Interview with Siemens Healthineers
»Generative AI is the next big thing in radiology«
Despite pioneering in digitalization and AI, radiology needs even more efficient workflows to combat rising case numbers and a shortage of specialists. Florian Reinhold from Siemens Healthineers explains how GenAI benefits clinical processes, radiologists, MTAs and patients.
»GenKI not only has the potential to increase efficiency, but also to improve diagnostics and patient care,« says the Head of Digital and Automation Product Marketing at Siemens Healthineers. The aim of the medical technology manufacturer is to reduce the workload of routine tasks for radiologists and medical technical assistants (MTAs) and increase productivity.
According to Reinhold, the Erlangen-based global corporation is focusing on four central application areas in GenKI: the consolidation and processing of patient data, the optimisation of scans and protocols, support in diagnostic procedures and the improvement of communication between radiologists, patients and referring physicians.
»Generative AI is based on the ability to generate new information from existing data. In radiology, we can therefore extract usable information from patient histories and other unstructured data, which then flows into the diagnostic process,« says Reinhold, explaining the fundamental benefits of large language models. Siemens Healthineers is pursuing »an integrative approach that combines both generative and pixel-based AI«. This makes it possible to create individualised patient protocols and thus significantly improve image quality and diagnostics.
| Generative AI in Clinical Use: Lung Cancer Screening |
|---|
|
A patient who has become conspicuous during a lung cancer screening can be diagnosed more comprehensively than before through the use of generative AI. Analysing the patient data enables the entire patient history to be taken into account and the subsequent CT scan and the necessary procedures to be individually tailored to the patient. The scan protocols and radiation values are optimised to the patient's specific needs and previous illnesses, which potentially leads to higher quality images and a more precise diagnosis. After the scan, classic »pixel-based« AI algorithms can then automatically generate findings that the radiologist only needs to check and confirm. GenKI is intended to provide an optimal process here, with the aim of a more precise and individualised diagnosis and ultimately a more holistic therapeutic approach. The generative AI is also intended to serve as a »translator« that translates medical terminology into a language that the patient can understand. Medicine knows that patients who really understand their findings are also more likely to accept the therapy and its healing measures. For radiologists, the use of conventional AI has already significantly reduced the manual workload; generative AI promises a further reduction in workload as well as increased efficiency and accuracy in reporting. |
For Florian Reinhold, a key advantage of generative AI lies in the optimisation of scan settings. By analysing the patient's history, the scan can be individually tailored to the patient, resulting in higher image quality and more precise diagnoses. For example, in the case of contrast agent allergies, it is possible to automatically switch to alternative examination methods, says the product expert. This not only reduces the risk for the patient, but also increases the efficiency and accuracy of the examination.
Even Better, Automated Findings
Not new, but much smarter with GenKI: For Siemens Healthineers, generative AI plays a decisive role in the creation of findings in addition to the technical optimisation of scan settings - in combination with pixel-based AI. By pre-processing and analysing image data from MRI or CT devices, preliminary findings can be generated automatically.
»Pixel-based AI generates the best images,« says Reinhold. »And the best images can only be generated if the patient history is also included via GenKI.« This means that radiologists can be relieved of even more time-consuming routine tasks, with more time for more complex diagnostic issues. For Siemens Healthineers, generative AI acts as a supporting tool that will further rationalise clinical workflows in the future and increase their productivity.
Another aspect that the German medical technology flagship is focussing on when using GenKI is improving communication between radiologists, patients and referring physicians. According to Florian Reinhold, »patients can be better informed about their state of health by generating comprehensible and visualised findings«. This not only promotes understanding and acceptance of the diagnosis, but can also lead to better compliance. It also makes it easier for referring physicians to interpret the findings, which optimises interdisciplinary collaboration.
| Classic AI | Generative AI |
|---|---|
| Technical image and scan quality | Inclusion of the patient history |
| Image recognition of lesions and other indications and changes | Optimised scan and recording settings |
| Automated measurement data and values | More effective report creation and workflows |
| Post-processing for incidental findings | Automated doctor's letters |
| Guidance for medical professionals | Improved patient communication |
Regulatory and Technological Hurdles
However, the introduction of generative AI in medicine in general and in radiology in particular faces regulatory and technological challenges. According to Reinhold, Siemens Healthineers relies on its own LLMs for development, but also utilises existing models available on the market.
»Our innovation department always tries to filter out what makes the most sense for the respective project.« In the case of the collaboration with Essen University Hospital, the jointly developed algorithms delivered a better result for the intended use case than ChatGPT, for example. »Above all, the models need to be well trained. Not only for the correct provision of information, but also to rule out hallucinations or discrimination against individual patient groups, for example.«
The authorisation of new, AI-based medical devices and software solutions requires extensive proof of safety and efficacy as well as country-specific adaptation of the software stacks to regional laws, regulations and data protection guidelines. »The regulatory hurdles are a major challenge, especially due to the strict data protection guidelines in Europe,« says Florian Reinhold. The medical technology giant is relying on close cooperation with regulatory authorities and, above all, a local approach and hospital partnerships for the compliant and safe use of AI.
»We are planning to train the algorithms directly on site in the clinics so that the data does not have to leave the hospital. We are also working on updating our scanner software so that it automatically applies the new, optimised protocols.« Seamless integration into existing hospital information systems is also essential, according to Florian Reinhold.
When Will GenKI Come Into Clinical Practice?
While generative AI has been in the world since 2023, the steps and authorisation procedures required for use in medical devices, hospitals and therefore 'on patients' take time. »We are conducting extensive studies to prove the benefits and safety of our technologies. This is complex and takes time, but we are well prepared and work closely with the relevant authorities,« emphasises Florian Reinhold with regard to device and patient safety. Pixel-based algorithms have already been approved for Europe and the USA, and the Group is currently working on their approval for China. »There won't be a big bang,« says Reinhold, outlining the gradual transfer of the GenKI models into radiological practice. »But we are also working intensively with collaboration customers and hope to certify the first implementations in the next few years.«
Siemens Healthineers sees enormous potential for the use of generative artificial intelligence, precisely because radiology is currently facing enormous challenges. According to Reinhold, these include a higher number of examinations due to an ageing population and the corresponding massive shortage of skilled labour, which will become even more acute in the coming years. For the product expert, technology could make a decisive contribution: »We need new approaches. Generative AI stands for individualised and more precise diagnostics and should relieve the burden on radiologists and MTAs and improve patient communication«.
In the long term, the Erlangen-based global corporation not only wants to improve the efficiency and accuracy of diagnostics, but also promote good working conditions for medical professionals and the attractiveness of radiology jobs. »Our aim is to automate routine tasks so that specialists can concentrate on more complex and value-adding activities,« says Reinhold, explaining the resulting economic benefits for the clinics. In addition to the clear benefits for doctors and hospital processes, generative AI will also score points through the so-called soft factors, concludes Reinhold: »For patients, we hope for faster diagnosis and treatment as well as better chances of recovery.«
