MUMBAI, India, Oct. 5 -- Intellectual Property India has published a patent application (202641116172 A) filed by Sr University, on September 28, 2026, for An Ai-Enabled Wearable Sensor Fusion System For Real-Time Detection And Prediction Of Abnormal Human Physiological Conditions.
Inventors include Dr. Bharani Sethupandian S; and Dr. Durgesh Nandan.
The application for the patent was published on October 02, 2026, under issue no. 40/2026.
Abstract: ABSTRACT OF THE INVENTION: The present invention provides an artificial-intelligence-enabled wearable sensor-fusion system for real-time detection and short-term prediction of abnormal human physiological conditions. A compact wearable device continuously acquires synchronised multi-modal signals that include photoplethysmogram (PPG), electrocardiogram (ECG), inertial measurement unit (accelerometer and gyroscope) data, skin temperature and galvanic skin response. On-device pre-processing removes artefacts and extracts both hand-crafted physiological features and learned embeddings. A hybrid fusion engine adaptively combines early (feature-level), late (decision-level) and quality-aware (attention- or Kalman-filter-based) fusion strategies, thereby maximising the complementary information of the different modalities while suppressing unreliable channels. The fused representation is processed by a dual-head lightweight neural network optimised for edge deployment. One head performs real-time classification of abnormal states such as cardiac arrhythmias, hypoxic episodes, falls and acute autonomic stress responses. The second head forecasts the probability of these states occurring within a configurable prediction horizon of typically 5–30 minutes. When a calibrated risk threshold is crossed, the system issues graded local and remote alerts. Because the majority of computation occurs on the wearable device, latency is minimised, privacy is preserved and continuous cloud connectivity is unnecessary. Quantitative evaluation demonstrates detection accuracy exceeding 95 %, early-warning lead times of approximately 12 minutes and substantially reduced false-positive rates compared with single-sensor or rule-based multi-sensor baselines. The invention therefore enables proactive, privacy-preserving ambulatory monitoring suitable for chronic-disease patients, elderly individuals, post-operative care and athletic performance applications.The technology that we will be discussing today is a wearable AI-driven sensor-fusion system. The design is to be worn at all times. The system is not only capable of detecting abnormal human physiological conditions in real time, but also built to be able to forecast such abnormalities in the short term. This is the major aim of the system. A tiny wearable gadget has the potential to continuously acquire synchronized multi-modal signals. The device can do this, among other things. The readings consist of galvanic skin response, photoplethysmogram (PPG), electrocardiogram (ECG), inertial measuring unit (accelerometer and gyroscope) and skin temperature. The galvanic skin reaction is also provided. In this study, we add also information that was acquired from these measures in the previous sentence. The pre-processing carried out on the device performs several jobs including the removal of artifacts, the extraction of hand-crafted physiological features and learnt embeddings, and the fusion of all these types of information. The accomplishment of each of these obligations occurs in the same time frame as the others. A hybrid fusion engine might also serve to amplify the complementary information of the many modalities while at the same time dampening the unreliable channels. This is not out of the question. The fact of the issue is that this is something that is achievable. This is achieved via the use of a combination of early (feature-level), late (decision-level) and quality-aware (attention- or Kalman-filter-based) fusion techniques in an adaptive manner. This allows to do this. The fused representation is processed by a lightweight neural network with two heads for edge deployment. This implementation is optimized for the neural network to be run at the edge. This network is designed to operate at the highest possible level, wherever it is situated, so as to maximize its efficiency. One of the heads is in charge of real time classification of abnormal conditions and will be responsible for doing this work. This is their responsibility as it falls within their category. Arrhythmias of the heart, episodes of hypoxia, falls and acute autonomic stress reactions are abnormal situations that may arise. But this is not a complete list and does not contain all. The second head makes predictions on the probability of different states occurring in this window using an adjustable predictable prediction horizon, usually between five and thirty minutes. The aim of this is to produce predictions of the probability of different states to occur . This horizon can be tuned to the needs of the system. These forecasts are designed to assess the probability of certain occurrences occurring, which is the overall purpose of the forecasts. If the specified danger threshold is exceeded, the system issues graded warnings locally and remotely. Both ways, these warnings will be issued. At the same time, each and every one of these messages will be delivered. The warnings should be conveyed to the interested parties without further delay. The majority of the processing is done by the wearable device. Therefore, the latency is brought down to an acceptable level, privacy is protected, and the need for cloud connectivity that is maintained continuously is eliminated. This result has been reached since the wearable device is responsible for the majority of the processing that goes place. The quantitative evaluation indicates that the detection accuracy is 95%, the early- warning lead durations are around 12 minutes, and the false-positive rates are dramatically reduced compared to the baselines of single-sensor or rule-based multi-sensor systems. All this is proven by the fact that the detection accuracy is remarkable. Every one of these discoveries is being made against the baselines that are being provided by the systems. And all this researches are based on the assumption that the leads for early warning are about twelve minutes. The comparative research of the two different sorts of systems serves as the basis for each and every one of these discoveries. Thus the innovation makes it possible to carry out a proactive ambulatory surveillance while maintaining the confidentiality of the patients at the same time. This scenario is made possible by this invention. One of the biggest benefits is that this is the case. Monitoring of patients can serve for a large number of purposes including but not limited to post-operative care, patients suffering from chronic diseases, elderly folks and applications related to increased athletic performance. These are only a few of the possible applications.
Disclaimer: Curated by HT Syndication.
