271 Non-invasive imaging biomarkers 3 learning, the ability of a system to extract information from raw data and to learn from experience. Decision trees, random forests, and support-vector machines are well-known examples of machine learning algorithms. More recently, deep learning, which is a subset of machine learning that uses a (convolutional) neural network structure loosely inspired by the human brain, emerged, providing even more sophisticated algorithms (Figure 1) [11]. Growing amounts of data and the availability of powerful computational hardware have empowered AI, allowing computers to better represent and interpret complex data [369]. The development and, to a lesser extent, use of AI in oncology are rapidly emerging, also in thyroid cancer. Applications vary from detection of abnormalities, lesions characterisation and the prediction of treatment response [370-373]. Whereas the first AI algorithms performed simple tasks with subhuman performance, more recent algorithms sometimes surpass humans in task-specific applications. As a result, tasks that, until a couple of years ago, could only be performed by humans, can now be executed by AI algorithms. In addition, AI algorithms have the potential to reduce variation, improve efficiency and prevent avoidable medical errors, when integrated in clinical practice as tools to assist clinicians [374]. Quantitative assessment by an algorithm reduces subjectivity that comes with visual assessment, because of the education and experience of a human reader, thereby preventing inter- and intraobserver variability [369]. In addition, a human reader can consider only a few variables at a time, quickly approaching the information processing capacity [375]. In contrast to qualitative assessment by a human reader, AI algorithms evaluate a large number of complex quantitative variables together, consistently, fast and efficiently. A major challenge of AI, however, is that the quality of a model highly depends on the input data, which is also referred to as ‘garbage Figure 1. Differences between artificial intelligence, machine learning, and deep learning
RkJQdWJsaXNoZXIy MTk4NDMw