Toward the transparency of deep learning in radiological imaging: beyond quantitative to qualitative artificial intelligence

In the near future, nearly every type of clinician, from paramedics to certificated medical specialists, will be expected to utilize artificial intelligence (AI) technology, and deep learning (DL) in particular (1). In terms of exceeding human ability, DL has been the backbone of computer science. DL mostly involves automated feature extraction using deep neural networks (DNNs), which can aid in the classification and discrimination of medical images, including mammograms, skin lesions, pathological slides, radiological images, and retinal fundus photographs.