Personalized anti-aspiration feeding method based on multi-modal sensor and intelligent algorithm

Abstract

The invention relates to a personalized anti-aspiration feeding method and device based on a multi-modal sensor and an intelligent algorithm. The method comprises the steps that biological information of a patient is collected through a multi-modal biosensor matrix; inputting the biological information of the patient into a self-adaptive feeding model to generate a personalized feeding scheme; the food backflow condition is detected, and feeding is conducted by controlling the double-channel feeding pipe and the spiral extrusion assembly according to a personalized feeding scheme; the spiral extrusion assembly drives a screw to rotate through a servo motor to achieve quantitative feeding, and the extrusion speed and direction are adjusted. According to feeding of the double-channel feeding pipe and the spiral extrusion assembly, a food flowing path, a mistaken inhalation risk area and a corresponding mistaken inhalation risk score are generated; and when the false inspiration risk score exceeds a plurality of preset thresholds, different levels of early warning and intervention measures are triggered respectively. According to the method, the mistaken inspiration risk score is calculated through the multi-modal sensor data, the personalized feeding scheme is dynamically generated, and the feeding safety and the rehabilitation efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical device auxiliary, in particular to a personalized anti-aspiration feeding method and device based on multi-modal sensors and intelligent algorithms, and a computer device. BACKGROUND

[0002] Aspiration pneumonia is a common complication in clinical practice, especially in patients with neurological diseases such as stroke, Parkinson's disease, and Alzheimer's disease. The incidence rate is relatively high. The traditional feeding method can only observe the swallowing action of the patient, and cannot comprehensively assess the risk of aspiration. For example, the feeding parameters such as flow rate and consistency are usually based on fixed standards, and are not dynamically adjusted according to the physiological characteristics of the patient. In addition, measures are taken only after aspiration occurs, and real-time prevention is not possible. The cleaning of the pipeline relies on manual operation, which may lead to the risk of cross infection.

[0003] To solve the above problems, the present application discloses a personalized anti-aspiration feeding method based on multi-modal sensors and intelligent algorithms, which calculates the aspiration risk score through multi-modal sensor data and dynamically generates a personalized feeding plan to improve feeding safety and rehabilitation efficiency. SUMMARY

[0004] The present application provides a personalized anti-aspiration feeding method and device based on multi-modal sensors and intelligent algorithms, and a computer device to solve the problems of the prior art.

[0005] The technical solution of the present application to solve the above technical problems is as follows: a personalized anti-aspiration feeding method based on multi-modal sensors and intelligent algorithms, comprising:

[0006] A multi-modal biosensor matrix is used to collect biological information of the patient, wherein the multi-modal biosensor matrix includes surface electromyography sensors, infrared pyroelectric sensors, swallowing sound signal sensors, tongue pressure sensors, and blood oxygen saturation sensors;

[0007] The biological information of the patient is input into an adaptive feeding model to generate a personalized feeding plan, and the adaptive feeding model is matched with a videofluoroscopic swallowing study (VFSS) data through a cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model uses a hierarchical LSTM neural network model;

[0008] A micro pressure sensor arranged near the oral cavity is used to detect the backflow of food, and the feeding is performed according to the personalized feeding plan by controlling a double-channel feeding tube and a spiral extrusion assembly, wherein the double-channel feeding tube controls the shunt ratio through an electromagnetic valve, and includes a food channel and a liquid channel; the spiral extrusion assembly drives the screw to rotate through a servo motor to realize quantitative feeding, and adjusts the extrusion speed and direction to adapt to different food consistency and feeding angle requirements;

[0009] According to the double-channel feeding tube and the spiral extrusion assembly, a food flow path, an aspiration risk area, and a corresponding aspiration risk score are generated;

[0010] When the aspiration risk score exceeds a plurality of preset thresholds, different levels of early warning and intervention measures are triggered respectively, wherein the intervention measures include slowing down the feeding speed, adjusting the food consistency, and pausing the feeding.

[0011] In an optional manner, the surface electromyography sensor is arranged between the submental muscle group and the thyrohyoid muscle, used to collect the swallowing electromyography signal and extract the energy proportion in the 50-150Hz frequency band as the electromyography feature;

[0012] The infrared pyroelectric sensor is used to detect the respiratory cycle to determine the inspiration and expiration phases;

[0013] The swallowing sound signal sensor is arranged at the suprasternal fossa of the neck, used to collect the swallowing sound signal and perform time-frequency analysis to extract the mel-frequency cepstral coefficient;

[0014] The tongue pressure sensor is used to measure the pressure of the tongue on the food to obtain the tongue movement control ability data;

[0015] The blood oxygen saturation sensor is used to monitor the real-time change of the blood oxygen level of the patient.

[0016] In an optional manner, the first layer LSTM of the hierarchical LSTM neural network model is used to process the historical feeding data and the biosensor data;

[0017] The second layer LSTM of the hierarchical LSTM neural network model is used to generate the optimal flow rate, food consistency, feeding interval, food granularity, and feeding angle according to the patient's subjective feedback and the expert rule base.

[0018] In an optional manner, the method further comprises:

[0019] When food reflux is detected, the feeding is stopped to prevent regurgitation aspiration;

[0020] When the blood oxygen saturation is lower than 95% or the blood oxygen saturation rapidly decreases within a preset time, an emergency suction channel is started to connect a negative pressure pump for negative pressure suction, wherein the negative pressure suction force is adaptively adjusted according to the aspiration degree and the suction effect is monitored by a pressure sensor.

[0021] In an optional manner, after the personalized feeding scheme is generated, the method further comprises:

[0022] The cheek muscles are massaged for 3-5 minutes by the vibration massage device at a preset massage frequency, massage duration and amplitude to stimulate facial sensory nerves; wherein the preset massage frequency, massage duration and amplitude are personalized adjusted by the adaptive feeding model according to VFSS data and biological feedback information of the patient;

[0023] The ice stick or ice water cotton swab is placed at the root of the tongue for 1-2 seconds by the ice stimulation device to activate the swallowing reflex.

[0024] In an optional mode, the calculation formula of the aspiration risk score is:

[0025]

[0026] Among them, is the energy proportion of the 50-150 Hz frequency band of the surface electromyogram standardized value; is the dynamic change rate of the mel frequency cepstral coefficient MFCC of the swallowing sound signal; is the time domain complexity of the pressure collected by the tongue pressure sensor; is the blood oxygen saturation drop rate; is the reflux velocity; is the time; is the weight coefficient.

[0027] In an optional mode, the calculation formula of the dynamic change rate of the mel frequency cepstral coefficient MFCC of the swallowing sound signal is:

[0028]

[0029] Among them, is the step function; is the abnormal voiceprint threshold; is the i th mel frequency cepstral coefficient; is the total number of coefficients in the MFCC vector;

[0030] The weight coefficient The constraint optimization solution expression of is:

[0031]

[0032] Among them, is the clinical gold standard score; is the L2 regularization coefficient; is the predicted value of aspiration risk.

[0033] In an optional mode, the double-channel feeding tube built-in positive pressure pulse flushing module;

[0034] The positive pressure pulse flushing module adopts a three-stage cleaning procedure composed of cleaning agent injection, vortex oscillation and sterile water flushing; and the frequency optimization formula of the vortex oscillation is:

[0035]

[0036] wherein, is the pipe wall shear force; is the fluid kinetic energy; is the energy efficiency balance coefficient; is the optimal frequency of vortex oscillation; is the frequency;

[0037] The double-channel feeding tube is automatically wound on the built-in UV-C LED disinfection roller when the pipeline of the double-channel feeding tube is retracted, and when the biological fluorescence value of the ATP fluorescence detection point arranged at the inlet of the pipeline of the double-channel feeding tube exceeds 50RLU, a strengthened disinfection program is started and a replacement reminder is pushed to the medical staff; and a one-way duckbill valve is arranged at the connection between the food channel and the liquid channel to prevent the cleaning liquid from flowing back and contaminating the servo motor, and a humidity sensor is arranged to monitor the dryness of the pipeline.

[0038] According to another aspect of the present application, a personalized anti-aspiration feeding device based on a multi-modal sensor and an intelligent algorithm is provided, comprising:

[0039] A data acquisition module is configured to acquire biological information of a patient through a multi-modal biological sensor matrix, wherein the multi-modal biological sensor matrix comprises a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor and a blood oxygen saturation sensor;

[0040] A scheme generation module is configured to input the biological information of the patient into an adaptive feeding model to generate a personalized feeding scheme, and the adaptive feeding model is matched with a videofluoroscopy (VFSS) data through a cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model adopts a hierarchical LSTM neural network model;

[0041] A precise feeding module is configured to detect food backflow through a micro pressure sensor arranged near the oral cavity, and to feed through a double-channel feeding tube and a spiral extrusion assembly according to the personalized feeding scheme, wherein the double-channel feeding tube controls the shunt ratio through an electromagnetic valve, and comprises a food channel and a liquid channel; the spiral extrusion assembly drives the rotation of a screw rod through a servo motor to realize quantitative feeding, and adjusts the extrusion speed and direction to adapt to different food consistency and feeding angle requirements;

[0042] A risk assessment module is configured to generate a food flow path, an aspiration risk area and a corresponding aspiration risk score according to the feeding of the double-channel feeding tube and the spiral extrusion assembly;

[0043] The early warning intervention module is configured to trigger different levels of early warning and intervention measures when the aspiration risk score exceeds a plurality of preset thresholds, wherein the intervention measures include slowing down the feeding speed, adjusting the food consistency, and pausing the feeding.

[0044] According to another aspect of the present application, a computer device is provided, comprising a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface being in communication with each other through the communication bus;

[0045] The memory is configured to store at least one executable instruction, and the executable instruction causes the processor to perform the operations corresponding to the above-mentioned individualized anti-aspiration feeding method based on the multi-modal sensor and intelligent algorithm.

[0046] According to the scheme provided by the present application, the biological information of the patient is collected through a multi-modal biological sensor matrix, wherein the multi-modal biological sensor matrix includes a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor and a blood oxygen saturation sensor; the biological information of the patient is input into an adaptive feeding model to generate an individualized feeding scheme, and the adaptive feeding model is matched with a swallowing radiography VFSS data through a cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model adopts a hierarchical LSTM neural network model; a micro pressure sensor arranged near the oral cavity is used to detect the food reflux condition, and the individualized feeding scheme is used to feed through the control of a double-channel feeding tube and a screw extrusion assembly, wherein the double-channel feeding tube controls the shunt ratio through an electromagnetic valve, and includes a food channel and a liquid medicine channel; the screw extrusion assembly drives the rotation of the screw rod through a servo motor to realize quantitative feeding, and adjusts the extrusion speed and direction to adapt to different food consistency and feeding angle requirements; a food flow path, an aspiration risk area and a corresponding aspiration risk score are generated according to the feeding of the double-channel feeding tube and the screw extrusion assembly; when the aspiration risk score exceeds a plurality of preset thresholds, different levels of early warning and intervention measures are triggered, wherein the intervention measures include slowing down the feeding speed, adjusting the food consistency and pausing the feeding. The present application calculates the aspiration risk score through the multi-modal sensor data and dynamically generates an individualized feeding scheme, thereby improving the feeding safety and rehabilitation efficiency.

[0047] The above description is only a summary of the technical scheme of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a better understanding of the preferred embodiments, and are not intended to be a limitation of the application. Moreover, in the drawings, like reference numerals denote same or similar components. In the drawings:

[0049] Figure 1 A flowchart of a personalized anti-aspiration feeding method based on a multi-modal sensor and an intelligent algorithm is shown.

[0050] Figure 2 A frame diagram of a personalized anti-aspiration feeding device based on a multi-modal sensor and an intelligent algorithm is shown.

[0051] Figure 3 A structure diagram of a computer device is shown. DETAILED DESCRIPTION

[0052] Exemplary embodiments of the present application will be described herein below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thoroughly and completely understood, and so that the scope of the present application will be completely conveyed to those skilled in the art.

[0053] Figure 1 A flowchart of a personalized anti-aspiration feeding method based on a multi-modal sensor and an intelligent algorithm is shown. Specifically, as shown in Figure 1 the following steps are included:

[0054] In step S101, biological information of a patient is collected by a multi-modal biosensor matrix, wherein the multi-modal biosensor matrix includes a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor, and an oxygen saturation sensor.

[0055] In this embodiment, the multi-modal sensor can capture various abnormal signals during swallowing, such as abnormal electromyography signals, changes in swallowing sound signals, and decreased tongue pressure, as early warning indicators of aspiration risk. The feeding regimen can be adjusted according to the patient's immediate state, such as adjusting the feeding speed according to the electromyography signal changes, or triggering emergency measures according to the decrease in blood oxygen saturation, thereby improving the adaptability of the feeding process. In this embodiment, the surface electromyography sensor is arranged between the submental muscle group and the thyrohyoid muscle, which is used to collect swallowing electromyography signals and extract the energy ratio of the 50-150Hz frequency band as electromyography features; the infrared pyroelectric sensor is used to detect the respiratory cycle to determine the inspiration and expiration phases; the swallowing sound signal sensor is arranged at the suprasternal fossa of the neck, which is used to collect swallowing sound signals and perform time-frequency analysis to extract mel-frequency cepstrum coefficients; the tongue pressure sensor is used to measure the pressure of the tongue on the food to obtain tongue movement control ability data; the blood oxygen saturation sensor is used to monitor the real-time changes of the patient's blood oxygen level.

[0056] In step S102, the biological information of the patient is input into the adaptive feeding model to generate a personalized feeding regimen, and the adaptive feeding model is matched with the VFSS data through the cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model adopts a hierarchical LSTM neural network model.

[0057] In this embodiment, the first layer LSTM processes the original sensor data and historical feeding data, and the second layer LSTM generates optimal feeding parameters in combination with an expert rule base (such as a swallowing flow rate threshold). The hierarchical LSTM model can simultaneously process time series data (such as historical feeding records and real-time biological signals) and dynamically adjust feeding parameters (flow rate, consistency, angle, etc.), which is more suitable for the state changes of the patient.

[0058] In an alternative way, after generating the personalized feeding regimen, the method further comprises:

[0059] The facial cheek muscles are massaged for 3-5 minutes by the vibration massage device at a preset massage frequency, massage duration, and amplitude to stimulate the facial sensory nerves; wherein the preset massage frequency, massage duration, and amplitude are individually adjusted by the adaptive feeding model according to the VFSS data and biological feedback information of the patient;

[0060] The ice stick or ice water cotton swab is placed at the base of the tongue for 1-2 seconds by the ice stimulation device to activate the swallowing reflex.

[0061] In this embodiment, the facial sensory nerves and swallowing reflexes are activated through dual pathways of mechanical vibration (vibration massage) and cold stimulation (ice stimulation), which can better cover the differences in neural responses of different patients than a single stimulation method, and is particularly suitable for swallowing disorders caused by central nervous system damage (such as stroke). The massage frequency, duration and amplitude are dynamically adjusted by the adaptive feeding model based on the patient's VFSS (swallowing radiography) data and real-time biofeedback (such as electromyographic signals, tongue pressure) to avoid discomfort or ineffective stimulation that may be caused by general parameters. For example, VFSS shows that the swallowing delay time is extended to 2.5 seconds, and the blood oxygen saturation is Descent rate The vibration massage was performed at a frequency of 5Hz, an amplitude of 1mm, and a duration of 3 minutes (to prevent muscle stiffness in Parkinson's patients). It was triggered once before and after each feeding, with the cotton swab in ice water in contact for 1 second. The swallowing delay time was shortened to 1.8 seconds, and the aspiration risk score R was reduced from 0.7 to 0.3.

[0062] In step S103, food reflux is detected by a micro pressure sensor located near the oral cavity, and feeding is performed according to the personalized feeding plan by controlling a dual-channel feeding tube and a spiral extrusion assembly. The dual-channel feeding tube includes a food channel and a liquid medicine channel, and a solenoid valve controls the diversion ratio. The spiral extrusion assembly achieves quantitative feeding by driving the screw to rotate via a servo motor, and adjusts the extrusion speed and direction to accommodate different food consistencies and feeding angles.

[0063] In this embodiment, the diversion ratio of the food channel and the liquid medicine channel is controlled by a solenoid valve, which can achieve precise feeding and avoid food blockage or overfeeding that leads to aspiration. The servo motor drives the screw to rotate to achieve quantitative feeding and improve feeding safety by adjusting the extrusion speed and direction to adapt to different food consistencies (such as liquid, semi-solid, solid) and feeding angles (such as vertical, inclined). A micro pressure sensor is set near the mouth to detect food backflow in real time. Once backflow is detected, feeding is automatically stopped to prevent reflux aspiration. The feeding speed, angle and consistency can be dynamically adjusted according to the personalized feeding plan to adapt to the swallowing ability of different patients. The dual-channel feeding tube and spiral extrusion component are integrated into a set of equipment, which is easy to operate and suitable for hospitals, nursing institutions or home care scenarios.

[0064] In an optional manner, the dual-channel feeding tube has a built-in positive pressure pulse flushing module;

[0065] The positive pressure pulse flushing module adopts a three-stage cleaning procedure consisting of detergent injection, vortex oscillation and sterile water flushing; the frequency optimization formula of the vortex oscillation is:

[0066]

[0067] in, Shear force for tube wall; Kinetic energy for fluid; Energy efficiency balance coefficient; Optimal frequency of vortex oscillation; Frequency;

[0068] The tube of the double-channel feeding tube is automatically wound on the built-in UV-C LED disinfection roller when it is retracted. When the biological fluorescence value of the ATP fluorescence detection point set at the inlet of the tube of the double-channel feeding tube exceeds 50 RLU, the enhanced disinfection program is started and a replacement reminder is pushed to the medical staff. A one-way duckbill valve is arranged at the connection between the food channel and the liquid channel to prevent the backflow of cleaning liquid from contaminating the servo motor. A humidity sensor is provided to monitor the dryness of the tube.

[0069] In this embodiment, the combination of cleaning agent injection, vortex oscillation and sterile water flushing ensures thorough cleaning of the inner wall of the tube, avoiding the risk of infection caused by food residue accumulation. The energy efficiency balance frequency dynamically adjusts the oscillation frequency, maximizes the shear force to remove stubborn dirt, and controls the kinetic energy to save energy. The tube is automatically wound on the roller when it is retracted, and continuous disinfection is achieved through ultraviolet irradiation, reducing the risk of contamination caused by manual operation. Real-time monitoring of the biological fluorescence value at the inlet of the tube (> 50 RLU triggers enhanced disinfection) combined with reminders from the medical staff ensures that the disinfection effect is traceable. In addition, the food channel and the liquid channel are physically isolated to prevent the backflow of cleaning liquid from contaminating the servo motor, ensuring the safety of the drive system. The humidity sensor monitors the dryness of the tube to prevent bacteria from growing in a humid environment and prolong the service life of the tube.

[0070] Step S104: generating a food flow path, an aspiration risk area and a corresponding aspiration risk score according to the double-channel feeding tube and the spiral extrusion assembly.

[0071] In this embodiment, the food flow path and the aspiration risk area are dynamically generated in combination with the motion trajectory of the spiral extrusion assembly, improving the accuracy of aspiration warning.

[0072] In an alternative way, the calculation formula of the aspiration risk score is:

[0073]

[0074] Wherein, The surface electromyogram 50-150 Hz frequency band energy ratio standardization value; The dynamic change rate of the swallowing sound signal mel-frequency cepstral coefficient MFCC; The time domain complexity of the tongue pressure sensor collected pressure; The blood oxygen saturation rate of decline; The reflux velocity; The time; is a weight coefficient.

[0075] In this embodiment, the reflux trend is monitored in real time in combination with the reflux velocity integral term, and the early warning level is dynamically adjusted. For example, a patient has dysphagia due to stroke, the muscle electrical signal energy proportion = 0.6 (standardized value); the swallowing sound MFCC dynamic change rate = 0.8; the tongue pressure time domain complexity = 0.4; the blood oxygen decline rate = 0.3% / s; the reflux velocity = 2 mL / s. The weight coefficients are α = 0.3, β = 0.4, γ = 0.2, δ = 0.1, and η = 0.5, and R ≈ 0.85 (high risk). Trigger the pause feeding and push the alarm to the medical side, and suggest adjusting the food consistency or using the ice stimulation device to activate the swallowing reflex.

[0076] In an alternative way, the calculation formula of the dynamic change rate of the swallowing sound signal mel-frequency cepstrum coefficient MFCC is:

[0077]

[0078] wherein, is a step function; is an abnormal voiceprint threshold; is the i th mel-frequency cepstrum coefficient; is the total number of coefficients in the MFCC vector;

[0079] weight coefficient The constraint optimization solving expression of the weight coefficient is:

[0080]

[0081] wherein, is the clinical gold standard score; is the L2 regularization coefficient; is the predicted value of aspiration risk.

[0082] In this embodiment, the instantaneous change of the swallowing sound signal is captured by time domain difference, and the abnormal voiceprint (such as coughing and wheezing) is filtered in combination with the step function, thereby improving the accuracy of aspiration detection. Compared with the traditional static MFCC feature, the abnormal voiceprint (such as air flow sound caused by incomplete glottis closure) in the swallowing process is more sensitive. By minimizing the difference between the model predicted value and the clinical gold standard score, the model is ensured to be consistent with the clinical standard.

[0083] Step S105, when the aspiration risk score exceeds a plurality of preset threshold values, triggering different levels of early warning and intervention measures respectively, wherein the intervention measures include slowing down the feeding speed, adjusting the food consistency and pausing the feeding.

[0084] In this embodiment, the multi-threshold value can trigger targeted measures according to different levels of aspiration risk scores, avoiding "one-size-fits-all" interventions and reducing false positives or false negatives. For example: low risk (R < 0.3) only slows down the feeding speed (such as servo motor speed reduction of 20%). Medium risk (0.3 ≤ R < 0.7) adjusts the food consistency (such as increasing the proportion of thickening agent through a double-channel feeding tube). High risk (R ≥ 0.7) suspends feeding and triggers emergency suction channel.

[0085] In an alternative way, the method further comprises:

[0086] When food backflow is detected, stop feeding to prevent regurgitation aspiration;

[0087] When the blood oxygen saturation is less than 95% or the blood oxygen saturation decreases rapidly within a preset time, the emergency suction channel is started to connect the negative pressure pump for negative pressure suction, wherein the negative pressure suction force is adaptively adjusted according to the aspiration degree and the suction effect is monitored by the pressure sensor.

[0088] In this embodiment, emergency measures are triggered immediately once food backflow or blood oxygen abnormalities are detected, significantly reducing the risk of suffocation caused by aspiration. The negative pressure suction force is adaptively adjusted according to the aspiration degree (such as regurgitation speed, blood oxygen drop rate), avoiding excessive suction causing throat injury or secondary aspiration. For example: mild aspiration (R < 0.3) suction force is set to -50 kPa. Severe aspiration (0.3 ≤ R < 0.7) suction force is set to -100 kPa.

[0089] ​​According to the scheme provided by the application, biological information of a patient is collected through a multi-modal biosensor matrix, wherein the multi-modal biosensor matrix comprises a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor and a blood oxygen saturation sensor; the biological information of the patient is input into an adaptive feeding model to generate a personalized feeding scheme, and the adaptive feeding model is matched with a videofluoroscopic swallowing study (VFSS) data through a cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model adopts a hierarchical LSTM neural network model; a miniature pressure sensor arranged near the oral cavity is used to detect food reflux, and the personalized feeding scheme is used to feed through control of a double-channel feeding tube and a screw extrusion assembly, wherein the double-channel feeding tube controls the shunt ratio through an electromagnetic valve and comprises a food channel and a liquid channel; the screw extrusion assembly rotates the screw rod through a servo motor to realize quantitative feeding and adjusts the extrusion speed and direction to adapt to different food consistency and feeding angle requirements; a food flow path, an aspiration risk area and a corresponding aspiration risk score are generated according to the feeding of the double-channel feeding tube and the screw extrusion assembly; when the aspiration risk score exceeds a plurality of preset thresholds, different levels of early warning and intervention measures are triggered, wherein the intervention measures comprise slowing the feeding speed, adjusting the food consistency and pausing the feeding. The application calculates the aspiration risk score through multi-modal sensor data and dynamically generates a personalized feeding scheme, thereby improving the feeding safety and rehabilitation efficiency.

[0090] Figure 2 A framework schematic diagram of a personalized anti-aspiration feeding device based on multi-modal sensors and intelligent algorithms is shown. The personalized anti-aspiration feeding device based on multi-modal sensors and intelligent algorithms comprises:

[0091] The data acquisition module 210 is configured to collect biological information of a patient through a multi-modal biosensor matrix, wherein the multi-modal biosensor matrix comprises a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor and a blood oxygen saturation sensor;

[0092] The scheme generation module 220 is configured to input the biological information of the patient into an adaptive feeding model to generate a personalized feeding scheme, and the adaptive feeding model is matched with a videofluoroscopic swallowing study (VFSS) data through a cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model adopts a hierarchical LSTM neural network model;

[0093] The precise feeding module 230 is used for detecting the food backflow condition through the micro pressure sensor arranged near the oral cavity, and feeding according to the personalized feeding scheme through the control of the double-channel feeding tube and the spiral extrusion assembly, wherein the double-channel feeding tube controls the shunt ratio through the electromagnetic valve, and contains a food channel and a liquid medicine channel; the spiral extrusion assembly realizes quantitative feeding through the rotation of the screw driven by the servo motor, and adjusts the extrusion speed and direction to adapt to different food consistency and feeding angle requirements;

[0094] The risk assessment module 240 is used for generating a food flow path, an aspiration risk area and a corresponding aspiration risk score according to the feeding of the double-channel feeding tube and the spiral extrusion assembly;

[0095] The early warning intervention module 250 is used for triggering different levels of early warning and intervention measures when the aspiration risk score exceeds a plurality of preset thresholds, wherein the intervention measures include slowing down the feeding speed, adjusting the food consistency and pausing the feeding.

[0096] Figure 3 The structure schematic diagram of the computer equipment embodiment of the present application is shown, and the specific implementation of the computer equipment is not limited in the specific embodiment of the present application.

[0097] As Figure 3 shown, the computer equipment can include a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0098] Among them: the processor 302, the communications interface 304, and the memory 306 complete the communication between each other through the communications bus 308. The communications interface 304 is used for communicating with network elements of other devices such as clients or other servers. The processor 302 is used for executing the program 310, and specifically can execute the related steps in the above-mentioned personalized anti-aspiration feeding method based on multi-modal sensors and intelligent algorithms.

[0099] Specifically, the program 310 can include program code, and the program code includes computer operation instructions.

[0100] The processor 302 can be a central processing unit CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present application. One or more processors included in the computer equipment can be the same type of processor, such as one or more CPUs; or can be different types of processors, such as one or more CPUs and one or more ASICs.

[0101] a memory 306 for storing a program 310. The memory 306 can include a high-speed RAM memory and can also include a non-volatile memory, such as at least one disk memory.

[0102] According to the scheme provided by the application, the biological information of the patient is collected by a multi-modal biosensor matrix, wherein the multi-modal biosensor matrix includes surface electromyography sensors, infrared pyroelectric sensors, swallowing sound signal sensors, tongue pressure sensors and blood oxygen saturation sensors; the biological information of the patient is input into an adaptive feeding model to generate a personalized feeding scheme, and the adaptive feeding model is matched with a swallowing radiography VFSS data by a cloud collaborative diagnosis system to generate a rehabilitation trend report; wherein the adaptive feeding model adopts a hierarchical LSTM neural network model; a micro pressure sensor arranged near the oral cavity is used to detect the food reflux condition, and the food is fed according to the personalized feeding scheme by controlling a double-channel feeding tube and a spiral extrusion assembly, wherein the double-channel feeding tube controls the shunt ratio by a solenoid valve, and includes a food channel and a liquid medicine channel; the spiral extrusion assembly rotates the screw rod by a servo motor to realize quantitative feeding, and adjusts the extrusion speed and direction to adapt to different food consistency and feeding angle requirements; a food flow path, an aspiration risk area and a corresponding aspiration risk score are generated according to the feeding of the double-channel feeding tube and the spiral extrusion assembly; when the aspiration risk score exceeds a plurality of preset thresholds, different levels of early warning and intervention measures are triggered, wherein the intervention measures include slowing down the feeding speed, adjusting the food consistency and pausing the feeding. The application calculates the aspiration risk score by multi-modal sensor data, dynamically generates a personalized feeding scheme, and improves the feeding safety and rehabilitation efficiency.

[0103] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and furthermore can be divided into multiple sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the present specification (including the accompanying claims, abstract and drawings), and any method or apparatus so disclosed, can be taken in any combination, except that at least some of such features and / or processes or units are mutually exclusive, unless explicitly stated otherwise. Each feature disclosed in the present specification (including the accompanying claims, abstract and drawings) can be replaced by alternative features serving the same, equivalent or similar purpose, unless explicitly stated otherwise. Furthermore, the skilled person will appreciate that the combination of features of different embodiments implies that the features of the different embodiments are meant to be combined, unless explicitly stated otherwise. For example, in the claims below, any of the embodiments can be used in any combination. The application can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unitary claim, several of the devices mentioned in the embodiments can be implemented by means of one and the same hardware item. The steps of the above-described embodiments, unless explicitly stated otherwise, are not to be understood as having to be carried out in the order in which they are described.

Claims

1. A personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms, characterized in that: include: Collecting biological information of the patient through a multimodal biosensor matrix, wherein the multimodal biosensor matrix includes a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor, and a blood oxygen saturation sensor; Inputting the patient's biological information into an adaptive feeding model to generate a personalized feeding plan, and matching the adaptive feeding model with swallowing contrast imaging (VFSS) data through a cloud-based collaborative diagnostic system to generate a recovery trend report; wherein the adaptive feeding model uses a hierarchical LSTM neural network model; A micro pressure sensor located near the oral cavity detects food reflux and controls a dual-channel feeding tube and a spiral extrusion assembly to deliver food according to the personalized feeding plan. The dual-channel feeding tube, which includes a food channel and a medication channel, is controlled by a solenoid valve to control the diversion ratio. The spiral extrusion assembly uses a servo motor to drive the screw to achieve quantitative feeding and adjust the extrusion speed and direction to accommodate different food consistencies and feeding angles. generating a food flow path, an aspiration risk area, and a corresponding aspiration risk score based on feeding using the dual-channel feeding tube and the spiral extrusion assembly; When the aspiration risk score exceeds multiple preset thresholds, different levels of warnings and intervention measures are triggered, wherein the intervention measures include slowing down the feeding speed, adjusting the consistency of the food, and suspending feeding.

2. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 1, characterized in that: The surface electromyography sensor is arranged between the submental muscles and the thyrohyoid muscle, and is used to collect swallowing electromyographic signals and extract the energy proportion of the 50-150 Hz frequency band as the electromyographic feature; The infrared pyroelectric sensor is used to detect the respiratory cycle to determine the inhalation and exhalation phases; The swallowing sound signal sensor is arranged at the suprasternal fossa of the neck, and is used to collect the swallowing sound signal and perform time-frequency analysis to extract the Mel-frequency cepstral coefficients; The tongue pressure sensor is used to measure the pressure of the tongue on food to obtain tongue movement control ability data; The blood oxygen saturation sensor is used to monitor real-time changes in the patient's blood oxygen level.

3. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 1, characterized in that: The first LSTM layer of the hierarchical LSTM neural network model is used to process historical feeding data and biosensor data; The second LSTM layer of the hierarchical LSTM neural network model is used to generate the optimal flow rate, food consistency, feeding interval, food particle size and feeding angle based on patient subjective feedback and expert rule base.

4. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 1, characterized in that: The method further comprises: When food reflux is detected, feeding is stopped to prevent reflux aspiration; When the blood oxygen saturation is lower than 95% or the blood oxygen saturation drops rapidly within a preset time, the emergency suction channel is activated to connect the negative pressure pump for negative pressure suction. The negative pressure suction force is adaptively adjusted according to the degree of aspiration and the suction effect is monitored by a pressure sensor.

5. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 1, characterized in that: After generating the personalized feeding plan, the method further includes: Massaging the cheek muscles for 3-5 minutes using a vibrating massage device at a preset massage frequency, duration, and amplitude to stimulate facial sensory nerves; wherein the preset massage frequency, duration, and amplitude are individually adjusted by the adaptive feeding model based on the patient's VFSS data and biofeedback information; The swallowing reflex is activated by placing a popsicle or a cotton swab in ice water at the base of the tongue for 1–2 seconds using an ice stimulation device.

6. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 1, characterized in that: The calculation formula of the aspiration risk score is: in, It is the normalized value of the energy proportion of the surface electromyographic signal in the 50-150 Hz frequency band; is the dynamic change rate of the Mel-frequency cepstral coefficient MFCC of the swallowing sound signal; The temporal complexity of pressure acquisition for tongue pressure sensors; is the rate of decrease in blood oxygen saturation; is the reflux velocity; For time; is the weight coefficient.

7. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 6, characterized in that: The calculation formula for the dynamic change rate of the Mel-frequency cepstral coefficient MFCC of the swallowing sound signal is: in, is a step function; is the abnormal voiceprint threshold; is the i-th Mel frequency cepstral coefficient; is the total number of coefficients in the MFCC vector; Weight coefficient The constrained optimization solution expression is: in, Scoring for clinical gold standard; is the L2 regularization coefficient; is the predicted value of aspiration risk.

8. The personalized anti-aspiration feeding method based on multimodal sensors and intelligent algorithms according to claim 1, characterized in that: The dual-channel feeding tube has a built-in positive pressure pulse flushing module; The positive pressure pulse flushing module adopts a three-stage cleaning procedure consisting of detergent injection, vortex oscillation and sterile water flushing; the frequency optimization formula of the vortex oscillation is: in, is the pipe wall shear force; is the kinetic energy of the fluid; is the energy efficiency balance coefficient; is the optimized frequency of eddy current oscillation; is the frequency; When the dual-channel feeding tube retracts, it automatically wraps around the built-in UV-C LED disinfection roller. When the bioluminescence value of the ATP fluorescence detection point set at the tube entrance of the dual-channel feeding tube exceeds 50RLU, the enhanced disinfection program is activated and a replacement reminder is pushed to the medical end. In addition, a one-way duckbill valve is set at the connection between the food channel and the liquid medicine channel to prevent the cleaning liquid from flowing back and contaminating the servo motor, and a humidity sensor is equipped to monitor the dryness of the tube.

9. A personalized anti-aspiration feeding device based on multimodal sensors and intelligent algorithms, characterized in that: include: A data acquisition module, configured to acquire biological information of the patient through a multimodal biosensor matrix, wherein the multimodal biosensor matrix includes a surface electromyography sensor, an infrared pyroelectric sensor, a swallowing sound signal sensor, a tongue pressure sensor, and a blood oxygen saturation sensor; A plan generation module is configured to input the patient's biological information into an adaptive feeding model to generate a personalized feeding plan, and to match the adaptive feeding model with VFSS data via a cloud-based collaborative diagnostic system to generate a recovery trend report; wherein the adaptive feeding model utilizes a hierarchical LSTM neural network model; A precision feeding module detects food reflux via a micro-pressure sensor positioned near the oral cavity and controls a dual-channel feeding tube and a spiral extrusion assembly to deliver food according to a personalized feeding plan. The dual-channel feeding tube, which includes a food channel and a medication channel, is controlled by a solenoid valve for diversion ratio control. The spiral extrusion assembly achieves quantitative feeding by rotating a screw driven by a servo motor and adjusts the extrusion speed and direction to accommodate varying food consistencies and feeding angles. a risk assessment module for generating a food flow path, an aspiration risk area, and a corresponding aspiration risk score based on feeding via the dual-channel feeding tube and the spiral extrusion assembly; The early warning and intervention module is used to trigger different levels of early warning and intervention measures when the aspiration risk score exceeds multiple preset thresholds, wherein the intervention measures include slowing down the feeding speed, adjusting the consistency of the food, and suspending feeding.

10. A computer device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the personalized anti-aspiration feeding method based on the multimodal sensor and intelligent algorithm.

Images