The report of a CT scan can change a patient's life. Consequently, it is essential that it is interpreted correctly by the radiologist, and that the subject's privacy is protected to the best. Israeli researchers have published one study which demonstrates how fragile hospital safety systems are, and how easy it is to obtain and modify radiology data. Moreover, their experiment also showed their malware's ability to deceive experienced doctors.
The video in the article shows how a system of deep-learning can add or remove nodules from a chest CT scan, simulating the presence of a tumor or (worse) its absence. This potential could be easily exploited by hackers, who would be able to organize insurance scams, damage the career of politicians running for the election or simply attempting to attack the life of unsuspecting patients, who would not receive the necessary medical follow-up in the presence of a cancer.
The installation of the system, moreover, seems very easy, above all considering the poor security in the average hospital and the ease of entering the premises at night.
in blind study furthermore 70 TACs have been shown altered by the malware to three expert radiologists, who proved to be completely incapable of discovering the additions or removals of the program. The doctors were not able to distinguish the changes even when they were aware of the fact that some sequences could be false. Even a software used for the reporting of images was not able to discriminate, almost always wrong. All this is understandable considering the accuracy of the modification, which is really very realistic and can easily mislead even the most trained eyes.
A real bucket of icy water, which forces us to reflect on the safety of health data and take action as soon as possible.