A picture speaks a thousand words.

Artificial intelligence (AI) may not be ready for the bedside, but we never said anything about the MRI reading room. This Chinese tech-savvy study reports configuration of a convolutional neural network (CNN) trained with >28K pelvic MRIs depicting nodal metastases. Let’s stop right here for those who still struggle with Siri and explain what we mean by CNN. Besides a mainstream news outlet where you can reliably get your daily Trump tweet analyses, CNN is a type of machine learning especially designed for visual imagery. The neural part comes from the fact the CNNs are biologically inspired by our own visual cortex, with distinct “neural” components each with a very specific job. This enables almost instantaneous image recognition, much like a toddler distinguishing a dog from a cat. Now back to the CNN of interest: it was prospectively tested against “imaging experts” (i.e., radiologists with >20 years experience) for 414 pelvic MRIs of patients with rectal cancer. CNN achieved an equally high area under the ROC curve (> 0.9) as did expert imagers, representing a high level of discrimination in terms of both sensitivity and specificity. What’s more, CNN accurately read images at an average of 20 seconds versus roughly 600 seconds needed by radiologists. TBL: Consider hedging your risk of future unemployment by investing early in CNN. | Lu, Cancer Res 2018