RSNA 2018: How AI can Help, but also Hack into Medical Imaging

The use of Artificial Intelligence (AI) and web-based services have largely benefitted the world of medicine. From sharing medical records and diagnostic images to transferring critical reports, the internet has immensely helped patients all over the world. But, the increasing use of such technology has also ignited security concerns.

A group of researchers based in Switzerland believes, increased connectivity can expose the medical community to several vulnerabilities and threats. It can even put patients’ lives at risk due to data theft. The findings of the study were presented in the RSNA annual meeting held in Chicago. Here, researchers demonstrated how medical images can be hacked with AI if precautionary measures were not implemented by health care service providers as well as computer software and hardware companies.

Hacking Medical Images with AI

The researchers, under the leadership of Anton S Becker, MD Radiology resident at University Hospital Zurich and ETH Zurich, showed how an AI algorithm can be used to tamper mammograms. For the study, an AI-based application was prepared with 680 screenings from 334 patients.

The cycle-consistent generative adversarial network (CycleGAN) was used to test its ability to change cancer-specific features from the mammograms. The network used healthy images and included cancerous features in them and vice versa. Radiologists then studied the images to find out whether they could easily identify tampered images. However, none of them were able to correctly distinguish the modified images from the untampered ones.



Cyber attacks with AI

Becker’s study concluded that AI could successfully modify radiological images. CycleGAN successfully learned how to detect breast cancer cells. With this newly acquired knowledge, it could remove the cancerous lesions from the diagnostic radiological image. It could also replace them with healthy tissue. Similarly, it was also capable of turning a normal image into a cancerous one. According to Becker, this could be a cause of concern in the immediate future. He believes, with the growing advancement of AI in clinical procedures, cyber attacks cannot be ruled out. A patient’s medical images can be altered for life-threatening consequences and may give rise to a new form of cybercrime.

Prevention is Key

The main objective of the study was to create awareness about potential threats. Although AI is playing a crucial role in major treatments, its misuse cannot be ruled out. Therefore, Becker cautions physicians as well as manufacturers of AI technology and recommends the implementation of preventive measures.

Investment in the IT department of all hospitals and health care service providers can go a long way in stopping such nuisance. Secure patient portals can drastically reduce any untoward incident and therefore, hospitals must make an effort to ensure safety. Use of encryptions and digital watermarks on medical images can offer better security. Access to patient portals must be stringently regulated to prevent any misuse of data.

sepStream® is committed to providing the best EMR/RIS/PACS solutions. We help to digitize the workflow from ordering to the final report stage and seamlessly integrate it according to your requirements. Our solutions are 100% web-based and can be securely accessed from anywhere. We are committed towards customer satisfaction and offer the best solutions at extremely nominal rates.