Medical AI is currently trained for isolated use-cases: predict pneumonia in a chest X-ray, find cancer in a histopathology slide or distinguish benign from malignant skin lesions.
However, that’s not how medicine works! Information in clinical routine is multimodal: images, numbers, text, voice, video. All these channels of information can be harnessed for diagnostic purposes and clinical decision-making. To this end, a wealth of data has to be integrated from different sources and locations in order to train a robust model.
The branch of AI that tackles such multimodal challenges are so-called foundational models. The demand for such models is no less than providing full scale decision support in the clinical routine.
Together with colleagues from surgery, radiology, radiotherapy and pathology, we aim to develop such a unified AI tool for the clinical routine. Stay tuned for more!
