Everyone has something to say about artificial intelligence and yet everyone has their own idea about what it means. The future of the healthcare sector does not belong to “Dr Algorithm”, but to self-learning, intelligent decision-support systems that help doctors, holistically factor in the patient, and enhance the quality of medical care.
Photos: Sebastian Gabriel
Artificial intelligence, or AI, is no longer exclusively a topic for computer nerds. Newspapers and magazines devote in-depth reports to the issue, often with reference to medicine. Many consider “Dr Algorithm” to be the future of healthcare. But is that really the case? Will we be generating algorithms instead of training medical students?
“AI won't do away with jobs”
During the German Federal Government's Digital Summit, at the Siemens Healthineers symposium, “Quantum Leap or Hype: AI in Medicine & Medical Care”, Chris Boos, CEO of Arago and a member of the government's Digital Council, dispelled some common misconceptions: “Machines do not understand anything. They have nothing to do with brains. Artificial intelligence is more than just machine learning. And it won't put everyone out of a job.”
Essentially, it is about using AI to make currently inflexible software programs more flexible. That sounds dry, but it actually has disruptive potential: “About 80 percent of everything can be automated using AI,” said Boos. Basically there is a need to distinguish between “narrow AI” and “general AI”. The former refers to applications that train (self-learning) algorithms to carry out very specific tasks that the algorithm then fully masters. The more ambitious “general AI”, on the other hand, trains algorithms to solve a range of problems, thereby lowering the overall amount of training required.
Volkmar Weckesser, MD, CIO of CentoGene, illustrated the healthcare areas destined for AI solutions using the example of patients with suspected rare diseases. The company from Rostock in Germany has developed a medical care and research workflow that extends from laboratory analysis to clinical management of patients by the treating physician. AI algorithms support doctors in collating clinical data. They are used to identify new biomarkers and will correlate biomarkers with quality of life data in a patient app in the future. “I'm certain that as a result we will make significant progress in identifying rare diseases, and will also be able to help with finding new treatments,” Weckesser said.
Dramatic decrease in the image analysis error rate
The fact that innovative companies today can make extensive use of AI, not least in the healthcare sector, is thanks to the advances in machine learning made possible by greater processing power, better graphics, and more extensive training data sets, stressed Professor Dr.-Ing. Joachim Hornegger, President of the University of Erlangen-Nuremberg and previously Chair of Pattern Recognition there: “Thanks to deep learning the error rates in image analysis have fallen dramatically since around 2012. Since 2015, AI methods surpassed humans in some areas.”
In image-based medicine, deep learning is facilitating medical assistance systems that until recently would have been unimaginable. Hornegger gave as an example digital subtraction angiography (DSA) software, a widely used method of imaging blood vessels using a contrast medium. However, until now DSA procedures could not be easily used on the coronary arteries because the blood vessels constantly move as the heart pumps. “Deep learning allows us to train the software to recognize the vascular tree so that we can use a DSA technique on the heart that requires only one image.” Hornegger sees other promising applications for deep learning in the detection of anatomical landmarks in CT data sets, as well as in improving image quality in 3D reconstructions.
“AI can help us improve precision and individual care. It helps enhance quality and eliminate errors, especially at a time when medical staff are facing an ever-increasing workload,” said Thomas Friese, PhD, Senior Vice President Data Architecture and Technology Platforms at Siemens Healthineers. “We are entering an era where there is more room for autonomy. It is no longer simply about detection but increasingly also about evaluation.”
An example of a simple but very useful AI application that already has product status at Siemens Healthineers is the automatic positioning of patients undergoing CT scans with the help of a camera mounted above the CT table. “This allows us to reduce the error rate and improve the dose rate,” said Friese. In other words, less radiation is required, or the image quality is improved while the dose stays the same.
At the annual meeting of the Radiological Society of North America (RSNA) in November, Siemens Healthineers unveiled the AI-Rad Companion, an AI platform that uses extensive automation to support radiologists with diagnostics – for example with the AI-Rad Companion Chest CT1 , a thoracic CT assistant. Using CT images of the chest, the software can differentiate between various structures in that region of the body, highlight them individually, and mark and measure potential abnormalities. AI-Rad Companion Chest CT is designed to help radiologists interpret images via automation for potentially reduced time spent on results documentation. The plan is to rapidly develop the AI-Rad Companion as an AI platform for radiology, said Christiane Bernhardt, Global Head of Sales and Marketing for CT at Siemens Healthineers: “We cannot afford to wait long. The next assistants are already in the pipeline.”
About the Author
Philipp Grätzel von Grätz is a medical doctor turned freelance writer and book author based in Berlin, Germany.