Explainable Artificial Intelligence (XAI) in Healthcare

Abstract

Artificial Intelligence (AI) describes computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and language translation. Examples of AI techniques are machine learning, neural networks and deep learning. AI can be applied in many different areas, such as econometrics, biometry, e-commerce and the automotive industry. In recent years, AI has found its way into healthcare as well, helping doctors to make better decisions (‘clinical decision support’), localizing tumors in magnetic resonance images, reading and analyzing reports written by radiologists and pathologists, and much more. However, AI has one big risk: it can be perceived as a ‘black box’, limiting trust in its reliability, which is a very big issue in an area in which a decision can mean life or death. As a result, the term Explainable Artificial Intelligence (XAI) has been gaining momentum. XAI tries to ensure that AI algorithms (and the resulting decisions) can be understood by humans. In this narrative review, we will have a look at the current status of XAI in healthcare, describe several issues around XAI, and discuss whether it can really help healthcare to advance, for example by increasing understanding and trust. Finally, alternatives to increase trust in AI are discussed, as well as future research possibilities in the area of XAI.

A Comprehensive Survey of ArtificialIntelligence in Precision Healthcare:Shedding Light on Interpretability

“…Interactive methods that represent brain images with prominent activations in multiple regions of the brain and using clustering methods to represent multiclass classifications will be more effective [54,55,56].…”

Explainable AI: Bridging the Gap between Machine Learning Models and Human Understanding

“…This study explored the problem of the link between mechanistic (or hybrid/natural languagebased) biological models and machine learning substitutes, showing the necessity of using domain knowledge in AI models to improve their analysis and applicability [18]. The paper "Explainability in AI healthcare" by Hulsen (2023) targeted the concepts and challenges of explainable AI in healthcare, stressing its crucial importance for the presented decision-making processes to be explainable, transparent, accountable, and so well trusted [19]. Jean-Quartier et al (2023) studied the computational cost of XAI algorithms and proposed sustainable machine learning methods that combine the interpretability with the computational efficiency, thus, achieving the balance [20].…”

Leveraging complexity frameworks to refine theories of engagement: Advancing self‐regulated learning in the age of artificial intelligence

“…These event data are typically organized into features of student learning behaviour including counts and patterns of usage of single items, or classes of items, merged with outcome variables and historical achievement data such as GPA or pretest scores, and used to train and test prediction models (Arizmendi et al, 2022;Baker et al, 2015;Brooks & Thompson, 2017). These models can be useful for identifying students likely to be at risk of a poor educational outcome, but are often not aligned with educational principles or theory in a meaningful or explainable way (Turek, 2018). Knight and Shum (2017) described one of the foundational concepts of learning analytics as going from clicks to constructs.…”