Abstract
This chapter introduces basic human-in-the-loop (HITL) techniques, focusing on the integration of human expertise into the training and testing of artificial intelligence (AI) models for the analysis of magnetic resonance (MR) imaging. We first examine the role of AI in automating MR data analysis, from segmenting anatomical structures to predicting disease outcomes. Despite recent successes, AI is vulnerable to biases and errors present in datasets, which can be mitigated by the collaboration with clinicians, who can provide oversight during training and testing to ensure diagnostic accuracy, improved generalization, transparency, reliability, and clinical relevance. The chapter covers essential HITL training techniques like active learning and reinforcement learning, as well as HITL testing methods like learning to defer and complement. It concludes with a discussion on practical challenges, ethical concerns, and future directions in advancing human-AI collaborative MR imaging analysis for clinical applications.