New artificial intelligence applications for emergency medicine

A new project is developing a software that automatically searches CT images for suspicious signs of critical symptoms, alerting the treating physician of developing complications.

The technique filters important information from X-ray and CT images that are decisive for further treatment

A new project is developing a software that automatically searches CT images for suspicious signs of critical symptoms, alerting the treating physician of developing complications.

The aim of the project, known as KI-RAD (in German: "Künstliche Intelligenz für radiologische Bildgebung in der Notfall- und Intensivmedizin" and English: "Artificial Intelligence for Radiological Imaging in Emergency and Intensive Care Medicine”), is to develop an intelligent X-ray assistant that helps to filter important information from X-ray and CT images that is crucial for the ongoing care of patients. 

Three critical application areas have been selected for this new technology: stroke, bone injuries, and chest x-ray. "Especially in emergency and intensive care medicine, having an intelligent X-ray assistant can be life-saving, as the physician can quickly recognize critical signs and make sure that nothing is overlooked," explains project coordinator Dr. Claus-Christian Glüer. Glüer is a Professor of Medical Physics at the Medical Faculty of the Christian-Albrechts University of Kiel (known also as CAU) and Head of the Molecular Imaging North Competence Center (MOIN CC).

An intelligent X-ray assistant to identify urgent problems

In practical terms, if a stroke is suspected, every minute counts. The most important thing is to quickly differentiate whether a blocked blood vessel or a cerebral hemorrhage is causing the symptoms. "The symptoms are similar in both cases, but the consequences are completely different," stresses Dr. Glüer. In the first case, the blood supply to the affected brain area must be restored as quickly as possible by administering special drugs. In the second case, the bleeding must be stopped and any damage caused by the leaking blood must be avoided.

In the case of bone injuries, the AI-supported analysis system should distinguish between fresh fractures resulting from an accident and old fractures, for example as a result of osteoporosis (bone loss). On the other hand, unstable fractures that require special care in dealing with those affected must be detected. In the case of vertebral body fractures, there is a risk that the spinal cord will be injured and paralysis may occur.

Dr. Jörg Barkhausen, professor at the University of Lübeck and Director of the Clinic for Radiology and Nuclear Medicine at the University Medical Center Schleswig Holstein, is responsible for the area of application of chest X-ray images. For such cases, the aim is to identify problems that require rapid treatment, such as in cases of respiratory distress due to pneumothorax or to check the correct placement of catheters.

Sample images for automated diagnostic procedures

The basis for the development of the intelligent X-ray assistant is CT images that have been interpreted and categorized by specialized radiologists. The AI system learns from these sample images and recognizes patterns and regularities. In addition, information from other imaging methods, such as magnetic resonance imaging (MRI), can also be used to train the system. "In MRI, for example, edema (fluid retention), which indicates a fresh fracture of vertebral bodies, can be detected more easily and clearly than, for example, on a CT image," says Dr. Glüer. Artificial intelligence might be able to compensate for this disadvantage.

There are many potential applications for an AI-supported analysis method. For example, smaller hospitals that lack the necessary radiological expertise could use it. It is also conceivable that it could be used for training purposes. Dr. Glüer explains: "First of all, we have to see how meaningful and specific the results are. This is what we want to find out in the expert research network on radiology and AI, and together with companies develop a pre-prototype. If the results are conclusive, the prototype can then be tested in clinical settings.