Development and validation of a smartphone-based deep-learning-enabled system to detect middle-ear conditions in otoscopic images.