Clinical correlation recommended.

Artificial intelligence has now made the jump from learning to interpret radiology images to learning to interpret hedged radiology reports. In this study, deep learning models were used to extract cancer outcome data from archived rad reports. These models were trained by human curators who abstracted the reports into a structured framework. When tested, the model was accurate at picking up the presence of disease, improvement or response, and worsening or progression from reports. In addition, it was able to generate accurate disease-free survival and progress-free survival times. TBL: The next step to a smart electronic medical record is autonomous extraction of accurate clinical outcomes data. | Kehl, JAMA Oncol 2019