ISBI Daily - Friday

which includes the borders of the disease that are usually the most difficult to define. Although deep learning is state of the art in the medical field – some reports say it can perform better than experts – there are still problems that have to be addressed. Without enough data, a deep learning model cannot be built. This work combines a convolutional network, which uses patches from regions of high confidence from a sparsely annotated dataset, and uses it in the full convolutional network, which uses image annotation on a pixel level. Work is continuing and in terms of next steps, the team are trying to create other models and use more advanced deep learning architectures in order to achieve better and more accurate results, and to improve comparisons with other doctors that annotate the datasets. Guillaume points out how the main difficulty is nowadays the lack of annotated datasets: the chance to combine own annotations with others which were done in different conditions and are made available online is a major advantage. Maria tells us that working with a doctor helped her understand the disease better and that researchers should cooperate with medical doctors to advance research in medical imaging. Guillaume adds: “ It is also very good for medical doctors to work with engineers, because it is the future. ” ISBI DAILY Friday 15 Guillaume and Maria invite you to their poster session today at 11:30- 12:30 in the Ambassador Ballroom to find out more about this work. Guillaume Chassagnon This work combines a convolutional network, which uses patches from regions of high confidence from a sparsely annotated dataset, and uses it in the full convolutional network …