The Digital Transformation in Hospitals and Medical Practices

The Fraunhofer Institute for Digital Medicine MEVIS develops innovative software solutions for more precise diagnoses and more effective therapies. The possibilities are as varied as the challenges: Artificial intelligence and self-learning algorithms will bring enormous advancements to the clinical routine in hospitals and medical practices. Simultaneously, these solutions call for new structures, for example, in data privacy and interdisciplinary cooperation.


“Most of today’s medical data is gathered in digital form, including MR and CT images, laboratory values, or genetic data,” says Horst Hahn, institute director of Fraunhofer MEVIS. “However, these data have not yet been sufficiently connected with each another, meaning that not all of the information contained within can be used.” Fraunhofer MEVIS is working on software-based systems that can help accomplish this, including software platforms that refine MRI and ultrasound diagnoses, optimize minimally invasive therapies, or provide targeted assistance to determine the best medication. The goal is precision medicine personalized to each individual patient.

The journey towards this goal is filled with challenges. The glut of available medical data and the growing number of therapy options offer new opportunities, but also contribute to the growing complexity of everyday medical life that must be managed. “Computers can help manage this complexity,” Hahn emphasizes. “Using methods from artificial intelligence (AI), for example, we want to achieve medical care that can act with more precision, integration, and efficiency”

Today’s oncologists, for instance, are hardly able to memorize all available knowledge regarding the detailed differences between different tumor types. A digital support system will accompany them in the future: Based on self-learning pattern recognition, such a system could, for example, detect similar past cases according to unusual disease patterns. Valuable information could then be extracted: What treatments were used at the time, and what were the results?


Detecting hidden patterns

Medical AI systems should be able to analyze numerous clinical parameters simultaneously, including blood values, image data, and ECG measurement data. The algorithms are likely to detect patterns that, because of their highly complex nature, are inevitably invisible to the naked eye. These patterns may improve, for example, the early detection of certain tumors and thereby increase the chances of recovery.

High-quality data is the main prerequisite for data-driven methods to function reliably. “Garbage in, garbage out,” is how deputy institute director Matthias Günther sums it up. “Even the best algorithms can’t turn low-quality data into pure gold.” It is also important to monitor image acquisition automatically, for example, during an MRI scan to detect and correct imaging errors as quickly as possible. Fraunhofer MEVIS has already developed algorithms that promise exactly this.

A further challenge is that different clinics and equipment manufacturers often collect data according to varying standards. “This causes problems, for example, when comparing image data within a multicenter study,” explains Günther. “This is why we are advancing methods to standardize such processes and ensure image quality at the time of image acquisition.” Data privacy must be guaranteed by methods such as decentralized learning: Patient data should not leave the clinics; the algorithm should come to the data and be trained on it.


The digital patient model

Carefully handling data is fundamentally essential for the digital patient model, a concept that likely will play a significant role in the future. “Ideally, everything we know about a particular patient flows into this mathematical model,” says Tobias Preusser, former deputy institute director of Fraunhofer MEVIS, “This might include different image data, various lab values, and DNA genotypes and phenotypes.” This computer model then acts as a sort of digital twin: To identify the best treatment, clinicians can play through various therapy scenarios virtually.

“For example, this could help evaluate whether a patient is a candidate for chemotherapy,” said Preusser, “or even a combination therapy. The evaluation of therapy options on the basis of a virtual patient model has the potential to spare the patient from unnecessary risks and side effects.” It is also plausible to use digital patient models in clinical studies, for instance, when developing devices or drugs.

In some cases, this might eliminate the need to perform an experiment on animals. “A digital model of the entire patient is still a dream for the future,” says Preusser. “However, we can already model individual organs such as the liver, parts of the heart, and some joints for planning operations.”

"With all these new tools it is important that it is not the machine that makes the decision," emphasizes Horst Hahn. “Making a decision must always be reserved for medical professionals and their patients.” The statements made by AI must, therefore, always be as transparent and comprehensible as possible, and sources of error and uncertainties in the statements must be clear. “One thing seems evident,” Hahn believes. “The use of AI systems will change the structures of medicine.” The various disciplines are likely to come closer together. To exploit this new technology, some hospitals are already considering establishing departments for the new discipline of artificial intelligence.