Wearable medical monitoring device with microneedle array and artificial intelligence (AI)-based health anomaly detection

A method for continuous health monitoring, said method comprising the steps of:

• extracting, by a microneedle array of a wearable device adhered to skin of a user, interstitial or capillary fluid from said user;
• transporting said extracted fluid through microfluidic channels from said microneedle array to sensor chambers, said microfluidic channels comprising flow rectifiers configured to maintain uniform fluid flow rates and heparin-coated anti-clotting surfaces configured to prevent sensor fouling;
• detecting, by a plurality of biosensors of said wearable device, a plurality of biomarkers from said extracted fluid, wherein said plurality of biomarkers comprises glucose and at least one complete blood count parameter;
• generating, by a microprocessor of said wearable device, biomarker data based on signals from said plurality of biosensors, wherein said microprocessor performs signal conditioning and pre-processing of said signals;
• transmitting, by a wireless communication module of said wearable device, said biomarker data to a user device;
• forwarding, by said user device, said biomarker data to a server via a network;
• analyzing, by a cloud analytics layer of said server, said biomarker data using one or more machine learning models including at least one of a time-series forecasting model using Long Short-Term Memory neural networks configured to detect rate-of-change metrics for each biomarker and to predict biomarker trajectory based on detected trends, an unsupervised anomaly detection model configured to detect rare patterns not present in training data, a classification model for risk scoring, and an ensemble model configured to integrate outputs from said time-series forecasting model, said unsupervised anomaly detection model, and said classification model, and to require agreement across said models, to detect anomalies indicative of health conditions, wherein said one or more machine learning models are configured to be updated via federated learning using anonymized data;
• and
• generating, by an alerting and triage layer of said server, a health alert based on a detected anomaly.

A health monitoring system, comprising:

• a wearable medical monitoring device configured to be adhered to skin of a user, said wearable medical monitoring device comprising:
• a microneedle array configured to extract interstitial or capillary fluid from said user;
• a plurality of biosensors configured to detect a plurality of biomarkers including glucose and at least one complete blood count parameter from the extracted fluid;
• microfluidic channels configured to transport said extracted fluid from said microneedle array to sensor chambers, said microfluidic channels comprising flow rectifiers configured to maintain uniform fluid flow rates and heparin-coated anti-clotting surfaces configured to prevent sensor fouling;
• a microprocessor configured to perform signal conditioning and pre-processing of biosensor signals and to generate biomarker data based on signals from the plurality of biosensors;
• and
• a wireless communication module configured to transmit said biomarker data;
• a first user device configured to receive said biomarker data from said wearable medical monitoring device and to forward said biomarker data via a network;
• and
• a server configured to receive said biomarker data from said first user device,
• said server comprising:
• a cloud analytics layer configured to analyze said biomarker data using one or more machine learning models to detect anomalies indicative of health conditions, said one or more machine learning models comprising a time-series forecasting model comprising a Long Short-Term Memory neural network, an unsupervised anomaly detection model, a classification model for risk scoring, and an ensemble model configured to integrate outputs from said time-series forecasting model, said unsupervised anomaly detection model, and said classification model, and to require agreement across said models before generating alerts;
• and
• an alerting and triage layer configured to generate a health alert based on a detected anomaly,
• wherein said one or more machine learning models are configured to be updated via federated learning using anonymized data.