AI intelligent system for early warning Alzheimer's disease and Parkinson's disease of old people
By collecting gait and voice data through an AI intelligent system and combining it with multimodal fusion analysis, the problem of data acquisition and algorithm accuracy in the early diagnosis of Alzheimer's and Parkinson's diseases in the elderly has been solved, enabling non-invasive, convenient home monitoring and efficient early warning.
Patent Information
- Application Number
- CN202511227026.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-29
- Publication Date
- 2025-11-07
AI Technical Summary
Existing technologies for the early diagnosis of Alzheimer's and Parkinson's diseases in the elderly suffer from problems such as difficulty in data acquisition, data accuracy being affected by environmental interference, low algorithm accuracy, and poor universality, resulting in a lack of sensitivity and timeliness in traditional diagnostic methods.
Employing an AI intelligent system, gait video is captured by a 4K high-definition camera deployed on the top of a TV or smart screen, and voice data is collected by a 6-microphone array. The system combines 3D-CNN, LSTM, and Transformer models to analyze gait and voice features, enabling multimodal fusion early warning.
It enables non-invasive and convenient home monitoring, improves the coverage and accuracy of early warning, reduces reliance on professional equipment and family care knowledge, and supports real-time early warning and long-term trend tracking.
Smart Images
Figure CN120899189A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field, and in particular to an AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly. BACKGROUND
[0002] Currently, Alzheimer's disease and Parkinson's disease pose great challenges to the health of the elderly. Early diagnosis of these two diseases is difficult, and the clinical latency of Alzheimer's disease is as long as 15-20 years, and by the time of diagnosis, it is often in the middle and late stages; more than 50% of dopamine neurons in the brain of patients with Parkinson's disease are damaged and die by the time of diagnosis. Traditional diagnosis relies on subjective methods such as scales, and lacks sensitivity and timeliness. However, with the development of AI technology, it has shown potential in the early warning and diagnosis of these two diseases, such as through the analysis of gait, speech, etc. to achieve early warning, and some technologies have broken through 90% accuracy, bringing new hope for early detection and early intervention of the disease.
[0003] Currently, AI technology for early warning of Alzheimer's disease and Parkinson's disease in the elderly still has many shortcomings. On the one hand, data acquisition and quality problems are prominent. It is difficult to collect a large amount of comprehensive physiological and behavioral data of the elderly, and data such as speech and gait are easily affected by the complex home environment, affecting data accuracy. At the same time, the problems of patient privacy protection and data labeling also hinder the accumulation of large amounts of high-quality data. On the other hand, the algorithm maturity is insufficient. The existing AI algorithm has poor accuracy and stability in identifying subtle features of the disease, and it is difficult to accurately distinguish between normal aging and early disease manifestations. Moreover, the technology has poor universality, and the physiological differences and different living habits among different individuals make it difficult for the model to maintain high efficiency in diverse scenarios, and there is still a long way to go before clinical application. SUMMARY
[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide an AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly, to solve the limitations of early symptoms being hidden and relying on subjective scales in traditional diagnosis; to break through the scene limitations of professional equipment detection and achieve convenient home monitoring; to reduce the resistance of the elderly to medical examinations and improve early warning coverage; to reduce the dependence on professional knowledge of family caregivers and help timely intervention.
[0005] To achieve the above purpose, the following technical solutions are adopted: The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly comprises: An image acquisition module for acquiring gait video of a user; A speech acquisition module for acquiring speech data of a user; A data processing module comprising a gait analysis unit, a speech analysis unit, and a multi-modal fusion unit; The gait analysis unit extracts gait features based on the gait video, and outputs a Parkinson's disease risk warning signal when the gait feature value exceeds a preset threshold value. The voice analysis unit extracts voice features based on the voice data, and outputs an Alzheimer's disease risk warning signal when the voice feature value exceeds a preset threshold value. The multi-modal fusion unit fuses and calculates the gait features and voice features in a preset weight ratio, and outputs a comprehensive risk warning signal of Alzheimer's disease and Parkinson's disease when the output comprehensive risk score exceeds a preset threshold value.
[0006] Further, the image acquisition module is a 4K high-definition camera deployed on the top of a television or smart screen, with a frame rate of 30fps and a coverage range of 3-5 meters.
[0007] Further, the 4K high-definition camera can automatically adjust the light sensitivity to ensure that the bone point recognition accuracy is greater than 90% in backlight scenes.
[0008] Further, the voice acquisition module is a 6-microphone array that uses noise reduction and beamforming technology, with an effective collection distance of 3 meters.
[0009] Further, the gait analysis unit uses a 3D-CNN model to extract gait features, including step standard deviation and swing phase proportion. The voice analysis unit uses an LSTM model to extract voice features, including pitch jitter rate and lexical repetition rate. The multi-modal fusion unit uses a Transformer model to fuse and analyze the gait features and voice features in a preset weight ratio, and outputs a comprehensive risk score.
[0010] Further, when the step standard deviation extracted by the gait analysis unit is greater than 20 cm or the swing phase proportion is less than 35%, the gait analysis unit outputs a Parkinson's disease risk warning signal. When the pitch jitter rate extracted by the voice analysis unit is greater than 3% or the lexical repetition rate is greater than 20%, the voice analysis unit outputs an Alzheimer's disease risk warning signal. When the comprehensive risk score is greater than or equal to 60 points, the multi-modal fusion unit outputs a comprehensive risk warning signal of Alzheimer's disease and Parkinson's disease.
[0011] Further, the gait analysis unit presets three body type models according to the height of the old person, and selects the corresponding short / medium / tall body type model through the calibration program when first used.
[0012] Further, the voice analysis unit supports Mandarin and multiple dialects, and the voice model is trained on a dialect corpus.
[0013] By adopting the above technical solution, the beneficial effects of the present invention are as follows: This invention employs multimodal fusion perception, capturing gait and facial micro-expressions through cameras on TVs and smart screens, and collecting voice features through microphones, combined with AI algorithms for modeling and analysis; it achieves non-invasive monitoring without the need for wearable devices, adapting to the daily usage habits of the elderly; it has real-time early warning and long-term trend tracking functions, linking home and medical terminals. Attached Figure Description
[0014] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a schematic block diagram of the AI intelligent system of the present invention. Detailed Implementation
[0015] like Figure 1 As shown, the present invention provides an AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly, comprising an image acquisition module, a voice acquisition module, and a data processing module.
[0016] The image acquisition module is used to capture the user's gait video. This module is a 4K high-definition camera deployed on top of a TV or smart screen, operating at 30fps with a coverage range of 3-5 meters. The 4K high-definition camera can automatically adjust its light sensitivity (ISO adjustment range 100-800) to ensure a skeletal point recognition accuracy of >90% in backlit scenes.
[0017] The voice acquisition module is used to collect users' voice data (collecting everyday conversations, such as "changing the channel" and "adjusting the volume"). The voice acquisition module is a 6-microphone array with an effective acquisition distance of 3 meters. It employs noise reduction and beamforming technology to effectively filter out interference from the volume of the TV and smart screen itself (≤60 decibels) on voice analysis.
[0018] The data processing module includes a gait analysis unit, a speech analysis unit, and a multimodal fusion unit.
[0019] The gait analysis unit extracts gait features from gait videos (inputting a 10-second video of the user's gait) using a 3D-CNN model. These features include stride standard deviation (normal range 5-15cm, abnormal >20cm) and swing phase ratio (normal 40%-50%, early Parkinson's disease <35%). When gait feature values exceed preset thresholds, a Parkinson's disease risk warning signal is output. Specifically, when the stride standard deviation extracted by the gait analysis unit is greater than 20cm or the swing phase ratio is less than 35%, a Parkinson's disease risk warning signal is output. For example, if an elderly person's stride standard deviation is greater than 22cm for three consecutive days, the model outputs a Parkinson's disease risk probability of 0.78.
[0020] When the elderly swing phase ratio continues to be greater than 45% for 2 consecutive days (the normal elderly usually stabilizes at 32%-40%), and is accompanied by a unilateral swing phase fluctuation amplitude greater than 8%, the model outputs a Parkinson's disease early warning risk probability of 0.81; When the elderly single-day swing phase ratio breaks through 50%, while the step length coefficient of variation is greater than 15% and the double-foot support time is prolonged to more than 28%, the model comprehensively determines the Parkinson's disease early warning risk probability of 0.85, prompting that the problems of motor retardation and muscle tension abnormality need to be prioritized; When the elderly swing phase ratio shows a stepwise increase (from 38% to 47%) for 4 consecutive days, and the swing phase fluctuation frequency is greater than 3 times / hour at night, the model outputs a Parkinson's disease early warning risk probability of 0.76, which needs to be further evaluated in combination with clinical symptoms such as hand tremor and gait freezing.
[0021] The voice analysis unit extracts voice features based on voice data using an LSTM model, including pitch jitter rate (pitch jitter rate > 3% is abnormal) and lexical repetition rate (lexical repetition rate > 20% is abnormal, indicating cognitive decline). When the voice feature value exceeds the preset threshold, an Alzheimer's disease risk warning signal is output, i.e., when the pitch jitter rate extracted by the voice analysis unit is greater than 3% or the lexical repetition rate is greater than 20%, an Alzheimer's disease risk warning signal is output. As an example, when the frequency of "forget" and "don't know" and other words in the elderly increases by 40% compared to the baseline within 1 week, the model outputs an Alzheimer's disease early warning risk probability of 0.72.
[0022] When the elderly have a pitch jitter rate greater than 3.5% for 5 consecutive days in language communication (healthy elderly usually ≤ 2.2%), and the number of sentence interruptions increases by 35% compared to the baseline, the model outputs an Alzheimer's disease early warning risk probability of 0.79, indicating that the language center cognitive function may have early degradation; When the elderly have a peak pitch jitter rate of 5.1% in a single day of conversation, and the frequency of "repeating the same event" (such as repeatedly mentioning "just had a meal" within 10 minutes) is greater than 4 times, the model comprehensively determines an Alzheimer's disease early warning risk probability of 0.83, which needs to focus on memory extraction and language organization ability abnormalities; When the elderly have a fluctuating increase in pitch jitter rate within 1 week (from 2.0% to 4.3%), and the pitch jitter is accompanied by a reaction delay of more than 10 seconds for more than 6 times when answering basic questions such as "what day is it" and "where is home", the model outputs an Alzheimer's disease early warning risk probability of 0.75, and it is recommended to further confirm in combination with cognitive scale evaluation.
[0023] The multi-modal fusion unit fuses and calculates the gait features and speech features with a preset weight ratio, and outputs a comprehensive risk warning signal of Alzheimer's disease and Parkinson's disease when the output comprehensive risk score exceeds a preset threshold. Specifically, the multi-modal fusion unit uses a Transformer model to fuse and analyze the gait features and speech features with a preset weight ratio (such as attention weight distribution: gait features 40%, speech features 60%), and outputs a comprehensive risk score (0-100 points). When the comprehensive risk score is greater than or equal to 60 points, the multi-modal fusion unit outputs a comprehensive risk warning signal of Alzheimer's disease and Parkinson's disease.
[0024] Individual difference adaptation: In the present application, the gait analysis unit presets three gait models according to the height of the old person, and selects the corresponding gait model through the calibration program during the first use. Specifically, for old people with a height of 150-180 cm, three gait models (short / medium / tall body types) are preset, and the 3-minute calibration program (the old person walks 3 times along a 5-meter straight line) is used during the first use.
[0025] Dialect adaptation: The speech analysis unit in the present application supports Mandarin and multiple dialects (such as Cantonese, Sichuanese, etc.), and the speech model is trained with dialect corpus (accuracy improved to 92%).
[0026] The application field of the present application covers family care for the elderly, community medical care, and remote health management. In the family scenario, the old person's daily behavior, speech, and movement characteristics are monitored in real time with the help of smart TVs and smart screens; when a danger is foreseen, the family members' mobile phones (SMS, phone calls, APP, etc.) are automatically notified in a timely manner. The community end can integrate data to form a regional health record, assisting grassroots screening; the remote medical field provides continuous disease tracking basis for medical staff. The purpose is to realize early warning of Alzheimer's disease and Parkinson's disease through a non-invasive method, solve the problem of relying on subjective scales and delayed diagnosis in traditional diagnosis, and strive for the golden period of intervention for the elderly. At the same time, it reduces the pressure of family care, promotes the efficient allocation of resources for the elderly, and constructs a collaborative neurodegenerative disease prevention and control system of "family-community-hospital".
[0027] The above describes a specific embodiment of the present application, but those skilled in the art should understand that this is only an example, and those skilled in the art can make various changes or modifications to this embodiment without departing from the principles and essence of the present application, and these changes and modifications all fall within the protection scope of the present application.
Claims
1. An AI-powered intelligent system for early warning of Alzheimer's and Parkinson's diseases in the elderly, characterized by: The application relates to an AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease of the elderly. An image acquisition module is used for acquiring gait videos of users; A voice acquisition module is used for acquiring voice data of users; A data processing module comprises a gait analysis unit, a voice analysis unit and a multi-modal fusion unit; The gait analysis unit extracts gait features based on the gait videos, and outputs a Parkinson's disease risk early warning signal when the gait feature value exceeds a preset threshold value; The voice analysis unit extracts voice features based on the voice data, and outputs an Alzheimer's disease risk early warning signal when the voice feature value exceeds a preset threshold value; The multi-modal fusion unit fuses the gait features and the voice features in a preset weight ratio and performs calculation, and outputs a comprehensive risk early warning signal of Alzheimer's disease and Parkinson's disease when the output comprehensive risk score exceeds a preset threshold value. 2.The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly according to claim 1, characterized in that: The image acquisition module is a 4K high-definition camera arranged on the top of a television or a smart screen, the frame rate is 30fps, and the coverage range is 3-5 meters. 3.The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly according to claim 2, characterized in that: The 4K high-definition camera can automatically adjust light sensitivity, and ensure that the bone point recognition accuracy is greater than 90% in a backlight scene. 4.The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly according to claim 1, characterized in that: The voice acquisition module is a 6-microphone array, adopts noise reduction and beam forming technology, and the effective acquisition distance is 3 meters. 5.The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly according to claim 1, characterized in that: The gait analysis unit adopts a 3D-CNN model to extract gait features, and the gait features include step standard deviation and swing phase proportion; The voice analysis unit adopts an LSTM model to extract voice features, and the voice features include pitch jitter rate and vocabulary repetition rate; The multi-modal fusion unit adopts a Transformer model to fuse and analyze the gait features and the voice features in a preset weight ratio, and outputs a comprehensive risk score.
6. The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease of the elderly according to claim 1, wherein: When the step standard deviation extracted by the gait analysis unit is greater than 20 cm or the swing phase proportion is less than 35%, a Parkinson's disease risk early warning signal is outputted; When the pitch jitter rate extracted by the voice analysis unit is greater than 3% or the vocabulary repetition rate is greater than 20%, an Alzheimer's disease risk early warning signal is outputted; When the comprehensive risk score is greater than or equal to 60 points, the multi-modal fusion unit outputs a comprehensive risk early warning signal of Alzheimer's disease and Parkinson's disease. 7.The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly according to claim 1, characterized in that: The gait analysis unit presets three gait models of short, medium and high according to the height of the old people, and selects the corresponding gait model through a calibration program during the first use. 8.The AI intelligent system for early warning of Alzheimer's disease and Parkinson's disease in the elderly according to claim 1, characterized in that: The voice analysis unit supports Mandarin and multiple dialects, and the voice model is trained by a dialect corpus.