Trendreport MedtecLIVE 2024

When will artificial intelligence replace doctors?

AI has had a major impact on medical technology in recent years – both in the development and manufacture of medical products as well as in diagnostics and therapy.

But what impact does AI have on the medical role? How can ethical aspects be taken into account when an AI system makes medical diagnoses and treatment decisions? And what about the protection of sensitive patient data?

Artificial intelligence (AI) is able to analyze large volumes of medical data and identify patterns. Its use in medical technology promises great potential for many areas. Not just in diagnostics – AI can already improve processes and quality in the manufacture of medical products. AI will influence the entire value chain of medical technology in the future - a topic that this year's MedtecLIVE from June 18 to 20, 2024 in Stuttgart has also put on its agenda. “We are experiencing a rapid development of AI, are guiding the regulation issue in Europe regarding the AI Act and are highlighting specific fields of application and regulatory framework conditions for AI at MedtecLIVE - in diagnostics as well as in production,” says Christopher Boss, Managing Director of MedtecLIVE GmbH and Executive Director of the event.

“The complexity of processes and products has increased significantly in many areas, including medical technology and biotechnology, and is becoming increasingly difficult for humans to master intellectually. One solution is offered by intelligent and networked sensors and sensor systems that use artificial intelligence to process data into information and derive well-founded decisions from it,” says Dr. Thomas Velten, Head of Innovation Management at the Fraunhofer Institute for Biomedical Engineering IBMT. So, in which areas of medicine is AI already supporting people?

On the one hand, AI-driven prediction models are already being used during the production of medical devices to preventively identify potential device failures. AI also supports quality control in production by analyzing large amounts of data and identifying patterns that could indicate quality problems. On the other hand, there is a wide field of application for artificial intelligence in medical devices themselves – for diagnostics and therapy: distributed patient data sets are analyzed using swarm learning, and AI is already being used successfully for automatic image recognition and in laboratory medicine.

AI-based prediction models for increased reliability of medical devices

The use of artificial intelligence in the production of medical technology, for example, has an impact on the reliability of medical devices. In particular, the use of AI algorithms for the preventive identification of potential device failures increases operational safety. These advanced models continuously analyze large amounts of device data to detect anomalies and irregularities that could indicate potential failures at an early stage. This proactive monitoring identifies impending defects before they can lead to more serious problems.

The power of AI in medical systems for pattern recognition has already been proven and there are many reasons to analyze data from medical devices using AI algorithms. The quicker and higher availability of devices has a positive impact on patient safety and also optimizes operational processes in the healthcare sector.

For example, AI-based prediction systems are used in Siemens Healthineers products to predict the outcome of software tests with a high degree of accuracy. The Fraunhofer Institute for Cognitive Systems IKS is also conducting research into trustworthy artificial intelligence and is dedicated to developing reliable AI-based predictions. The institute integrates cause-based prediction models for causal analyses to enable well-founded decisions. Time series analyses are used to evaluate historical data series in order to create reliable, AI-supported predictions for future developments. These approaches are used in various areas, such as predictive maintenance of production facilities. Methods for robust, reliable and early predictions of future events are being developed.

AI algorithms for quality inspection

AI algorithms help to reduce human error in production and increase the consistency of quality inspection. By implementing AI in quality inspection, manufacturers can not only ensure product quality, but also reduce costs by minimizing scrap and increasing manufacturing efficiency.

The automated quality inspection AI.SEE™ by manufacturer elunic, for example, combines the intelligence of artificial neural networks with state-of-the-art image processing. Its modules are equipped with smart cameras and AI-based evaluation tools that, according to elunic, observe, learn and adapt. Each component passes AI.SEE™ during production, without any manual effort and completely automatically, where every small deviation on complex surface structures is detected. Deep learning algorithms can reliably detect even tiny defects on heterogeneous or reflective surfaces, which is often not possible for the human eye.

This automated test is used in medicine, for example in diagnostics, in the evaluation of microscopic images, where it examines the smallest anomalies in cell morphology that may indicate disease or can also count cells or bacterial colonies. 

Automated image analysis through AI

Together with development partners and supported by several clinical partners, the Fraunhofer Institute for Biomedical Engineering IBMT is developing an intelligent and interactive assistance system for ophthalmologists in the joint project “Ophthalmo-AI”. This system uses methods of explainable artificial intelligence to create comprehensible diagnoses and treatment suggestions. The aim is to help ophthalmologists make a correct diagnosis and the best possible treatment decision based on image data and clinical information. “The AI system first identifies biological structures and pathological features in the image data in order to generate comprehensible suggestions for the medical staff. Special AI models then derive diagnoses from the image findings and other patient data, make therapy suggestions and predict the success of the therapy. The doctors' knowledge is integrated into the process through interactive machine learning,” explains Velten from the Fraunhofer IBMT. “Extensive and processed treatment data is used in a special data integration platform to develop the system. And the extensive data-driven processing fully complies with the data protection aspects of the GDPR.”

Thomas Velten © Fraunhofer
Thomas Velten, Fraunhofer IBMT © Fraunhofer IBMT

Siemens Healthineers also uses AI algorithms as part of its MRI portfolio. Some of its MAGNETOM MRI scanners, for example, use algorithms for automated patient positioning. In this way, the movement of the heart will also be automatically recognized in the future, thus avoiding the time-consuming application of electrodes. In the Syngo Breast Care diagnostic solution for mammography screening, AI algorithms help to assess individual lesions more accurately and reduce the number of false positive findings – thus avoiding unnecessary invasive diagnostics. To further simplify the assessment of mammograms, Syngo Breast Care provides an automatic classification of the probability of breast cancer.

According to the Prague-based company carebot, doctors using artificial intelligence are able to detect breast cancer at an early stage. Their AI application AI MMG automatically analyzes mammographic images and creates a quantitative assessment of breast density. In this way, the AI helps radiologists to improve diagnostic accuracy and reduce the variability of their assessments. It therefore supports the early detection of breast cancer. According to carebot, with its AI solution, lung cancer can also be detected earlier, increasing the patient's chance of surviving the first year to more than 87 percent. In the fourth stage, it is already less than 19 percent. Without the use of artificial intelligence in practice, a doctor's ability to detect early-stage lung cancer on an X-ray when the tumor is still very small is very limited, and in most cases practically impossible.

In many cases, automatic image recognition using AI therefore increases patient safety. Faster analysis of medical images improves diagnosis and speeds up treatment planning.

Use of AI in medical laboratory

The field of laboratory medicine is a very important area of application for AI and offers great advantages and support for doctors, including for imaging procedures: at Essen University Hospital, for example, a system was trained for pulmonary fibrosis using a self-learning algorithm. After just a few learning cycles, the computer made a better diagnosis than a doctor, as the computer does not forget what it has learned and is superior to the human eye when comparing patterns, according to Johannes Haubold, Head of the Clinic for Diagnostic Radiology at the hospital.

A study by the University of Nijmegen in the Netherlands also shows that AI systems outperform humans in cancer diagnostics, particularly in terms of time: a team of developers with their own AI software solutions as well as a group of pathologists was asked to detect cancerous tissue. The best AI system achieved almost 100% detection accuracy and was significantly faster than a pathologist, who needed 30 hours to recognize the affected samples. Research is also being carried out in dermatology at Heidelberg University using an AI system that scans patients' skin to distinguish cancerous melanomas from moles. The hit rate in an evaluation of over 100,000 images is already 95 percent.

Siemens Healthineers uses self-learning algorithms in its Atellica Solution for laboratory diagnostics. Their Drawer Vision System (DVS) visualizes each individual sample tube. For example, user errors in handling the sample tubes can be reliably detected by automatic comparison with an image library comprising more than 60,000 images.

Another example of the use of artificial intelligence is the cultivation of 3D cell aggregates (spheroids) in a bioreactor. “Temperature and CO2 content are already controlled as standard in the incubator, but parameters are not usually measured directly in the cell culture (liquid). This means that no statements can be made about the pH value, oxygen content or other values relevant to cultivation. Furthermore, the quality control of the spheroids has so far only been carried out from time to time and outside the bioreactor,” says Fraunhofer Innovation Manager Velten. “As part of the Fraunhofer ZSI, sensors are being developed for direct integration into the bioreactor's cell culture tubes. This enables online measurement of various parameters that are relevant to produce 3D cell aggregates, such as oxygen, pH value, glucose and lactate. In terms of online process control, the sensors are read out wirelessly and the sensor data is processed and analyzed directly on site and almost in real time by AI-based evaluation software.” An additional continuous optical analysis of the spheroids, which also uses AI methods to evaluate the camera images, then provides information about the size, shape and number of spheroids. According to Velten, this enables a direct correlation between the cell culture parameters and the condition of the spheroids. In the future, this should not only significantly improve quality assurance in the production of 3D cell aggregates, but also optimize the cell culture and thus increase production yield.

Data aggregation through swarm learning

A key challenge in the application of AI is that it relies on data in order to be trained. In Germany, patient and health data is considered extremely sensitive and is strictly protected. Merging data from different clinical institutions, studies or projects into a single database to effectively train an AI is practically unfeasible. For this reason, an alternative solution has been developed for collecting and analyzing patient data. Within the enormous amount of data on patient symptoms, the hope is to identify key components for innovative, customized therapeutic approaches. “Collecting medical data can benefit the development of new, better therapies. The medicine of the future will be networked using artificial intelligence and everyone involved will benefit equally,” says Prof. Joachim Schultze, spokesperson for the research consortium and Director of Systems Medicine at the DZNE, the German Center for Neurodegenerative Diseases. “Patients report their symptoms, we measure a few more things and then try to identify which disease all this could indicate. We are now underpinning this process with a great deal of very precise data. However, we have so much data that we need technical help to recognize the patterns.”

Joachim Schultze, DZNE © Frommann
Joachim Schultze, DZNE © Frommann

In collaboration with several German research centers, the DZNE has therefore developed an AI-based evaluation system based on swarm learning. The AI technology makes it possible to analyze distributed data sets. “We have developed a technology together with the IT company Hewlett Packard Enterprise. We call it Swarm Learning and see it as a game changer when it comes to dealing with big data.” Schultze and his team carried out an analysis of thousands of medical data records in a specific application area. These were X-ray images of the lungs and molecular patterns in the blood from various sources. With the help of swarm learning, the artificial intelligence was able to identify pathological changes in the lungs and diagnose leukemia, tuberculosis and COVID-19. In the future, an AI trained in this way could potentially also take on a supporting role in analyzing brain scans or X-rays. Although doctors cannot be replaced by AI, it is still an extremely useful tool for their daily work. And swarm learning helps to further increase the performance of AI.

How safe is our data?

In addition to legal feasibility due to regulatory limits, there are ethical and social issues in the context of AI applications in medicine. Processing biomedical data with ethical and data protection issues is not easy. Collecting data and storing it centrally is usually difficult to implement given the legal regulations on data protection. “Data protection and trust are key in medicine. Digital data is worth protecting, especially personal data,” says Schultze. “With swarm learning, for example, the data protection requirements are fully met because the actual data is not exchanged.” This means that the data is not collected but remains stored locally on the individual devices.

The assessment of compliance risks in AI-supported systems and medical devices is also carried out by experts at TÜV SÜD. “The assessment of AI in medical devices by TÜV SÜD is always an important step in ensuring the quality and safety of such products. TÜV SÜD is one of the leading bodies for the assessment of AI in medical devices and medical applications. The assessment is based on the requirements of the respective conformity assessment procedures and the requirements of the state of the art", says Dr. Abtin Rad, Global Director Software Product Assessment at TÜV SÜD. This ensures that the data used to develop and validate the AI model is of predetermined and verified quality and complies with applicable data protection regulations.

Dr. Abtin Rad, TÜV SÜD
Dr. Abtin Rad, TÜV SÜD 

AI in compliance monitoring

Compliance monitoring, i.e. the monitoring of and adherence to regulatory standards, is extremely important in highly regulated areas such as medical technology. Under certain circumstances, artificial intelligence can also make this process more reliable. By using AI algorithms, companies can continuously analyze data or documents and identify patterns to detect potential violations.

For example, has developed a regulatory compliance solution that uses purpose-built AI models to automatically monitor the regulatory environment for relevant changes and align them with your internal policies. The system ensures that relevant regulations and requirements are tracked and reported on promptly. Their software is designed to minimize risk and reduce costs by automatically monitoring regulatory updates.

Dr. Abtin Rad from TÜV SÜD, a body responsible for the assessment of AI in medical devices and medical applications, disagrees: “In our opinion, it is not expedient to evaluate or test one autonomous system by another autonomous system.” Compliance risks are assessed by experts.

Addition or replacement for doctors?

The use of AI in medicine cannot yet be seen as a replacement for doctors, but rather as a supporting tool that improves medical activities. The targeted integration of AI systems can increase the efficiency and accuracy of medical diagnoses and therapies, ultimately improving the quality of patient care.

Christopher Boss ©NürnbergMesse.jpg (1.4 MB)
Christopher Boss, MedtecLIVE GmbH © NürnbergMesse

The upcoming MedtecLIVE, which will take place in Stuttgart from 18 to 20 June 2024, will also focus on artificial intelligence and digitalization. “At the upcoming trade fair, we are placing a strong focus on the future-relevant topic of AI in medical technology. In a constantly evolving world, it is crucial that we shed light on the latest advances in this field. Our trade fair will therefore provide a platform to explore the groundbreaking developments and pioneering importance of artificial intelligence in medical technology,” says Christopher Boss, Managing Director of MedtecLIVE GmbH and Executive Director of the event.