This innovation area focuses on developing more efficient deep learning methods for diagnosis support and decision support for diseases such as cardiovascular diseases and cancer.
Medical images captured from inside the body using various scanning and imaging techniques have traditionally been challenging and time-consuming to analyze by trained experts. Well-performing deep learning models in the medical domain have the potential to assist healthcare professionals by increasing certainty and streamlining analyses of medical images.
Visual Intelligence researchers have developed several innovations which aim to assist healthcare professionals in the clinical workflow. For instance, our research efforts have resulted in novel deep learning methods, such as for:
Major obstacles of developing deep learning methods in medicine and health include the availability of training data, the estimation of confidence and uncertainty in the models’ predictions, as well as lack of explainability and reliability. The innovations mentioned above address these research challenges in different ways, enabling progress within this innovation area.
For instance, research on the challenge of learning from limited data is at the core of the clinically inspired data augmentation technique for CT images mentioned above. This method also leverages context and dependencies by exploiting knowledge about the signal-generating process.
Research on explainable and reliable AI constitutes a significant part of our method for detecting cancer in mammography images. This is also the case for our novel content-based CT image retrieval method.
When developing deep learning solutions for concrete medical and health challenges that our user partners face, it is important to transfer knowledge and methodologies across innovation areas. Our proposed methodologies within medicine and health synergize well with other work within this innovation area, as well as our other three innovation areas.
For instance, the development of a semi-automatic landmark prediction in cardiac ultrasound depends on context provided in the form of a scan line in the echocardiography. This is inspired by other developed solutions which leverage context in the form of anatomical knowledge, e.g. for cancer detection in mammography.
Self-supervised deep learning, which several of our medical innovations are based on, has not only proven useful within medicine and health, but also in “Marine science” “Energy” and “Earth observation”. For example, the framework for CT image retrieval shares similarities with a content-based image retrieval system for seismic data.