Edge computing

In one line

Edge computing moves computation next to the data source — a ward gateway scoring an ECG stream locally instead of shipping every sample to a distant data centre.

The problem it solves

The cloud-only model — stream everything to a data centre, compute there, send results back — breaks for health monitoring: it's too slow for real-time alarms, too bandwidth-hungry for high-rate waveforms, too fragile when the internet drops, and it pushes raw PHI off-site. Edge computing answers all four by doing the work near where the data is born.

The compute hierarchy

• Device — a microcontroller running a tiny model ("TinyML"): basic detection on the sensor itself.
• Gateway / edge node — a small box aggregating a ward's devices, running heavier inference, buffering, and protocol translation.
• Cloud — model training, fleet management, long-term storage, cross-site analytics.

The edge filters, compresses, and reacts; only events and summaries travel upstream.

The four wins

• Latency — an alarm fires in milliseconds, not after a cloud round-trip.
• Bandwidth — don't stream 250 Hz waveforms continuously; send beats and summaries.
• Resilience — keeps working through internet outages (critical in a hospital).
• Privacy — raw data can stay on-site, reducing the PHI that ever leaves (HIPAA/DPDP surface area).

Where it shows up in digital health

Real-time vital-sign analytics and early-warning scores at the bedside; ambient fall detection that must not depend on connectivity; and rural / low-bandwidth deployments where the link to the cloud is the weakest component — a first-class concern for India-scale health IT.

Common pitfalls

• Edge sprawl — many small nodes are hard to update and secure; plan fleet management.
• Model drift on-device — an edge model still needs monitoring and periodic refresh from the cloud.
• Assuming edge = private — local processing helps, but governance and encryption still apply.
• Inconsistent results — device, edge, and cloud models must agree, or alarms differ by layer.

Key takeaways

• Edge computing processes data near the source across a device → edge → cloud hierarchy.
• It wins on latency, bandwidth, resilience, and privacy — all central to medical monitoring.
• TinyML pushes inference onto the device itself.
• Especially valuable where connectivity is the weak link — a key India-scale design choice.