Topic: Key challenges for deep learning in medicine Consider twosystems for melanoma classification when viewed using dermoscopy(an instrument to closely examine a patient’s skin). one thesystems is relying on conventional machine learning pipeline whilethe other one is relying on the deep learning.
The first one’s inputs are features extracted by thedermatologists, shape-based features computed from the segmentedlesion, features describing color.
The second system’s input is the actual image. Here is what anexperienced dermatologist thinks about the systems: ” I like the1st system, because I need help with borderline cases and the firstone is more helpful with these cases, indeed, I can trust it… thesystem asks me to provide features and also show me the shape-basedand color-based features. I can play with the features – change thefeature set, I’m providing and see whether the system’s outputchange…”
Considering the above-mentioned comment, what is one of the mainchallenges of deep learning algorithms in medicine? Explain indepth
