How We Built a Sub-50ms DICOM Streaming Pipeline

When we started PrizMed in 2019, the biggest pain point we heard from healthcare providers was simple: "Why does it take 45 seconds to load an MRI study acquired 2 minutes ago?"

The Challenge

Medical imaging data is uniquely demanding. A single CT study can contain 5,000+ slices at 512×512 pixels, totaling several gigabytes of raw DICOM data. Multiply that by hundreds of concurrent studies across a health system, and you're looking at terabytes per hour.

Architecture Overview

Our streaming pipeline is built on three core principles:

• Chunked streaming — We break studies into variable-size chunks and stream them through /api/v2/imaging/stream with sequence tracking.
• Edge processing — Data is processed and cached at 40+ edge locations.
• Speculative pre-fetch — ML model predicts which studies a radiologist will view next.

Connection Multiplexing

We moved from one-connection-per-study to multiplexed streaming. Our SDK maintains 16–32 concurrent HTTP/2 streams per connection. Upload throughput improved by 13x, p50 latency dropped from 180ms to 12ms.

Results

Today we process over 2.4 billion images annually with p95 latency of 47ms. During peak hours, we sustain 800+ Gbps of aggregate throughput across our edge network.