Digital Diagnostics and Artificial Intelligence

Artificial intelligence, or AI, can combine and analyze large amounts of data in the shortest time faster than any human. How can diagnostics in the field of medicine use this advantage for themselves and who benefits from it?

On the protection of data, current uses of AI and what is still possible in the coming years in digital diagnostics, learn in the interview with Dr. Ilja Hagen, Cluster Manager Healthcare at Biopark Regensburg GmbH.

Ilja, what is actually meant by digital diagnostics?

Dr. Ilja Hagen: The term digital diagnostics currently usually refers to the use of digital technologies, in particular artificial intelligence and Big Data analyses for evaluating large volumes of medical data or digital health applications, such as health apps and wearables.

The entire field of digital imaging, such as computed tomography (CT), magnetic resonance imaging (MRI) and telemedicine for remote patient monitoring, is also part of digital diagnostics, which has comparatively also been established for some time.

The goal of digital diagnostics is always to support and improve medical diagnosis. The speed can be increased by digital diagnostics, the precision or scalability can be improved and the evaluation methods can be automated by algorithms, so that the workload for the individual doctor or physician is reduced.

Are there any application examples for areas in medicine that currently benefit from digital diagnostics?

First and foremost are all imaging procedures that are at the forefront of the application of artificial intelligence, i.e. radiology, i.e. digital X-rays, computer tomography (CT), scans, magnetic resonance imaging (MRI) and ultrasound images. These allow high-resolution body images that enable rapid diagnoses and repeated analyses.

Another imaging technique, but one that deals with a different topic, is from the field of pathology. This involves the analysis of microscopic images such as cross-sectional images. Due to the new digital analysis methods, on the one hand there is an increased volume of samples and data, on the other hand this represents an additional workload for the pathologist. Artificial intelligence supports the pathologist in the evaluation by, for example, making a pre-selection in the cell annotation, and thus quantitative image analysis results in a precision in diagnostics that is clearly superior to visual assessment by eye.

Is there a concrete example, perhaps from the working group "Digital Diagnostics" for the use of digital diagnostics?

A regional example from gastroenterology is the early detection of esophageal cancer. In the laboratory Regensburg Medical Image Computing (ReMIC) of the East Bavarian Technical University Regensburg, the topic of heartburn or reflux disease was processed. When the reflux of gastric acid becomes chronic, it is also a cause of Barrett's esophagus. It is a tissue damage of the mucosa, which is also an increased risk for the development of esophageal cancer. In this case, the chances of survival for patients are very poor, as the disease is usually diagnosed at a very late stage. Earlier diagnosis by endoscopic examination is difficult because the damage is often not visible. Artificial intelligence, or deep learning, allows physicians to detect this mucosal damage ideally at a stage before this cancer has developed. This is a diagnostic that was not possible before.

A second example from cardiology is textile-integrated sensor technology from the Fraunhofer Institute IIS in Erlangen, which is also a member of the working group. This technology enables continuous heart monitoring, which can use AI to detect heart disease, for example, that previously often went unnoticed even with a 24-hour electrocardiogram (ECG).

It is precisely because of this diversity that artificial intelligence is one of the key technologies in the healthcare sector. At the "Digital Diagnostics" working group, which we launched together with Bayern Innovativ almost a year ago, we meet monthly with representatives from research institutions and companies and exchange ideas on this topic. It is amazing how different the development stakes are in the community and how we learn from each other. For example, unusual markers are used in diagnostics that were not previously common, or voice analysis is used to detect infectious diseases. Another example is the determination of vital parameters through an optical analysis of the face using a camera. All these parameters would not be evaluable without AI.

On the subject of data protection and security: how is it ensured that medical data is protected in digital diagnostics?

As with all other data, encryption and access control is a crucial data security measure that must be ensured. Here you have to decide for yourself what kind of repository you want for your data: a regional database or a supra-regional cloud. However, I cannot say more about this, as I am not an expert in this field.

Sensitive health data should, however, be anonymized and pseudonymized as far as possible. That is, an identification of patients is more difficult, although just at the same time the analysis for diagnostics and also for research is possible. However, this is not possible for all data, such as genetic data. These cannot be anonymized quasi by definition, because each gene data set can uniquely identify a person.

In light of the electronic patient record, patient consent is essential. It must be possible for any patient to refuse clearance for sensitive findings, for example mental illness, so that they are not visible to any other physician or healthcare actor. On the other hand, it would be helpful for the best possible medical research and then accordingly the care of patients if as much data as possible were available. This is the only way to learn and to develop new diagnostic procedures efficiently.

Where does AI support the medical profession, and what role does it take when working with AI systems?

Generally, any form of AI in diagnostics is a supportive tool. The decision of treatment always lies with the physician. Through AI, the possibility is given to optimize the workflow, to make the flood of data more processable for the doctor or to point out findings that he himself might not have seen so, because an algorithm, more precise, more efficient, perhaps more sensitive than a human doctor in his analysis of findings. This is like the usual four-eyes principle or getting a second opinion. With the use of AI, four eyes or two opinions become thousands or millions, respectively, allowing for a better diagnosis to be made.

These "thousand eyes" can also likely provide medical professionals with appropriately life-saving guidance on rarer conditions where they have little or no experience?

Absolutely. In the course of comprehensive care or even in telemedical care, AI can support here very well. No doctor can have seen every finding several times a year in every hospital in the area. The attending physician would probably also consult an expert from another larger hospital or refer the patient there. AI systems can then make it possible to make a finding more quickly.

What might the future of digital diagnostics look like? In your opinion, what developments or innovations are conceivable in the coming years?

The integration of different data sources is one of the next logical or necessary steps. After all, the examples given always come from a single indication. To establish the connection of different diagnostic markers with each other, in order to achieve an even better and even more precise diagnosis, is certainly the area that will develop strongly. To achieve this, however, health data interoperability must also be optimized, i.e., the interaction of different systems must be improved. Suppliers and manufacturers of devices are in demand here, but the infrastructure must also be improved.

In addition, the integration of previously unused data sources, i.e., so-called quantified self, will be an issue. People will certainly use more wearable devices in the future and monitor their health and activity, so that the data can then also be used diagnostically and early warnings for health problems will be made possible.

Another step is the improvement of personalized medicine. To enable increasingly truly personalized healthcare, it is necessary to combine diagnostics with various omics sciences such as genomics. In this way, a patient's predisposition to certain diseases can be identified and treatment plans can be customized based on his or her diagnostic and genetic profile.

Early diagnostics and predictive diagnostics are certainly challenges that are still furthest in the future. When it comes to detecting diseases before they have broken out, AI can hardly help at present. Examples include neurodegenerative diseases such as Alzheimer's, Parkinson's, dementia or multiple sclerosis. The disease state is not known until the disease has occurred. So the question is, how can I tell that a person's movement patterns are changing, for example, before they get Parkinson's? Again, there are ways to train AI accordingly, but that's still a long way off. Very, very large amounts of data and very, very many patients are needed to actually arrive at meaningful algorithms at this point.

The interview was conducted by Dr. Petra Blumenroth, Project Manager Technology I Frugal Innovation at Bayern Innovativ GmbH.

If you have any questions about the topic or the "Digital Diagnostics" working group, then talk to our expert Julia Ott.

Listen to the full interview as a podcast: