A Review of Deep Learning-Based Lung CT Lesion Segmentation Methods

Deep learning now dominates lung lesion segmentation in CT, surpassing traditional and ML methods via end-to-end features and multi-scale context. We survey: (1) its advantages; (2) architectures—FCN, U-Net family, GANs, Transformers—for nodules, GGO, consolidation; (3) benchmarks on LIDC-IDRI, COVID-CT, Lung-PET-CT-Dx (Dice, IoU, sensitivity, specificity, speed); (4) open challenges—few-shot, compression, cross-domain generalization—and future directions.