How long will it take for deep-learning algorithms to be better than radiologists at their job? - Quora

The matrix of anatomy, pathology and image observations on multiple modalities is huge. The supervised learning algorithms are hampered by the lack of annotated data sets for each cell in that matrix. The unsupervised learning methods will be hampered by the few examples of many of the rare cells of the matrix. Algorithms are typically developed one by one, i.e., this image observation in this disease setting (possibly in this population). Even after an algorithm is developed, it has to be'perfected' to high sensitivity and specificity because humans are very good at this task.