Abstract
An acute exacerbation of chronic obstructive pulmonary disease (AECOPD) is a significant reason for avoidable admission to hospital, avoidable emergency department attendance and pressure on primary care. While existing remote monitoring system had helped in early symptoms of disease deterioration, they still rely much on cloud base data processing which introduces high latency and also lack real life integration with GP workflow for decision making. In this paper, a connected and disconnected Edge AI framework for COPD monitoring is proposed, enabling the collection of patient vital signs from various sensors on an ESP32 wearable device, the execution of local TinyML inference, offline storage of vital signs in a SQLite database, and the synch of COPD related risk summaries to the GP dashboard when Wi-Fi connectivity is available. The embedded AI header inference model is an int8 quantised 1D convolutional neural network capable of deploying into ESP32 graded controller with a firmware to drive the logic. The dashboard features Patient, Doctor, Analytics, and Appointment entities; ranks high risk patients at the top of GP view as alarm events; and enables doctors to schedule appointments if threshold levels are breached. The system also correlates PatientID with DeviceID during onboarding and is designed with NHS clinical safety, data security, UK GDPR and software/AI medical device in mind. The proposed architecture illustrates the integration of TinyML, local buffering and secure clinical decision support to enable early COPD risk detection – to actionable GP review.