Deep Interactive Segmentation of Medical Images: A Systematic Review and Taxonomy

Interactive segmentation is a crucial research area in medical image analysisaiming to boost the efficiency of costly annotations by incorporating humanfeedback. This feedback takes the form of clicks, scribbles, or masks andallows for iterative refinement of the model output so as to efficiently guidethe system towards the desired behavior. In recent years, deep learning-basedapproaches have propelled results to a new level causing a rapid growth in thefield with 121 methods proposed in the medical imaging domain alone. In thisreview, we provide a structured overview of this emerging field featuring acomprehensive taxonomy, a systematic review of existing methods, and anin-depth analysis of current practices. Based on these contributions, wediscuss the challenges and opportunities in the field. For instance, we findthat there is a severe lack of comparison across methods which needs to betackled by standardized baselines and benchmarks.