AI-driven segmentation techniques for COVID-19 pneumonia in medical imaging: a short review and future directions

The use of artificial intelligence (AI) for medical imaging has been greatly advanced and applied in response to the COVID-19 pandemic, particularly with regard to automated analysis of lung lesion patterns related to viral pneumonia. Among various AI applications, image segmentation has come to play a fundamental role, allowing accurate delineation of the infected regions such as ground-glass opacities (GGOs) and consolidations in chest CT and x-ray images. These segmented images are essential not only for the diagnosis, but also for measuring disease severity, guiding the treatment planning, and monitoring the patient recovery. In this chapter, we provide a targeted review on the AI-based lung segmentation for COVID-19, and discuss model designs, labeled datasets, clinical practices, as well as performance assessment approaches. We review how deep learning models—including those based on U-Net variants and attention mechanisms—have been adapted for this task of interest, typically designed by transfer learning, semi-supervised learning, or hybrid combinations with clinical prior knowledge. We consider benchmark segmentation datasets and address practical issues including sparse annotations, imaging variation and inter-observer label disparity due to human error. Lastly, we discuss the clinical relevance of the segmentation outputs, from automated severity scoring to the integration into real-time decision-support systems. In the future, the direction of the field is heading towards more robust and generalized segmentation systems, such as by applying federated learning, weakly/semi-supervised techniques and multimodal data fusion. A focus on explainability, uncertainty quantification and real-world uplift are also likely to improve clinical trust and utility. Integrating recent progress and persistent challenges, this review aims to bridge the gap between technical development and clinical adoption, ultimately supporting scalable and resilient diagnostic workflows in both pandemic and post-pandemic healthcare systems.