Site-to-site VPN vs internet-facing endpoints for DICOM
- Site-to-site IPsec VPNs route hospital DICOM traffic as if both networks were private, but need per-site setup work.
- VPN tunnels can hit private IP range conflicts, needing NAT rules, plus rekeying and dead-peer-detection failures.
- Internet-facing mTLS endpoints authenticate senders by certificate instead of relying on private network routing.
- Legacy modalities that cannot speak DICOM TLS still need a local TLS bridge or a site-to-site VPN as a fallback.
- mTLS scales better across many hospitals since each sender gets a certificate instead of a custom network design.
If you need to receive DICOM from hospitals, you have 2 common network choices. You can build a site-to-site IPsec VPN for each hospital, or expose a DICOM endpoint that requires mutual TLS (mTLS).
Both can encrypt traffic across the public internet. The main difference is where you want the operational work to live: network routing and tunnel support, or certificate-based device authentication and a public DICOM listener.
How do site-to-site VPNs work for DICOM?
A site-to-site VPN creates an encrypted tunnel between the hospital network and your cloud network. IPsec defines the security architecture in RFC 4301. Internet Key Exchange version 2, defined in RFC 7296, negotiates security associations and session keys.
After the tunnel comes up, both networks route packets as if they share a private network. CT scanners, MR scanners, or a PACS router can send C-STORE requests to your private DICOM listener. Many systems still use port 104 or 11112 for this traffic.
This is a familiar model for hospital IT teams. It also keeps old senders working when they cannot use TLS themselves. If a site already has a strong VPN process, this path can be easy to approve once.
Where do hospital VPN tunnels get hard?
VPNs become painful when you connect many hospitals, not when you connect the first one. Each site usually needs security review, exchanged settings, firewall changes, and a maintenance window. That can turn one integration into several weeks of coordination.
You can also hit private IP conflicts. Many hospitals use ranges such as 10.0.0.0/16 or 192.168.1.0/24. If 2 sites overlap with your network or each other, you need Network Address Translation (NAT) rules before a DICOM association can even start.
The failure mode is also awkward. Phase 1 or Phase 2 rekeying can fail, a firewall can clear state, or dead peer detection can decide the tunnel is down. The scanner may just queue studies locally until someone checks the tunnel.
These problems are solvable. Treat each hospital like a small network project, keep a clear runbook, and monitor the tunnel from both ends. The trade-off is ongoing network ownership, not just the first setup.
If your team does not want one VPN project per hospital, a managed gateway such as Dicomly’s endpoint model can handle public mTLS receiving. Hospitals connect outbound to an internet-facing endpoint, so you avoid a site-to-site VPN per hospital. You give up some low-level control and add a monthly service cost, but your engineers spend more time on the product itself.
How do internet-facing mTLS endpoints work for DICOM?
An internet-facing DICOM endpoint accepts TCP connections over the public internet. It still requires encryption and sender authentication. DICOM Part 15 describes secure transport profiles, and RFC 9325 gives current TLS and Datagram TLS guidance referenced by secure deployments. The IHE ATNA Profile describes authenticated node-to-node communication for healthcare systems.
With mTLS, both sides present X.509 certificates during the TLS handshake. The hospital sender checks your server certificate. Your endpoint checks the sender’s client certificate before DICOM association negotiation begins.
That changes the security boundary. You no longer rely on private IP routing to identify a sender. You identify the sending device by its certificate, then allow or reject the DICOM association.
This is still standard DICOM work, not a custom protocol. Our guide to mutual TLS for DICOM connections explains the certificate flow in more detail.
Site-to-site VPN vs internet-facing mTLS endpoint
| Dimension | Site-to-site IPsec VPN | Internet-facing mTLS endpoint |
|---|---|---|
| Security boundary | Network layer, IPsec | Transport layer, TLS 1.2 or TLS 1.3 |
| Typical onboarding work | Network tickets, routes, firewall rules | DNS name, port, Application Entity (AE) title, client certificate |
| IP address conflicts | Common with private ranges | None for the receiver |
| Failure to expect | Tunnel down, rekey failure, stale routes | Certificate mismatch, blocked outbound port, TLS configuration error |
| Fixed infrastructure cost | Often one gateway or tunnel cost per site | Usually tied to the endpoint and traffic model |
| Legacy sender support | Works with senders that only speak raw C-STORE | Needs DICOM TLS support or a local TLS bridge |
If I were building a new cloud imaging product, I would choose internet-facing mTLS first. It scales better when you add hospitals because each sender gets connection details and a certificate, not a private network design.
The honest downside is compatibility. A scanner from the 2000s may only speak unencrypted C-STORE over raw TCP. For that device, a VPN or a local TLS bridge may still be the practical answer.
What about legacy modalities that cannot use TLS?
Some older ultrasound, radiography, or workstation software cannot start a DICOM TLS connection. They can still send normal C-STORE inside the hospital network, but they cannot authenticate with a client certificate themselves.
You have 2 realistic options. First, place a local TLS bridge inside the hospital network. It receives raw C-STORE locally and sends outward over mTLS. A well-known tool such as Stunnel can play this role, or the site may prefer a managed appliance.
Second, use an IPsec VPN for that hospital. This is not a failure of the architecture. It is a compatibility decision for one sender or one facility.
The useful part is that you do not need one rule for every hospital forever. Use mTLS where the sender supports it. Keep VPNs as the exception for devices that cannot be changed.
How should the endpoint feed your application?
Network security only gets the study to your edge. Your product still needs to receive, validate, store, route, or analyze the images. Most web teams do not want a full DICOM Message Service Element (DIMSE) stack inside every application service.
A common pattern is to accept C-STORE at the edge and deliver each instance to your application as DICOMweb STOW-RS. That gives your product a normal HTTPS request path while keeping DICOM compatibility for the sender. The difference between those protocols is covered in DICOM C-STORE vs DICOMweb STOW-RS.
Running that listener and translation layer yourself means handling sender setup, certificate rotation, delivery failures, and on-call support for years. Dicomly’s delivery guide describes the managed version: each received instance reaches your URL as a standard HTTPS POST in STOW-RS format. The trade-off is that the gateway stays a narrow connectivity layer; storage, viewing, de-identification, and tag changes remain in your pipeline.
What should you do next?
Pick the next hospital sender you expect to connect. Ask whether it supports DICOM TLS with a client certificate, and note its Called AE title and outbound firewall rules. Then read the guide on receiving DICOM from a hospital over the internet. Use it to map that sender to either mTLS or VPN before you design the full integration.
Dicomly builds infrastructure for healthcare software teams that receive DICOM studies from hospitals. Why we built Dicomly →