AI in RadiologyEmpowering clinical decisions with AI

Neural networks are inspired by the brain. As within the brain, they consist of multiple layers. The input layer receives the data (such as a CT volume), while hidden layers extract characteristic data features, which are relevant for the specific diagnostic task and vary between various medical fields of application. In a next step, the detection and expression of the features is being learned by the network to assess them and assign respective interpretation. To train this neural network, a large amount of data is needed to learn, for example, how to differentiate a lung nodule from a healthy lung. For the accuracy of the algorithm, it is important to have a good quality of the data to feed the algorithm, so for supervised learning for example, annotated data is needed.

To ensure the high quality of our algorithms, we collaborate with well-known institutions around the world to understand customer needs to train our algorithms accordingly. 

Siemens Healthineers CEO Dr. Bernd Montag explains it through a “step-by-step” approach. The first step is to begin with building more and more digitalization into our devices and incorporating AI. As the systems become more intelligent and adapt to patients, they deliver the right quality automatically. The next step is to support diagnostic findings, for example, data from a CT or MR. This is then followed by solving an even bigger challenge, which is to bring all the data from various sources together and build a digital assistant for holistic decision-making in healthcare. It is not a sudden transformation, rather a work-in constant progress.

AI is already used in the workflow, image acquisition and reconstruction space. With the help of AI, we are able to get more accurate data, important for later diagnosis. A great example for this is myExam Companion with features like the 3D camera. With an AI-based algorithm, it analyzes the patient shape and identifies key anatomic landmarks for patient pose, body region, and iso-center detection. Such AI-powered systems not only have the potential to reduce errors due to operator dependency, standardize the acquisitions in a patient specific manner, and cut down the time but can also contribute to radiation dose reduction due to automatic table-height and scan range determination. With the rapid advances of AI also in the field of diagnosis, it is very likely that AI will be used as an additional diagnosis help for examinations and reporting. This is already done today with the intelligent solution AI-Rad Companion³.

• Order/Schedule: AI-powered connection of patients and physicians
• Preparation and Acquisition: AI-powered standardized, accurate patient positioning and planning of exams
• Postprocessing/Quantification: AI-powered automatic lesion scoring, automatic measurements
• Interpretation/Report Generation: AI-powered automatic highlighting, characterization and quantification of anatomies and abnormalities

However, there are other aspects like country regulations and reimbursements that are relevant for getting these innovations into the clinical routine. Siemens Healthineers works closely with clinicians, regulators, and organizations all over the world to standardize our products to the healthcare needs, to provide the best possible outcome in the radiology workflow. At the heart of developing AI products for us lies the focus to help clinicians and radiologists with their daily work to provide better patient outcomes. While we strive to digitalize healthcare, one aspect of this transformation is Artificial Intelligence, co-developed with our customers. Because only when the developed algorithms fit the needs of our customers and follow all rules of the regulators, can we get to the adoption rate where new technology can make a difference.