Healthcare Innovation’s recent Northeast Virtual Cybersecurity Summit featured a discussion with Benoit Desjardins M.D., Ph.D., an associate professor of radiology and medicine at the University of Pennsylvania, who has recently published research on the cybersecurity of medical images. Desjardins is a reformed hacker from Canada, who pursued a career in medicine after graduate studies in artificial intelligence and mathematics.
As a clinician and ex-hacker, what made you renew your interest in hacking and cybersecurity?
In 2017, I attended DEFCON in Vegas. When I went to attend a small evening session on cybersecurity in healthcare, 500 people were in line at the door waiting to get in. They moved us to a larger room and gave us an update on recent cyberattacks like Wannacry and NotPetya, and everything else going on in the field. This was fascinating and I decided to renew old interests. I then attended several conferences and completed several certifications in cybersecurity to bring my knowledge and hacking skills up to date. I also started interacting with some of the leaders in the field and with the cybersecurity team at my institution.
You recently published research on the cybersecurity of medical images. What can you tell us about the context for that research?
In the past 10 years, there have been over 3,000 reported breaches of medical records, including the Anthem breach, which included 78 million records. But these breaches involved mostly general medical records. Medical images are stored in different servers and are relatively safe from breaches, although this was safety through obscurity. Medical images use the DICOM standard, proposed in the 1980s for integration between different imaging devices from multiple manufacturers. Medical images are not stored on the electronic medical record, but rather on DICOM servers, which have different IP addresses and different access methods.
Have there been recent breaches of medical images?
Over the past two years, three different cybersecurity research groups have performed controlled breaches of DICOM servers: one from the Mass General Hospital, one from McAfee, and one from Greenbone Networks. They used either the Shodan search engine or did an exhaustive search of worldwide IP addresses for DICOM servers. They found that thousands of DICOM servers were unprotected and their images could be easily accessed from anywhere on the planet. News of this went all the way to Congress.
Then last year, a group from Spain found a way to embed malware in DICOM images and a group from Israel found a way to add lung nodules on intercepted CT scan images being transmitted from a CT scanner to a DICOM server. So I decided to undertake a thorough study of the exploitability of DICOM images and DICOM servers.
What kind of team did you assemble for that study?
I assembled an elite team of the best people on earth for that topic, including the cybersecurity researchers who had recently found the vulnerabilities I just talked about, as well as the two co-leaders of the DICOM Security workgroup, responsible for adding all security features to the DICOM standard. We did extensive brainstorming and a thorough analysis of the ways in which the DICOM standard could be abused and published our results and recommendations in one of the top radiology journals.
What major cybersecurity issues did you find in the study?
We found many ways to exploit the DICOM standard and DICOM servers, relating to the confidentiality, integrity and availability of medical images. I’m happy to elaborate on three of them: issues with data at rest, issues with data in transit and data integrity issues.
What are some issues with data at rest?
There are several problems with the security of data at rest on DICOM servers, coming from both outside and inside medical institutions. A minority of DICOM servers are poorly configured and can be accessed remotely. Not all of those offer the same level of access to their images. But even if a DICOM server is properly protected from access from outside an institution, hackers can simply walk into hospitals and connect their laptop to the hospital network, from a network jack in a hallway or a patient room and access the DICOM servers. The DICOM standard includes features to protect data on media and in emails, including encryption, but does not provide for encryption of data on DICOM servers. Furthermore, most of the encryption features are not implemented.
What about issues with data in transit?
There are several problems with the security of data in transit — for example, on network connections between CT scanners and DICOM servers. One of the co-authors on our paper is part of the Israel group that found a way to intercept DICOM images acquired by CT scanners on their way to the DICOM server and used an artificial intelligence approach to either add nodules or subtract nodules on chest CT scans. Most radiologists were fooled by the tampered images. This hack was made possible because of lack of encryption on network connections. The DICOM standard includes features for encryption of data in transit using TLS, which is implemented in most systems, but not always used.
What about data integrity issues?
There are several problems with the lack of integrity check in transmitted DICOM images. The current DICOM standard includes an image “creator digital signature” field which can be populated by the device acquiring the image, for lifetime integrity check. But this is often not implemented by the manufacturers, or if they are implemented, are simply not used in practice. I already mentioned the violation of image integrity when lung nodules were added to CT images in transit. Another example is also from one of the co-authors on our paper from the group in Spain who found a way to embed malware into DICOM images. He embedded malware into DICOM private attributes in the DICOM header and replaced the DICOM preamble by the header of windows executable files. Any hack involving remote command execution could trigger that malware and take over a computer system.
What are the main recommendations you made related to the security of medical images?
The use of encryption and digital signatures require a system of keys and certificates that are complex and require a lot of overhead. We need to figure out how to store, acquire and recover keys, and how to authenticate people requesting keys. DICOM security leaders need to integrate modern technologies to more easily deal with encryption keys and certificates.
Many security features in the DICOM standard have never been implemented by manufacturers because they thought medical images were safe at rest and in transit. We now know they are not, and this is a powerful incentive to implement all DICOM security features.
New technologies to manage encryption keys and certificates will help implementation of these features. Once an image is generated by a scanner, the “Creator Digital Signature” of the created DICOM file should be populated by the scanner. And every time that image is transmitted to another medical device, this signature should be checked, and warnings issued if it is incorrect or missing. Image validators should also be implemented to verify the internal consistency of DICOM images.
What about recommendations for local IT experts?
Local IT experts need to carefully monitor their networks for any suspicious activity from the outside or the inside. They need to properly authenticate all users sending and requesting images, limit access of images to only legitimate users, limit the network visibility of their DICOM servers and imaging devices, and use already implemented DICOM features to securely transmit images within the institution and with outside institutions. They should implement rate limiters to prevent denial of service attacks and should disable CD auto-loading on all radiology workstations.
What about recommendations for radiologists and technologists?
Radiologists are already primed to detect errors in medical images. For example, if the chest radiograph of a male patient looks normal, but on the next radiograph the patient has grown breasts, then radiologists instinctually know that this is probably the radiograph of a different patient which has been mislabeled. But if the next radiograph shows new lung nodules, then they do check if it is the same patient but should also consider in the back of their minds the possibility of corrupted or tampered images. They should also make sure to maintain confidentiality at all times, by encrypting laptops containing imaging data and never sending such data over public networks. And if they are handed a compact disk for a curbside consult at the imaging workstation, they should not load it in their workstation as it could contain malware.
How has the COVID pandemic affected the practice of radiology and what are the cyber implications?
The pandemic had considerable impact on medical practice, especially for remote work. At my institution, consultations by telehealth between doctors and patients went from 50 a day to 7,000 a day. Most radiologists have very limited interactions with patients. For a radiologist, the only differences between working at home and working at the hospital is that it’s a little bit slower and a lot lonelier at home. But radiologists must keep security in mind. The radiologist’s home workstation connects to a home router, which connects via the internet to the hospital VPN device, which itself connects to the hospital servers. Each of those points is vulnerable in many different ways. Radiologists must make sure that their home equipment has not been compromised. In particular they need to change their router’s default admin password, to avoid their router from being hijacked by hackers. This has happened with thousands of home routers, which can redirect links to hackers’ websites and can intercept data
