How AI works
For the past three years, medical imaging in Isala has worked closely with AIDOC, a company specializing in artificial intelligence. Rogier: ‘They primarily focus on prioritizing diagnostic imaging in emergency radiology. In large hospitals, for example in America, there are often long lists of CT scans that have not yet been seen. So it is important that the radiologists first see the images of patients in the emergency department (ED) with abnormalities that cannot wait too long. We use this application, but it is less important to Isala because we usually review all CT scans done in the ER within an hour. Still, we were interested in working together when that question came up.’
The system works as follows. For example, when someone with a head injury comes to the emergency room, a CT scan is often done to rule out bleeding or fractures. Martijn: ‘The images are immediately visible to us, so we can assess them. In addition, they are sent directly (anonymized) to AIDOC. Within 10 minutes of receiving all the images, I also receive a message from AIDOC on my screen. If the AI module has detected a bleed, I see on the screen where that bleed is and I know where to look. We also receive a notification when the computer has not seen any abnormalities. You can of course always check it again yourself. You want to be one hundred percent sure.’
It takes a lot of time and research to run an AI program “perfectly” in your own hospital. Martijn: ‘One of the problems at the beginning was that it took too long for us to receive the AI images again, sometimes more than 30 minutes after the scan was done. It should be faster. We were also not initially notified if no abnormalities were found. But how can you be sure that the computer has looked? So that has also been adjusted.’
Radiology is now using AI for possible brain bleeds, cervical vertebral fractures and pulmonary embolism. Rogier: ‘Everyone sees large bleeds in the brain or fractures in the neck vertebrae. The computer picks out the small bleeds or tears that patients may suffer from. It is certainly a real added value during the evaluation of the CT images during the night hours. AI keeps us on our toes. During our research into the use of AI, we had the computer evaluate, among other things, 2500 neck scans. Some of these scans showed fractures. The computer pulled almost all of them out, but eventually also saw fractures that the doctors doing the scan at the time hadn’t seen. This almost always involved very small tears that the patient most likely never had a problem with. The abnormalities that the computer had not found turned out not to be fractures, but displacements of the vertebrae’.
The program also often removes the small pulmonary emboli, says Martijn. ‘We will therefore also use AI for all contrast scans of the lungs. Oncology patients, for example, have a greater risk of developing pulmonary embolism. When you do a lung scan to assess how the tumors are doing, that is your focus. If the computer then shows a possible pulmonary embolism, it is only a gain and even better care.’
AI also helps the radiologists with busy shifts. Martijn: ‘When the service is busy and we receive several scans from different places at the same time, it requires one hundred percent attention. AI helps us react more quickly and adequately.’
But will it be computers that will completely take over healthcare? ‘No, that’s not going to happen’, reply Martijn and Rogier. ‘A computer cannot take the final responsibility, the doctor will always do that. AI helps us do our work better. All studies show that radiologists and AI perform better together than separately. Now that it’s getting busier in healthcare, including in the emergency room, the extra pair of eyes is a must. AI is one of the solutions to responsibly manage the increasing demand for care.’