Artificial Intelligence in Radiology

Basically, AI is a set of learning algorithms that are fed with a large data set (e. g. CT images), which are categorized by human input (e. g. by diagnosis). After several learning rounds the AI is then able to identify a given disease or condition (e. g. a fracture) with a high reliability (usually above 99.9 %). The precision can even be increased if additional data (e. g. symptoms) are added to the program. The medical expert may then confirm or correct the proposed diagnosis of the system. AI is a powerful tool that is about to change many facets of our modern lives.

Pre- examination die Untersuchung

Artificial Intelligence (AI) can help by reducing the workload of radiological staff in various fields in different ways. For instance, it can choose the optimal time for an appointment with a patient. It accomplishes that by calculating the factors of availability of the clinic and the patient as well as accumulating data about similar examinations. As such, there is no need to adjust the equipment after every patient. Before the appointment the AI – remembering all the right sequences – can recalibrate the equipment to the patient. AI, thus, helps to optimise logistics in clinics and surgeries.

During the examination

After an image is taken AI can immediately check whether it fulfills all criteria set by the radiologist. If not, a more suitable picture can be taken and the patient does not have to come to the clinic yet another time.

Maybe the most prominent and crucial field of AI medical application is diagnostics. For instance, AI can detect acute pathologies even during the computer tomography as well as produce insights into cerebral infarction diagnostics. The AI is provided with a huge amount of high-quality pattern-recognition data during the programming learning phase (prior to its registration for medical use). This helps the radiologist to make the diagnosis and, in the end, to find the correct treatment for the patient.

After the examination

Eventually, an application based on AI can help the doctor and the patient to communicate, to ensure effective aftercare.

Does the implementation of AI into medicine need to be promoted? There are many certified medical applications based on AI on the current market. However, they are still rather hesitantly applied in radiological clinics. Nonetheless, radiology is the medical field in which AI is most prominent.

Germany is one of the last developed countries to promote the digitalization of the health care market, despite trotz, ungeachtet the huge amount investment. A milestone on the way to a more digitalized healthcare system is the Digital Supply Act, which came into effect in November 2019. This law allows doctors to prescribe healthcare apps. Additionally to providing online consultation privately, it is now possible to share the provision of these options. Also, e-prescriptions are promoted as well as sending electronic sick leave notes. These digital services are paid for by the statutory insurance .

In the EU, it is only allowed to train the AI until the point of accreditation. This makes continuously-learning AI impossible in medical applications. Continuous Learning would be achieved through constant interaction between radiological staff and the AI, thus, constantly updating one local AI system instead of acquiring a standardised, up-to-date system by installing new systems when a new update is available.

On the other hand, stopping the training phase of the AI algorithm after registration enables quality control of the software by the developer. Therefore, the reason why continuous learning is prevented is the risk of adding possible false values into the program. Thus, the relatively small or erroneous data set of one radiological clinic may deteriorate their AI system and lead to wrong diagnoses. Until it is possible to gather all provided information on one central server which transmits information to all learning AI systems in a controlled way, it is not practical to accredit continuously-learning AI applications.

Ultimately , there are no established (legal) guidelines regulating who is responsible in the case of a false diagnosis by the AI and a subsequent false decision by the radiologist. Another difficult case would be that the radiologist might make the wrong diagnosis even though the AI suggested the correct diagnosis. Hearing a second or third opinion may help, or lead to legal complexity.

Even though there are still several scientific, medical and legal issues that need to be discussed when it comes to an extensive use of AI applications in the clinic, we should regard the technology as a powerful tool to assist physicians and patients in curing diseases. We may keep in mind that AI is not developed to make doctors redundant or even to completely substitute their diagnostic expertise. AI can support them by automating standard procedures and helping them to focus on diagnosing and curing diseases, and interacting with patients.

EXERCISE: 1. AI does not reduce the workload of radiological staff in various fields.

a) True b) False

2. AI can support radiologists after the image is taken, because it can immediately check whether it fulfills criteria set by the radiologist. If not, another picture can be taken immediately and the patient does not have to come to the clinic unnecessarily.

a) True b) False

You’ll find the answer on the next page

BIBLIOGRAPHY 1. Maryellen L X, Giger L, Min J K: Artificial Intelligence in Medicine 1st Edition: Technical Basis and Clinical Applications, Academic Press 2020. 2.http://dict.leo.org 3. www.dict.cc 4. www.webster-dictionary.org

CARMEN LOBITZ Carmen Lobitz ist freie Trainerin, Lehrerin und MTA. Sie gründete 1998 in Berlin lobitz seminare, die unter anderem Englischseminare für Health Professionals anbieten. Kontakt: mail@lobitz-seminare.de www.lobitz-seminare.de

###more###

Lösung für Exercise: 1. B 2. A

Artificial Intelligence in Radiology

Online-Angebot der MT im Dialog

Stellen- und Rubrikenmarkt

Artificial Intelligence in Radiology

Sie wollen den ganzen Text lesen?

Online-Angebot der MT im Dialog

Stellen- und Rubrikenmarkt