Abstract:
Due to the limitations of human vision, it is simple for radiologists to miss
small malignant tumors. At the initial screening, up to 35% of lung nodules are
missed. Deep-learning systems understand what a tumor is from real-world
instances rather than searching for tumor features that a programmer has
predefined in advance. Researchers provide the systems with a huge data set
made up of lung CT scans of thousands of individuals, some of whom had
cancer and others of whom did. The ability of the computers to identify
between lung tumors and benign increases with the number of training scans
they have seen. And they perform this task more precisely than earlier, non-AI
systems.