CN219439095U - Intelligent diagnosis equipment for early nerve function evaluation of infants - Google Patents

Abstract

The utility model provides intelligent diagnosis equipment for early-stage nerve function evaluation of infants, which mainly comprises a control main board, an image acquisition unit, an action sensing unit, an acoustic acquisition unit, data analysis software and a human-computer interaction interface. The infant early neural development evaluation equipment has the beneficial effects that the infant early neural development evaluation equipment with multi-angle monitoring and multi-parameter diagnosis is provided, the defects existing in the existing sensory monitoring and scale evaluation are avoided, the blank in the technical field is filled, and the infant early neural development evaluation equipment has good clinical significance and use value.

Description

Intelligent diagnosis equipment for early nerve function evaluation of infants

Technical Field

The utility model relates to intelligent diagnosis equipment for early nerve function evaluation of infants, which is used for medical institutions to screen early nerve development states based on spontaneous behaviors of infants and belongs to the category of medical appliances.

Background

Early identification of the risk of neurological or cognitive dysfunction in infants remains a challenge in the medical field due to the immature brain development of infants and based on safety considerations in infant imaging diagnostics. Moreover, there is still little clear association between brain structural changes and cognitive deficits, and no clear association between pathological results based on neuroimaging scans and cognitive functions has been detected at present. Spontaneous movement and free sounding are spontaneous behaviors of infants in early stages, and modern medical research finds that abnormal spontaneous movement, abnormal sounding and the like are one of the most reliable early signs of the infant neural development disorder. Spontaneous movement of an infant refers to movement under conditions not stimulated or induced by the outside, including trunk movement (twisting), neck rotation, limb movement, etc. From early fetal stage to the end of the second month after term, a turning motion (twisting motion) around the limb axis begins to occur. At 6-9 weeks after delivery, the twisting motion gradually disappears, the dysphoria motion appears, the dysphoria motion is characterized by tiny motions of the neck, the trunk and the limbs in all directions, the acceleration is variable, and the motions of the limbs are coordinated. For example, in spontaneous movement of infants, if the movements are monotonous, there is a lack of variability and complexity; stiff limb movements and lack of fluency in movements; also, the limb and trunk muscles shrink or relax almost simultaneously, and the intensity, speed and extent of spontaneous movement lack normal variability, etc., which are abnormal spontaneous movement characteristics, indicating that the infant may develop serious neurological dysfunction such as Cerebral Palsy (CP). At the same time, monotonous mechanical stiff movements and abnormal motor posture at 3-5 months in infants beyond term age also indicate a high risk of subsequent cognitive dysfunction, such as autism. At about 4-5 months of age, the infant starts exploring the full potential of his/her vocal organs, vocalizing, screaming, crying and more complex yawing become through early atypical features, these acoustic parameters (frequency or duration) being useful for identifying infants with Autism Spectrum Disorder (ASD) or rett syndrome (RTT).

Therefore, the research on specific indexes such as spontaneous movement, sounding and the like of infants can become an important sensitive index for early recognition of nerve dysfunction or cognitive dysfunction, and provides a new diagnostic technical direction for medicine. However, the existing analysis of spontaneous movement and acoustic characteristics of infants mainly depends on the sensory and scale modes of medical staff, and in the evaluation process, not only trained professional evaluation staff are needed, but also continuous observation and recording are needed, so that the defects of high subjectivity of evaluation results, easy fatigue of observers and the like exist. Therefore, the utility model provides an evaluation tool for providing continuous, objective and quantitative evaluation based on an artificial intelligent recognition technology.

Disclosure of Invention

The utility model provides intelligent diagnosis equipment for early-stage nerve function evaluation of infants, which mainly comprises a control main board, an image acquisition unit, a motion sensing unit, an acoustic acquisition unit, data analysis software and a man-machine interaction interface, wherein one or more parameters related to spontaneous movement and acoustic characteristics of infants are acquired, the data analysis software is used for finding out specific indexes, and the intelligent diagnosis equipment is used for screening and diagnosing early-stage nerve function diseases of infants, and suitable early-stage nerve function diseases comprise Cerebral Palsy (CP), autism Spectrum Disorder (ASD) and Rate syndrome (RTT).

The control main board is an integrated circuit constructed based on the core processor and comprises the core processor, a power management module, a storage module, a communication module and the like. Wherein, the core processor usually adopts any one of a Central Processing Unit (CPU), a single chip Microcomputer (MCU) and a Programmable Logic Controller (PLC); the storage module comprises any one of an SD card, a TF card or a computer disk; the communication module is used for communicating with an upper computer and sending monitoring data or diagnosis results, for example, to a medical terminal, a medical staff handheld terminal or a patient family mobile phone and the like. The communication module adopts traditional Bluetooth, wiFi, zigBee or RF modes and the like without limitation. The memory module of the control main board is also provided with embedded software which is mainly used for hardware driving, program control, information output and the like.

The image acquisition unit is communicated with the control main board, and mainly comprises a digital video camera and a photographic light source, wherein the digital video camera and the photographic light source are arranged above the test table, and an effective acquisition area covers the infant test table. In the process of evaluating spontaneous movement of the infant, the infant is horizontally placed on the test bench, and the image acquisition unit is used for completely recording limb movements, body movements and facial expressions of the infant to form a video file and transmitting the video file to the storage module of the control main board for storage.

The motion sensing unit mainly comprises an inertial sensor and a communication module, and is connected and communicated with the control main board and used for collecting spontaneous movement track signals of infants. The inertial sensor adopts a six-degree-of-freedom sensor comprising a 3-axis accelerometer and a 3-axis gyroscope and an electromagnetic motion tracking sensor; the communication module comprises Bluetooth, wifi and the like. In the evaluating process, the motion sensing unit is prepared as a wearable sensing element and is attached to or bound and fixed on a monitored part of an infant, for example, the motion sensing unit is fixed on wrist joints of left and right arms and ankle joints or soles and forehead leaves of left and right lower legs of the infant, the motion track signals collected by the motion sensing unit comprise head rotation tracks, limb motion tracks and torsion motion tracks, and the motion track signals are stored in a storage module of a Bluetooth or WiFi transmission control main board.

The acoustic acquisition unit mainly comprises a pickup and an audio signal processor, and is communicated with the control main board. The acoustic acquisition unit is arranged above or around the test bench, and in the evaluating process, the acoustic acquisition unit carries out complete recording on acoustic characteristics of the infant, including the occurrence of the miking, the laughing and the crying, forms an audio file and transmits the audio file to the storage module of the control main board for storage.

The data analysis software is installed in the storage module of the control main board, and classifies the video files acquired by the image acquisition unit, the motion track signals acquired by the motion sensing unit and the audio files acquired by the acoustic acquisition unit respectively. The data analysis software divides the video files into five categories of facial features, head movements, upper limb movements, trunk movements and lower limb movements according to infant moods and movement tracks; according to the motion track of the infant, dividing the motion track signals into four categories of head motion, upper limb motion, trunk motion and lower limb motion; audio file tracks are classified into two major categories, speech-like and non-speech-like, according to the sound characteristics of the infant.

The data analysis software also comprises an artificial intelligence (artificial intelligence, AI) analysis unit, wherein the artificial intelligence analysis unit is a pre-trained neural network algorithm model, and can independently and intelligently analyze classified video files, motion track signals and audio files respectively and automatically give out classification diagnosis results. The neural network algorithm model comprises a three-dimensional feature extraction model, a position detection model, a feature segmentation model, a classification model and other models, wherein the models are obtained through training in advance and have corresponding specific functions.

The artificial intelligent analysis unit also carries out multidimensional comprehensive analysis on the video file, the motion trail signal and the audio file, automatically gives out comprehensive diagnosis results, and improves the working efficiency of medical staff.

For example, the image acquisition unit acquires a video file of spontaneous movement of the baby, and the artificial intelligence analysis unit automatically gives a classification diagnosis result according to the video file. The artificial intelligent analysis unit adopts a neural network model which is obtained through training in advance, and extracts the space-time characteristics of infant emotion and motion track from 1000 healthy infant video files acquired by the image acquisition unit through a three-dimensional characteristic extraction model based on a three-dimensional (3D) convolution neural network model (3D ConvNet,3D CNN,C3D). According to the space-time characteristics of the video, classifying and marking actions in the video, including emotion classification marks of facial characteristics such as laughing, crying, slow-down, surprise and the like of infants. The infant head movement, upper limb movement, trunk movement and lower limb movement classification marks specifically comprise the steps of identifying the positions of the head, neck, shoulders, trunk, arms and legs and marking the positions of the head, neck, left shoulder, right shoulder, trunk, left elbow joint, right elbow joint, left palm, right palm, crotch, left leg knee joint, right leg knee joint and left sole/right sole respectively. Secondly, recovering the complete posture of the infant in a skeleton model form, classifying according to the motion characteristics of the skeleton, extracting typical actions after identifying the typical actions, dividing the typical actions into single-frame images, identifying and marking the single-frame images, and establishing an action identification neural network model. After the action recognition neural network model is built, automatically classifying and recognizing video files input in the spontaneous movement process of the infant, including whether the expression is stiff, whether the limb actions are asymmetrical, uncoordinated and the like, observing and judging whether the twisting action of the infant is natural and smooth, whether the head rotation, the neck rotation, the limb movements are coordinated and the like, carrying out limb movement coordination analysis of the infant, giving a result prompt for medical staff to refer to, and judging whether the infant has nerve dysfunction such as Cerebral Palsy (CP).

For example, the motion sensing unit acquires a motion trail signal of spontaneous movement of the baby, and the artificial intelligence analysis unit automatically gives a classification diagnosis result according to the motion trail signal. The motion sensing unit can sense unobtrusive motion changes of the head or limbs of the infant, including head motion, upper limb motion, trunk motion and lower limb motion, and record motion data in real time, so that the defect that the image acquisition unit acquires unobtrusive motion changes is overcome. The artificial intelligent analysis unit is combined with the acquired video file and the motion trail signal to more comprehensively evaluate the spontaneous motion state of the infant, analyze the coordination of the limb motion of the infant and give out a comprehensive analysis result.

The artificial intelligent analysis unit also includes an audio file into the comprehensive analysis element based on the comprehensive analysis and diagnosis of the video file and the motion trail signal. Early vocalization is a sensitive indicator of the neurological state of infants, and preschool children with Autism Spectrum Disorders (ASD) produce higher levels of non-linguistic, atypical vocalization than those of the same age with typical development. The artificial intelligence analysis unit evaluates the acquired audio files, including the baby's scale, consonant-vowel syllables with speech quality, and closed syllables (consonant-vowel-consonant), while evaluating the baby's spontaneous movement. If the occurrence of consonants and normative syllables in the audio file of the infant is small and more nonverbal sounding behaviors exist, the possibility of the autism symptoms is higher in the future. The comprehensive analysis of the video file and the motion trail signal combined with the audio file directly and in detail analyzes spontaneous movement and sound rows of infants at risk of ASD, provides more abundant information than single-dimension evaluation, and improves the reliability of diagnosis.

The man-machine interaction interface mainly comprises a display, operation function keys and a protective shell, and is communicated with the control main board. The man-machine interaction interface is mainly used for realizing other auxiliary functions such as on/off operation, function setting, parameter setting, information reading and printing.

The intelligent diagnosis equipment for early-stage nerve function evaluation of infants has the beneficial effects that the intelligent diagnosis equipment for early-stage nerve development of infants, which is used for multi-angle monitoring and multi-parameter diagnosis, avoids the defects of the existing sensory monitoring and scale evaluation, fills the blank in the technical field, and has good clinical significance and use value.

Drawings

Fig. 1 is a functional block diagram of an embodiment of the present disclosure.

Fig. 2 is an external structural schematic diagram of an embodiment of the present disclosure.

The figure shows: the system comprises an image acquisition unit (1), an action sensing unit (2), an acoustic acquisition unit (3), a human-computer interaction interface (4) and a test board (5).

Detailed Description

The present utility model will be specifically described below with reference to the drawings and examples.

Example 1: artificial intelligent analysis based on video files acquired by image acquisition unit

1. Video file acquisition

According to standard requirements, collecting video images of spontaneous movement of infants with age of 6 months, wherein the number of samples is not less than 1000, and the specific standard requirements are as follows:

(1) The infant lies on the back, is fully awake, has quiet emotion, and has no agitation, crying or frightening;

(2) The indoor temperature is proper, the temperature is kept at about 26 ℃, and the surrounding environment is quiet and is in an undisturbed state;

(3) The baby wears the baby to be thin, the body can freely move, the limbs and joints are exposed, and the baby can be clearly seen by naked eyes;

(4) The RGB camera is used to capture images, the video length is 10 minutes, and at least 3 consecutive minutes of video should contain limb movements.

2. Video file marking

(1) After the video file is encoded, each video is divided into video segments with equal length, and each segment is 10 seconds;

(2) Using gesture estimation algorithm software AlphaPose (pre-training gesture estimation model developed based on convolutional neural network architecture), position coordinates of twelve joints are automatically extracted from the acquired spontaneous motion video, spatial-temporal data of the infant's head, trunk, arms and legs are acquired, including the head, shoulder, elbow, wrist, hip, knee and ankle marked on the left and right, position coordinates of each joint are generated in each frame, confidence level is from 0 to 1, and confidence level is <0.5.

3. Algorithm model training

The segmented coded video segments are marked manually by a professional medical staff to evaluate coordination of spontaneous movement. Scoring head, torso, arm and leg motor coordination, respectively, for example: coordination, +1 points; uncoordinated, -1 score; and 0 point cannot be judged.

Then, the extracted skeleton point coordinates are input into a neural network for training, and a skeleton point coordinate sequence of the whole body of the baby in the video segment is extracted through a deep learning algorithm.

After the algorithm model is established, in the clinical diagnosis process, only a trained neural network is input every other multiple times in the collected infant video file, and automatic judgment results of which section of video is uncoordinated in motion, whether the specific mark is uncoordinated in upper limb motion or lower limb motion and the like can be automatically output.

Example 2: acquisition by using motion sensing unit

1. Data acquisition

And 6 motion sensing units are prepared by adopting a triaxial accelerometer, a gyroscope and Bluetooth as main components, wherein the volume is not more than 60mm and 40mm and 20mm, and the weight of each sensing unit is not more than 50g. The infant training device is respectively fixed at forehead leaves, left/right wrists, left/right ankles and navel positions of an infant, and is used for collecting movement track signals of spontaneous movement of the infant in 30 minutes, wherein the sample size is not lower than 500 groups. In the calm and supine state of the infant, taking a gyroscope at the navel position as a reference coordinate, dynamically calculating motion parameters (x, y and z) by a triaxial accelerometer at each part, and synthesizing linear acceleration, wherein a specific calculation formula is as follows:

the median value is subtracted to remove the gravitational component and any stationary noise or offset in the embedded signal, thereby eliminating the tendency of the resultant acceleration.

2. Algorithm performance verification

And respectively calculating the movement frequencies of the head, the left/right arm and the left/right leg according to the movement track data acquired by the movement sensing units distributed at each position. Then, using the synchronous video file, the professional observer compares the motion frequency data formed by the algorithm with the synchronous video file, records the motion times of each part, compares the motion frequency with the motion frequency calculated by the motion track data acquired by the motion sensing unit, verifies the reliability of the algorithm, and the accuracy of motion recognition is not lower than 95%.

Example 3: artificial intelligence analysis of audio files based on acquisition of acoustic acquisition units

1. Acquisition and storage of audio files: from a population of infants with a history of premature infants or high risk of brain injury for 6 months to 12 months, an acoustic acquisition unit is used to collect samples of audio files taken by infants from timed parent-child interactions for 10 minutes with the child playing and interaction time, and the microphone of the acoustic acquisition unit is placed close to the child's mouth, preferably with a headset microphone. Each person selects one section of the audio file with rich acoustic characteristics, each section is not lower than 5 minutes, and the number of samples of the obtained audio file is not lower than 5000 sections.

2. Classification: the audio files of infants are divided into two categories, speech-like and non-speech. The class of similar voices includes sounds characterized by the generation of consonants and/or vowels, which may be represented by a note number and contain the sound quality of the similar voices, so they resemble typical yawing sounds. Non-speech categories include utterances characterized by non-speech resonances and sound quality (e.g., screaming, shouting, growling) without identifiable consonants.

3. Marking: the audio files are randomly divided into three groups and two trained medical scoring persons by taking 6 months, 9 months and 12 months as age limits, all classified audio files are marked and encoded, and the point-to-point reliability of all encoding categories is not lower than 95%.

4. And (3) algorithm establishment: and extracting a set of acoustic parameters from each marked and coded segmented sounding by adopting openSMILE software, including sounding speed and sounding frequency (such as arithmetic mean, standard deviation, high-order moment, spectral band energy, harmonic noise ratio and the like), and establishing an acoustic feature library.

The development progress of 5000 segments of encoded audio files was followed to 30 months of age, and the encoded audio files finally diagnosed as Autism Spectrum Disorder (ASD) were named positive (+) and the encoded audio files of healthy infants were named negative (-) using the autism infant examination scale MCHAT diagnosis. The typical sound (producing consonants and/or vowels characterized by) of the infant diagnosed as positive (+) is significantly reduced compared to the negative (-) infant, the lower overall sound volume of the infant diagnosed as positive (+) is significantly reduced compared to the negative (-) infant, and the sound production of non-speech features such as screaming, shouting, growling, etc. of the infant diagnosed as positive (+) is significantly increased compared to the negative (-) infant. According to different acoustic parameter characteristics of positive (+) infants and negative (-) infants in the acoustic characteristic library, characteristic extraction is respectively carried out, an artificial intelligent analysis algorithm normal model is established after normalization based on a template method and a probability statistics method in an artificial intelligent algorithm, and an artificial intelligent identification diagnosis result is obtained by matching an infant audio file sample to be identified with a known template.

While the utility model has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that the foregoing drawings and embodiments are illustrative only and not restrictive, and that the utility model may be modified and practiced without departing from the spirit and scope of the appended claims.

Claims (5)

1. An intelligent diagnosis device for early-stage nerve function evaluation of infants mainly comprises a control main board, an image acquisition unit (1), an action sensing unit (2), an acoustic acquisition unit (3), data analysis software and a human-computer interaction interface (4), and is characterized in that: the image acquisition unit (1) is communicated with the control main board, the image acquisition unit (1) mainly comprises a digital video camera and a photographic light source, the digital video camera and the photographic light source are arranged above the test table (5), and an effective acquisition area covers the infant test table (5); the motion sensing unit (2) mainly comprises an inertial sensor and a communication module, and the motion sensing unit (2) is connected and communicated with the control main board; the acoustic acquisition unit (3) mainly comprises a pickup and an audio signal processor, and the acoustic acquisition unit (3) is communicated with the control main board; the data analysis software is installed in a storage module of the control main board; the man-machine interaction interface (4) mainly comprises a display, operation function keys and a protective shell.

2. An intelligent diagnostic apparatus for early childhood neurological assessment of claim 1, further characterized by: the inertial sensor adopts a six-degree-of-freedom sensor comprising a 3-axis accelerometer and a 3-axis gyroscope and an electromagnetic motion tracking sensor.

3. An intelligent diagnostic apparatus for early childhood neurological assessment of claim 1, further characterized by: the motion sensing unit (2) is fixed on the wrist joints of the left and right arms and the ankle joints or the soles and forehead leaves of the left and right lower legs of the infant, and the motion track signals collected by the motion sensing unit (2) comprise a head rotation track, an extremity motion track and a torsion motion track.

4. An intelligent diagnostic apparatus for early childhood neurological assessment of claim 1, further characterized by: the data analysis software divides the video files into five categories of facial features, head movements, upper limb movements, trunk movements and lower limb movements according to infant moods and movement tracks; according to the motion track of the infant, dividing the motion track signals into four categories of head motion, upper limb motion, trunk motion and lower limb motion; audio file tracks are classified into two major categories, speech-like and non-speech-like, according to the sound characteristics of the infant.

5. An intelligent diagnostic apparatus for early childhood neurological assessment of claim 1, further characterized by: the motion sensing unit (2) is prepared as a wearable sensing element and is attached to or bound and fixed on a monitored part of the infant.