Fraunhofer Institute for Cognitive Systems IKS
Fraunhofer Institute for Cognitive Systems IKS
Grüne Tropfen vor blauem Hintergrund
© iStock.com/in-future

AI in Healthcare

© iStock.com/in-future

Artificial Intelligence in Healthcare

Digitalization, automation and artificial intelligence (AI) are rapidly changing the healthcare sector. In clinics, hospitals and doctors' offices, electronic health records (EHR), data management systems, AI-supported evaluations, predictions and resource planning, robot assistants in the OR, intelligent assistants and many other technologies are on the rise. Doctors, healthcare professionals and patients are increasingly being supported by cognitive systems - from the initial telemedical consultation and AI-supported diagnosis to individualized therapy and aftercare at home. Digitally networking distributed patient data, public health data and data from health apps and smart wearables is the basis for individualized and optimized healthcare services.

How is AI used in healthcare? The digital patient journey

In the future, digital medicine and AI will accompany us as patients: From the prevention, screening, AI diagnosis and therapy to aftercare. This means that AI can be used to support patients and medical staff in every aspect of the patient journey.

Graphic about the digital patient journey
© Fraunhofer IKS

Challenges for trustworthy AI in healthcare

Medical AI promises great potential for many fields of application, for example in medical diagnostics, drug development, administration and process management in hospitals and doctors' surgeries, resource and capacity planning, patient education and the training of healthcare professionals.

In order to use AI in healthcare, various technological and organizational challenges must be addressed appropriately, from the database and algorithm development to the practical application of AI systems.

The database...

... has a significant influence on the quality of the AI system and is often the most time-consuming part of an AI project. Even before the actual algorithm development, collecting and preprocessing the data creates the necessary input to train and test the AI.

The quality of the algorithm...

...is comparable to the known differences in quality between technology products

The use of AI...

...must be evaluated from use case to use case. The rule here is: it depends. Even a high-quality AI algorithm cannot always be easily transferred from one context to another. And the use of AI is not equally sensible and feasible for every use case.

  • Small amounts of data ("little data")
    require special training and testing approaches in order to develop trustworthy AI models, e.g. in the case of rare diseases.
  • Multimodal data
    often adds complexity to clinical decision making and requires specialized AI processing methods.
  • Distributed & particularly sensitive data
    sensitive data often cannot be "simply" made available for the development of AI models, but require decentralized methods for secure data processing such as federated learning.
  • Data availability & quality 
    pose a major challenge in the case of rare diseases, for example, due to the scarcity of data.

  • Explainability of AI
    even for specialists is not always given if suitable technical methods are not used to understand which data and factors are decisive for the AI's decision.
  • Uncertainty & bias
    are often the result of training on incomplete or inaccurate data, which can lead to uncertainty in the results of the AI model.

  • AI proofs of safety
    are particularly important for critical application areas in order to ensure the reliability, quality and explainability of AI decisions.
  • Unknown scenarios
    occur in reinforcement learning when the model is used outside the 'closed world' in which it was trained. Such cases can be identified via out-of-distribution detection.

Research on AI in healthcare at Fraunhofer IKS

Our focus: Trustworthy digital health

At Fraunhofer IKS, we conduct research in the following areas, with especial focus on development of trustworthy AI-based systems in safety-critical areas, such as healthcare. ​

  • ​Optimizing patient journey: from screening and diagnosis to treatment and follow-up care
  • Medical decision support and time series
  • Clinical decision making based on causal inference ​
  • Robot-assisted hospitals ​
  • Data-efficient medical image processing in imaging and diagnostics
  • Optimization of healthcare processes, such as hospital resource management ​
  • Predictive maintenance  of medical devices​
  • Visual quality inspection of medical devices ​
  • Practical applications of quantum computing in healthcare​

Our AI in medicine services

Data-efficient medical imaging

with safe and explainable AI models for data scarcity and small sample sizes.

More info

Medical decision support & time series

with explainable and reliable AI for improved decision making, disease prediction and treatment.

More info

Optimization of healthcare processes

with scalable, transferable AI systems to facilitate manual tasks for healthcare professionals.

More info

Validation & verification of AI models

with our verification framework to ensure safety and trustworthiness.

More info
 

© iStock.com/yul38885

Our services

  • Idea generation workshops
  • Rapid prototyping
  • R&D
  • Trainings
Services

Digitalization opportunities in medicine

Digitalization generates huge amounts of data. Industry 4.0 demonstrates how this data can be used effectively. Data is networked to monitor processes, to identify trends at an early stage and to be able to respond to them using new business models. These advantages can also be applied to medicine. While taking into account all the medical and non-medical data, it becomes possible to make efficient and rational decisions, individualize therapies or detect diseases at an early stage.

Big Data and artificial intelligence (AI) are important keywords in the medicine of the future. AI can combine and analyze large amounts of data in a very short time, faster than humans ever could. This paves the way for intelligent applications in the fields of:

  •  Clinical decision-making
  •  Robot-assisted surgery
  •  Medical imaging and diagnostics
  •  Chronic disease monitoring
  •  Hospital Data Management

At Fraunhofer IKS, we conduct research on the validation of digital applications in these safety-critical areas.

Medical imaging and diagnostics

By analyzing image data, artificial intelligence can assist in medical diagnostics. Based on existing image data and associated diagnoses, patterns in the image are recognized by the AI, which are assigned to disease patterns. The analysis and availability of large amounts of data makes it possible to detect pathological changes in the image quickly and reliably, to individually adapt therapies to the patient and to provide prognoses concerning the further course of the disease.

We are also working on practical applications of quantum technology in medicine. You can find more information about this on our blog.

Robot-assisted surgery

Robotics in the operating theater is anything but science fiction. Today, robotic assistants are already being used in a wide variety of operations or in hospital logistics for the distribution of medications. Robotics enables higher precision, better visualization and a minimally invasive approach to surgery.

Medizinischer Roboter im OP
© iStock.com/ClaudioVentrella

Safeguarding of safety-critical AI applications in medicine

Holistic safety concepts are required to ensure the safety of users of digital health services. Intelligent systems collect and combine data and make decisions based on it. 

This decision-making process is complicated by two factors:

  • The availability of the data: Artificial intelligence is trained to recognize patterns using processed data sets. In the healthcare sector, these data are sometimes scarcely available, e.g. in the diagnosis of rare diseases.
  • Intransparency and traceability of AI decisions: Complex AI algorithms and their decisions, for example in the area of diagnostics, can only be checked by specialists with difficulty because it is not clear which data are of fundamental importance for the decision.

False diagnoses and prognoses constitute a major risk in the application of cognitive systems in the field of medical technology. At the Fraunhofer Institute for Cognitive Systems (IKS), we conduct research to make intelligent cognitive systems reliable and secure.

Our research encompasses several focal points:

  • Safeguarding perception of cognitive systems in medicine.
  • Dynamic security proofs of digital health services
  • Uncertainty estimations of the AI outputs
  • Explainability and comprehensibility of cognitive systems
  • Application of quantum computing in medicine

Research focus at Fraunhofer IKS

Safeguarding the perception of cognitive systems in medicine

Artificial intelligence recognizes patterns in images and sensor data that have been previously learned from training data. By means of this perception, i.e. interpretation of this data, objects are recognized and the environment perceived. Reliable and robust perception is a critical safety factor not only for the diagnostics, but also for the perception of surgical robots. In order to be able to act safely and avoid administration of the incorrect treatment and injuries to patients, the exact positioning of the procedure in the body is essential.

Dynamic proofs of security for digital health services

The regulatory demands upon the safety and performance of medical devices are high. Intelligent systems constantly further develop and build on their capabilities. In doing so, they continue to be a black box, i.e. it is not clear to the outside world how decisions are made. A dynamic and continuous proof of safety is necessary in order to be able to certify artificial intelligence. At Fraunhofer IKS, we are researching solutions to improve the explainability, transparency, and robustness of neural networks and to provide quality and proofs of safety of AI.

© iStock.com/Widewingsstudio

Determination of the uncertainty of the AI output

We use estimations of uncertainty to teach AI to doubt things in order to avoid making incorrect decisions. In the medicine field, there is often only a small amount of low quality data available for the training of cognitive systems. This may result in a situation whereby no clear decision can be made on the basis of existing data. In such situations, it is important that the cognitive system is aware of this uncertainty and draws attention to it, for example, to avoid misdiagnoses.

© istock.com/tavani

Explainability and comprehensibility of AI

Under the keyword of Explainable AI, Fraunhofer IKS is researching solutions to make the basis for decision-making in diagnostics comprehensible and verifiable for users. In doing so, we combine AI technology with classical software development.

Quantum computing in medicine

In a joint project with the LMU Munich Hospital, we are researching into solutions on how quantum computing and artificial intelligence can be used for intelligent diagnostics and healthcare.