For an average 20% of computed tomography and magnetic resonance images, radiologists see structure (pathology) with which they are not familiar. The process of finding information in order to write a report takes up to 20 minutes. There is a significant variation in the identification of pathologies in the same images by multiple radiologists leaving patients with different diagnosis for the same image.
for 20% of images
20 minutes to find information
for the same image
Led by contextflow GmbH, the pilot aims to reduce the time of diagnosis in radiology departments, and at the same time the quality of diagnosis by providing an efficient search engine for radiological data.
Finding relevant cases during search is a big data problem because of the high variability in natural anatomy and disease markers, the steady stream of new data and the need for fast access to the knowledge encoded in these cases during daily diagnosis. The technologies employed are deep learning, extremely large-scale image indexing and similarity search, semantic analysis of text in radiology reports and text search.