M Costa, F. S. Barros, I. Chifu, A. Restivo
Abstract
Automated detection of solar coronal loops in extreme ultraviolet imagery has remained an open problem for two decades, despite its practical importance for coronal magnetic field modelling. Loops are observed as bright, arch-like structures in EUV imagery from instruments such as AIA/SDO, EUVI/STEREO, and FSI/Solar Orbiter, but their optically thin nature means that recorded intensity integrates emission along the line of sight rather than isolating individual structures. The result is a superposition of overlapping photon intensities against a diffuse background that defeats simple intensity-based approaches and challenges carefully designed geometric filters. Despite two decades of methodological development, no structured review of computational methods for coronal loop detection currently exists, leaving the field without a consolidated view of what has been attempted, where the structural obstacles lie, and which directions from adjacent disciplines are most likely to close the gap. This paper presents a structured semi-systematic review organised around three themes: the evolution of coronal loop detection from geometric ridge-tracing algorithms to deep learning segmentation models; the suitability of available observational data sources for machine learning tasks; and the transfer of curvilinear structure detection methods from medical imaging and remote sensing to the solar domain. While recent work has demonstrated that machine learning models can reconstruct three-dimensional loop coordinates from two-dimensional projections with high accuracy, those methods assume the 2D loop geometries are already extracted. Robust automated detection and centreline tracing from single-viewpoint EUV images remains understudied and lacks the large-scale annotated datasets that would make modern supervised approaches tractable. The absence of standardised evaluation protocols further limits meaningful comparison across methods. This review identifies the structural obstacles that have prevented a complete end-to-end pipeline and maps the directions most likely to close the remaining gaps.
Keywords
Solar coronal loops / Extreme ultraviolet imaging / Automated loop detection / Curvilinear structure detection / Deep learning / Image segmentation / Magnetic field extrapolation
Astronomy and Computing
Volume 57, Number 101169
2026 October
