A groundbreaking study has demonstrated the potential for artificial intelligence (AI) to predict the risk of cerebrovascular diseases, such as stroke, by analyzing subtle changes in daily behavior patterns. Researchers utilized non-contact IoT sensors and AI algorithms within smart home environments to identify early warning signs and assess the imminent risk of these conditions. This innovative approach could revolutionize early detection, particularly for elderly individuals living alone.
The Challenge of Early Cerebrovascular Disease Detection
Cerebrovascular diseases, including stroke, cerebral infarction, and cerebral hemorrhage, necessitate prompt diagnosis and intervention for the best possible outcomes. However, current diagnostic methods often rely on hospital visits after significant symptoms have already manifested. Medical imaging techniques like MRI and CT scans are typically employed at this later stage. This leaves a critical gap in identifying the gradual behavioral shifts that may precede a major event, making proactive risk assessment challenging in everyday life.
A Novel Approach Using Smart Home Technology and AI
A collaborative research team, comprising Professor Jo Gyeong-hee from the Neurology Department at Korea University Guro Hospital, Professor Im Ri-sa from the Department of Construction and Environmental Engineering at Korea University of Science and Technology, and Professor Jeong Jo-un from the Department of Electronic and Electrical Engineering at Sungkyunkwan University, has pioneered a new method. Their research confirms the feasibility of predicting the pre-symptomatic stages and immediate risk of cerebrovascular diseases by leveraging non-contact Internet of Things (IoT) sensors installed in homes and artificial intelligence (AI).
Study Methodology and Participants
The study involved 1,224 Korean individuals aged 65 and older, all residing in smart home environments. The research team meticulously analyzed 13,362 data points collected over 14-day intervals. Participants were categorized into three groups:
- Control Group: 598 individuals with no prior diagnosis of cerebrovascular disease.
- Patient Group: 598 individuals already diagnosed with cerebrovascular disease.
- Precursor Group: 28 individuals who initially had no diagnosis but were later hospitalized due to cerebral infarction or hemorrhage.
This diverse participant pool allowed for a comprehensive analysis of behavioral changes associated with the progression of these conditions.
Data Collection and AI Analysis
The research team collected data from various non-contact sensors, including motion sensors, door sensors, and indoor temperature and humidity sensors. This information was used to analyze physical activity levels, sleep patterns, and indoor environmental conditions. The AI model was trained to comprehensively learn factors such as:
- Changes in overall activity levels within the home.
- Movement patterns before sleep.
- Activity during nighttime hours.
- Periods of inactivity.
- Sleep fragmentation.
Non-contact sensors are crucial in this methodology as they can detect movement and environmental changes without requiring the individual to wear any devices. The “precursor group” specifically refers to individuals who, while not yet diagnosed, were in a stage where their risk of developing the disease was significantly elevated.
AI Model Performance and Key Behavioral Indicators
The AI model demonstrated remarkable accuracy in distinguishing between different risk levels and patient groups. Key performance metrics included:
- AUPRC (Area Under the Precision-Recall Curve): 0.85 in distinguishing the precursor group from the control group. This metric is particularly important for identifying individuals at high risk.
- AUROC (Area Under the Receiver Operating Characteristic Curve): 0.91 in differentiating between the diagnosed patient group and the control group. This indicates a strong overall classification performance.
- Sensitivity and Specificity: When predicting the imminent risk within the precursor group, the model achieved a sensitivity of 95.12% and a specificity of 96.97%. The accuracy was also high at 96.53%.
Significant Behavioral Markers Identified
The AI analysis pinpointed specific behavioral indicators that were highly predictive of cerebrovascular disease risk:
- For the Precursor Group: A sustained increase in movement before bedtime (between 10 PM and 2 AM), reduced periods of inactivity, and delayed sleep onset times were significant predictors.
- For the Diagnosed Patient Group: Increased activity during early morning hours and frequent sleep interruptions were notable patterns.
- For Imminent Risk Prediction: Reduced physical activity during evening hours, sustained periods of high activity, and indoor humidity levels were identified as crucial factors.
Implications for Early Intervention and Future Care
This research holds significant promise as a supplementary tool to encourage early diagnosis and medical check-ups by continuously monitoring behavioral changes in daily life. The implications are particularly profound for elderly individuals living alone, who may delay recognizing symptoms or seeking medical attention. Technologies that can passively detect warning signs within the home environment could provide invaluable support.
Professor Jo Gyeong-hee emphasized the importance of early intervention, stating, “Cerebrovascular diseases are conditions where early response greatly influences the prognosis, but subtle changes are easily missed in elderly patients.” She added, “This study demonstrates that behavioral changes in daily life can serve as a digital biomarker for cerebrovascular disease risk.”
The study, titled ‘AI home monitoring for behavioral markers of cerebrovascular disease,’ was recently published in the esteemed international academic journal npj Digital Medicine, a leading publication in the digital health field.
