Receiving DICOM from a hospital over the internet

  • Plain DICOM C-STORE assumes a trusted local network and has no required transport encryption.
  • A DICOM AE title is a self-declared 16-character name and does not prove sender identity.
  • Two main patterns exist: an on-premises relay inside the hospital, or an internet-facing mutual TLS connection.
  • Mutual TLS requires both the server and the sending device to present and verify certificates before data moves.
  • DICOMweb STOW-RS lets a cloud application receive studies as a standard HTTPS multipart/related POST.

You can receive DICOM studies from a hospital over the internet, but you need more than a public IP address. You need encrypted transport, real sender authentication, and a path that does not depend on opening inbound ports at the hospital.

That matters if you build a cloud PACS, an AI diagnostic platform, a teleradiology service, or any product that needs studies from hospital equipment. A local DICOM setup is familiar. A hospital-to-cloud setup adds network and security work that you must design on purpose.

Why is receiving DICOM from a hospital different over the internet?

DICOM was designed for a trusted local network. The core transfer protocol, C-STORE from the DIMSE message set, assumes both sides can open a direct TCP connection. A scanner sends to a PACS. A PACS sends to a workstation. Inside the hospital, that often means port 104 or 11112.

Move the same C-STORE association to the public internet and 3 things change.

First, plain DICOM has no required transport encryption. Patient imaging data cannot travel across the internet as clear TCP traffic.

Second, a DICOM Application Entity (AE) title is not strong identity. It is a self-declared 16-character name. It helps route a DICOM association, but it does not prove who is connecting.

Third, the network path is harder. Hospitals usually will not open inbound firewall ports for each vendor. Your cloud service also should not depend on custom routing work for every site.

The good news is that none of this makes internet delivery unusual or experimental. The hard parts are known: protect the transport, identify each sender, and keep the hospital connection outbound from the hospital side.

Option 1: place a relay inside the hospital

Some vendors solve the problem with an on-premises relay. It lives inside the hospital network, receives C-STORE locally, and forwards each study out over HTTPS.

This can be a good fit when the hospital requires local software, or when you already have a support team for site installs. The DICOM sender stays on the local network, so scanner configuration looks familiar to the PACS team.

The trade-off is operations. You now ship software into someone else’s network. You handle OS patches, local outages, remote access, firewall changes, and upgrades at every site. One hospital is manageable. Many hospitals turn it into a long-running support job.

If you choose this route, make the relay boring. Give it one job, document the hospital handoff, and test upgrade recovery before the first real study arrives. That keeps the pattern workable instead of fragile.

Option 2: use an internet-facing mTLS DICOM endpoint

The other pattern is to expose a public DICOM endpoint that a hospital scanner or PACS can connect to directly. There is no relay inside the hospital. The sender opens an outbound connection to your endpoint.

This is only safe when the DICOM connection is encrypted and both sides authenticate. Mutual TLS (mTLS) does that. The server presents its certificate. The sending device presents its own client certificate. Both sides verify each other before DICOM data moves.

If you want the protocol detail, read how mutual TLS works for DICOM connections. If your team is comparing network models, the useful question is often whether to use site-to-site VPN tunnels or internet-facing endpoints.

I would usually pick the internet-facing mTLS pattern for a cloud product. It avoids per-hospital VPN projects and on-premises software. You give up some low-level control over the network path, so your certificate process and sender onboarding need to be clear.

How does C-STORE become something your app can receive?

Your product probably should not run a DICOM listener unless DICOM networking is your core business. Most cloud teams would rather receive an HTTPS request, validate it, and pass the study into their own pipeline.

That is where DICOMweb STOW-RS helps. STOW-RS is a standard HTTP-based way to store DICOM instances. Instead of your app accepting a raw DIMSE association, it receives a multipart/related POST. For a deeper comparison, read C-STORE vs DICOMweb STOW-RS.

The failure you will hit in production is not the first test file. It is the 30th sender, the expired certificate, or the hospital that retries after your HTTPS endpoint returns an error. You need a clear handoff from DICOM status to HTTP status, plus enough monitoring to see which sender is affected.

Building that bridge yourself costs engineering time and years of on-call ownership. A managed DICOM gateway such as Dicomly can handle the internet-facing mTLS endpoint. It delivers each instance to your HTTPS URL as a standard STOW-RS POST, so your team works on the product itself. The trade-off is less direct control and a monthly service cost. It also does not parse, de-identify, or fix DICOM tags for you.

What changes in your own architecture?

With this model, your application receives DICOM like a webhook. It handles HTTPS. It does not need to negotiate DIMSE associations, manage transfer syntax negotiation, or keep a PACS-style listener online.

That does not remove your medical imaging responsibilities. You still decide where studies go, how long you keep them, how you process tags, and how your product handles failed deliveries. The boundary just becomes clearer: DICOM networking on one side, your HTTP pipeline on the other.

This is a useful split for a software team. Your developers can build with normal web tools while still accepting studies from real hospital systems. The old DICOM parts stay at the edge. Your product logic stays in code your team already knows how to operate.

What should you do next?

Pick one hospital sender flow and write down the exact handoff: calling AE title, called AE title, host, port, certificate owner, and HTTPS destination. Then decide whether you want to run that edge yourself or use a gateway. If you are still comparing patterns, start with the VPN vs internet-facing endpoint guide and choose the simpler path for your first real sender.

Dicomly builds infrastructure for healthcare software teams that receive DICOM studies from hospitals. Why we built Dicomly →

Sources

  1. DICOM Standard Part 15 (PS3.15, Annex B — Secure Transport Connection Profiles) (dicom.nema.org)
  2. IHE ITI Technical Framework Volume 1 — ATNA Profile (profiles.ihe.net)
  3. RFC 8446 — The Transport Layer Security (TLS) Protocol Version 1.3 (www.rfc-editor.org)
  4. RFC 5382 — NAT Behavioral Requirements for TCP (www.rfc-editor.org)

Start receiving DICOM today.

Early access. No credit card. First endpoint free.