Share this article on:
There is growing concern about hackers gaining access to medical devices and conducting attacks to cause harm to patients. Now malware has been created that can add fake tumors to CT scans.
The malware is not being used in real-world attacks. It has been created by researchers at the Ben Gurion University Cybersecurity Center in Israel to demonstrate just how easy it is to exploit vulnerabilities in medical imaging equipment.
In addition to adding tumors to medical images the malware could be used to remove real tumors. The former could be conducted for political reasons such as preventing a candidate from running for office, the latter would prevent individuals from receiving treatment for a life-threatening illness. The technique could also be used for insurance fraud, sabotaging of medical trials, and cyber terrorism.
Prior to a patient being prescribed radiation therapy or chemotherapy additional tests would be performed and the incorrect diagnosis would be identified, but patients would still be caused considerable emotional distress. The removal of tumors to make the patient appear healthy could have much more serious implications. Treatment could be delayed until a point when it is too late to be effective.
The researchers used a deep learning neural network called a generative adversarial network to alter the CT scans intercepted by the malware.
The attack scenario demonstrated by the researchers would require a man-in-the-middle device to be built and physical access to a hospital. The device could be planted close to the scanner, such as at night when there is less chance of detection. With the device in place it would be possible to intercept CT scans and manipulate them at will.
The researchers created such a device from a Raspberry Pi 3 which was connected to a USB to Ethernet adapter. Both could be purchased for around $40.
The device was loaded with the Raspbian OS and was configured as a network bridge and set up as a WiFi access point. Once connected to the network, the device was capable of intercepting scan data as it was sent to the PAC system. The attacker had full control over scan data and could alter it at will and create or remove any number of tumors while retaining the same anatomy as the original scans.
But how effective is the malware at altering CT scans? Were the alterations good enough to fool trained radiologists?
In tests, 70 images were manipulated. The accuracy of the alterations was such that it was possible to fool three radiologists in 99% of instances where fake tumors were added and 94% of images where real tumors were removed. The altered images fooled AI systems every time.
When the radiologists were made aware that scans had been altered, in a second test using a mix of genuine and doctored images, they were still fooled by 60% of the images that had tumors added and 87% of images where tumors had been removed.
In the tests, the researchers used lung scans and injected fake tumors, but brain tumors could be created or removed just as easily and the system could be used on a wide range of health conditions such as bone fractures, blood clots, or spinal problems.
The alteration of images would be difficult to detect as scans are typically not encrypted nor digitally signed. Healthcare organizations are usually good at implementing robust perimeter controls to prevent attacks from remote threat actors but are less good at protecting internal networks. This eggshell approach to security leaves hospitals vulnerable to attacks conducted inside the facility by malicious insiders.
A video of the simulated attack can be viewed on the following link: https://youtu.be/_mkRAArj-x0