Application of Artificial Intelligence in Healthcare: The Need for More Interpretable Artificial Intelligence

Understanding artificial intelligence (AI) and its different types is of the utmost importance for the application of this technology in healthcare. Artificial intelligence is a field of knowledge which combines computer science and advanced statistics to support problem-solving. It is divided in two sub-fields: machine learning (ML) and deep learning. The ML concept resides in the ability of using computer algorithms that have the capability to recognize patterns and efficiently learn to train the model to predict, make recommendations or find data patterns. After a sufficient number of repetitions and algorithm adjustments, the machine becomes capable to accurately predict an output. Deep learning is a newer and more complex approach of AI that uses deep neural networks. The neural network starts with an input layer that then progresses to a variable number of hidden layers. Since the algorithm uses multiple layers with deep neural networks, it can successively refine itself, without explicitly programmed directions. It is a fact that, by using deep learning, the models usually achieve higher accuracy compared with ML. Still, when using ML, it is frequently possible to better understand which are the input variables that have more influence on the output variables. In both medical and clinical practices, it is often particularly relevant to understand why an AI technique is suggesting a certain classification or direction for a certain action. Not only in healthcare but also in other fields of knowledge, explain-able AI (also called XAI) is growing its influence. The current European legal regulation, specifically the General Data Protection Regulation (GDPR), requires that automated models provide meaningful information about the rationale on how the algorithm operates. The goal of this article is not to provide an exhaustive view about all existing AI models and explainable AI, but instead to provide a summarized and easy to understand view of what should be considered when implementing AI in healthcare and in clinical practice.