A preliminary glossary of artificial intelligence in radiology

The use of artificial intelligence (AI) and deep learning is progressively gaining credibility in medicine, particularly within the radiological sciences (1). In particular, convolutional neural networks (CNN) can solve even complex cases and provide diagnoses within short time frames with a level of accuracy that occasionally surpasses the capabilities of radiologists. In this vein, data scientists and data engineers have appeared before radiology society conferences and delivered interesting talks regarding AI. As a result of the use of AI in radiology, previously unheard-of terminology has been introduced to the field that will become common language in the near future. Thus, this paper seeks to define many of these common terms for the benefit of radiology practitioners. In their Canadian Association of Radiologists white paper on AI, Tang et al. very briefly mentioned an AI glossary (2). However, to the best of this author’s knowledge, this paper presents the first brief, practical glossary for AI in radiology in the English literature that can serve as a reference prototype for the purpose of simplifying the complex terminology. It is especially designed for radiology trainees and experienced radiologists who have an interest in AI. It is important to note that this is not intended as a precise or exhaustive glossary; therefore, some data science-specific, nonradiological terminology has been excluded from this paper. Instead, the most frequently encountered terms that relate to radiology are included below. Artificial intelligence (AI): Highly developed computer systems that have the ability to perform tasks that normally require human intelligence, such as visual perception, translation, speech recognition, image interpretation, interaction with humans (e.g. chatting), and decision-making. AI, sometimes called machine intelligence, is a type of intelligence demonstrated by machines (3,4). Algorithm: Step-by-step instructions completed by computers, including simple or complex tasks, such as setting reminders or identifying a group of people within a crowd (4). Backpropagation: The manner in which CNNs learn. They are able to recognize the differences between output and desired output and adjust calculations in reverse order of execution (5). Big data: A term for extremely large datasets that can be analyzed to reveal patterns, trends, and associations. Big data includes electronic medical records, which contain huge digital imaging archives, pathology department and laboratory archives, and millions of digital clinical notes and diagnoses (4). Blackbox: CNN’s (algorithm’s) unknown internal working pattern (6). Essentially unknown processing, especially within hidden layers. For example, it is difficult to comprehend how a CNN reaches an outcome. Cluster analysis (clustering): The task of grouping a set of objects similar to each other into clusters (7). Convolutional neural network (CNN): Deep learning intelligence network commonly used in diagnostic imaging (7). Data mining: A process used to extract usable data from a larger set of raw data. Data mining algorithms are particularly beneficial on complex datasets with a large number of variables and samples (8). Data science: A field that uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from data in various forms, both structured and unstructured. Similar to data mining (7). Data scientist: A person who analyzes and interprets data (9). Deep learning (machine learning): “A set of automatic pattern recognition methods that have been successfully applied across various problem domains, including biomedical image analysis” (10). Diagnosis (artificial intelligence): Concerned with the development of algorithms and techniques that are able to show whether the behavior of an AI system is correct