Artificial intelligence in neuro-radiology: overview of the current commercial landscape

The last decade has seen significant strides in the application of AI within the field of radiology. With the volume of imaging increasing at a rate far greater than recruitment of consultant radiologists, AI is often considered the panacea to a labour force in dire crisis. 1 Although there remain differences in one’s definition of AI, depending on computational experience, AI within radiology generally refers to the automation of fundamental cognitive tasks that would normally require human intelligence. While the principles of AI have been around since at least the 1950s, a lack of computational hardware and structured data meant that early iterations of AI could not surpass human performance. 2 Fast forward to the widespread adoption of deep learning, accessibility to graphical processing units and large repositories of annotated data, and it is now commonplace to see AI surpassing human performance in narrow radiological tasks, mainly involving image classification and object detection. Deep learning is the method by which an algorithm can ‘learn’a set of features that reflect a hierarchy of structures within data. 3 It uses an artificial neural network that is loosely based on biological neurons of the human brain and has allowed algorithms to be successful in performing a number of perceptual tasks, which were previously not possible.