When the X-ray is a deepfake. The Power of RNN Neural Networks

Today, with specific malware, it is possible to modify X-ray images by creating or deleting tumors. This has already happened and will continue to happen in an increasingly digitized healthcare system.

What happens if a radiologist is fooled by a hacker who has managed to modify the diagnostic images? The downside of artificial intelligence applied to the medical field.

One Thursday morning in early spring, a radiologist carefully observed the X-rays of his 33 patients, looking for traces of lung cancer. It is difficult to imagine, however, that the chest X-rays he is looking at are not the original ones, but the result of an accurate photo-retouching process, which was carried out perfectly thanks to artificial intelligence.

This is not science fiction. The news dates back to last May, then confirmed and detailed in a scientific paper written by an Anglo-Japanese team: a massive cyber-attack in which hackers have modified diagnostic images – even 3D images – succeeding in deceiving radiologists in 98% of cases. In some X-rays a tumor had been added, in others an existing neoplasm had disappeared completely. The secret to the success of the attack was to insert the malware in the right place, i.e. in the data transmission system between the diagnostic equipment and the doctor’s computer, so that neither the technician nor the radiologist could see that something was going wrong. An automated system based on a neural network did the rest.

Instead of lung cancer, the same system would have made it possible to create ghost fractures, hide a potential infection, accentuate arthritis or make a heart problem that could be diagnosed by that same image disappear. And if we hypothesize that a team of hackers could devote themselves to medical deepfakes – a technique that consists of superimposing images with artificial intelligence to create fake ones – the potential purposes can be endless: from insurance fraud to terrorist attacks, up to targeted and premeditated murder.

In January of this year, in the U.S., Google Health researchers announced that their artificial intelligence could beat a radiologist in the accuracy of breast cancer diagnoses. Criticism of this result has not been spared, especially because, according to radiologists, the real challenge today is not to identify the tumor masses, but to define how insidious they are and at what stage of disease progression they correspond.

Controversies aside, the entry of digital technology into healthcare is something that is already happening, but we still do not know what impact it will have on our health. The disruption due to artificial intelligence still has uncertain outcomes, and the only sure event is that it will disrupt the entire architecture of health knowledge.

However, the change is already underway because we have collected a huge amount of health data, but their use is potentially null and void due to the difficulties of linking them together and processing them with appropriate tools. Healthcare facilities are gradually equipping themselves with specialized systems and resources for data analysis and machine learning, but the difficulty of integrating the systems, connecting the results to obtain something useful remains high. In parallel, the concept of disease itself, as well as the range of diseases and treatments, is becoming much more complex and articulated, and with the ageing population the picture is even more complicated.

Which are the impacts of these changes on the professionals working in the health sector? The encounter between artificial intelligence, genomics and clinical practice, generates the demand for new professional figures, with staff specialised in assessing whether or not the content of the big data collected is clinically relevant. Moreover, the processing of data opens up new bioethical and legal scenarios, especially because there is a legal vacuum on these issues.

Leaving aside for a moment the predictions on technological developments, we have to consider the role of health regulations – everything is regulated by strict certifications and procedures and it is also on these more formal aspects that the game is played. In the next two or three years, we will feel an increasingly need to have clinical professionals and researchers more specialized and trained in technology.