 11. Language (9): The patient is asked by voice about his / her name, age, and location. The patient is prompted by voice to name the objects in the screen image ("cup", "key", "watch", "pencil"), and the patient's listening comprehension and naming accuracy are analyzed.

[0146] 0 points: Completely normal (all answers are correct).

[0147] 1 point: Partial aphasia, but partial communication possible (correct answer rate ≥ 30%).

[0148] 2 points: severe aphasia, difficulty in communication (correct answer rate <30%).

[0149] 3 points: Complete aphasia (no answer or completely wrong answer).

[0150] 12. Articulation disorder (10): Prompt the patient to read the following words: "Mom, Dada, Dad, Watermelon" and assess the patient's pronunciation clarity.

[0151] 0 points: Completely normal (pronunciation is clear and completely recognized).

[0152] 1 point: Mild to moderate dysarthria, understandable (some pronunciations are unclear and cannot be fully recognized).

[0153] 2 points: Severe articulation disorder, incomprehensible (completely unrecognizable).

[0154] 9 points: Unable to pronounce due to other reasons.

[0155] 13. Neglect (11): Tactile neglect: Simultaneously needle the patient's upper and lower limbs to see if the patient can perceive bilateral stimulation; Visual neglect: When a red grid appears in the left and right visual fields of the screen at the same time, see if the patient can see bilaterally.

[0156] 0 points: No neglect.

[0157] 1 point: Any neglect occurs.

[0158] 2 points: There is partial neglect.

[0159] The innovative scoring diagnostic system inspection and quantitative scoring method of the present invention are presented in a table format below.

[0160] NIHSS scoring diagnostic system examination and quantitative scoring method

[0161]

[0162]

[0163]

[0164] Quantitative evaluation results:

[0165] The severity of stroke was quantitatively assessed based on the above NIHSS score.

[0166] Methods: The NIHSS score is the sum of the above items. The higher the score, the more severe the neurological deficit.

[0167] NIHSS score 0 points 1-3 points 4-15 points 16 points and above Severity rating normal Minor stroke medium stroke Severe stroke

[0168] The technical solution of this invention can be applied to healthcare institutions, facilitating condition assessment, prognosis prediction, treatment effect monitoring, and clinical research for patients with cerebral infarction. This product can also be used at home or online by patients with cerebral infarction, facilitating their monitoring of their own neurological function and promoting their understanding of stroke.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based NIHSS scoring and diagnostic system, characterized by: The system includes a testing device, a data acquisition module, a data analysis unit, a standardized scoring algorithm module and an output module; The testing device is used to collect physiological data of the patient, including: a facial and limb movement detection camera for capturing the patient's facial expressions and limb movements; a pupil detection camera and an infrared light source for tracking eye movements and pupil size changes; a microphone and a voice interaction unit for collecting the patient's voice response; and a visual field detection display for presenting a dynamic grid to assess the patient's visual field range. The data acquisition module includes: a voice acquisition unit that acquires the patient's voice response to standardized questions through a microphone; an image acquisition unit that acquires facial, limb and eye movement images through a multi-angle camera; The data analysis unit includes: a speech recognition module that calls an open source API to convert speech into text and compares it with preset answers; an image processing module that uses the YOLO, U-Net and MediaPipe frameworks to identify facial key points, pupil segmentation, body posture and co-motion trajectory; The standardized scoring algorithm module quantifies the data analysis results into scores for each item according to the NIHSS scoring rules and generates a total score; The output module is used to display the scoring results and the corresponding neurological impairment severity grades.

2. The NIHSS scoring and diagnostic system based on artificial intelligence according to claim 1, characterized in that: In the test device, the pupil detection camera uses a CMOS camera with a resolution of 1920×1080 or above and a frame rate ≥30fps, and is equipped with a lens with a focal length of 5-10mm and an aperture of F2.0-F2.8, and the infrared light source wavelength is 850nm or 940nm.

3. The NIHSS scoring and diagnostic system based on artificial intelligence according to claim 2, characterized in that: In the image processing module, facial expression recognition uses a fine-grained deep learning model to determine the degree of facial paralysis through the geometric features of the facial features (eye closure, mouth corner displacement); limb movement analysis uses MediaPipe to extract 21 hand key points and 31 body posture key points to calculate joint angle changes and movement amplitude.

4. The artificial intelligence-based NIHSS scoring and diagnostic system according to claim 3, characterized in that: In the standardized scoring algorithm module, for the level of consciousness item, the patient's eye opening and command response are tested through a decibel gradient voice awakening test, and the score is automatically scored in combination with the voice recognition results and facial movement detection data; for the gaze item, the level of movement abnormality is determined based on the comparison between the eye movement trajectory and the preset normal range.

5. A method for establishing an artificial intelligence-based NIHSS scoring and diagnostic system, characterized by: The method comprises the following steps: S1. Collect the patient's voice, facial image, limb movement and eye movement data through the testing device; S2. parse the voice response using the voice recognition module and match it with the standard answer; S3, extracting facial expressions, pupil size, limb movement amplitude and coordination features through image processing module; S4. Based on the NIHSS scoring rules, the characteristic data is mapped into scores for each item and the total score is calculated; S5. Output the scoring results and the corresponding neurological impairment severity grade.

6. The method for establishing an artificial intelligence-based NIHSS scoring and diagnosis system according to claim 5, wherein: In step S3, the limb movement amplitude is calculated by the change of joint angles in continuous video frames, and the difference between the initial posture and the maximum lifting posture is compared; the coordination is evaluated by the dynamic error of the distance between the finger and the nose tip in the finger-to-nose test.

7. The method for establishing an artificial intelligence-based NIHSS scoring and diagnosis system according to claim 6, wherein: In step S4, the visual field item is scored based on the patient's visible range of the dynamic red grid, and the visual field defect level is determined by the preset normal visual field angles (60° nasal, 100° temporal, 60° above, and 75° below).

Images