Recent Advances in Medical Imaging Segmentation: A Survey

TL;DR

This survey analyzes recent advances in medical image segmentation (MIS) through four interconnected strands: Generative AI, Few-Shot Learning, Foundation Models, and Universal Models.It synthesizes theoretical foundations, state-of-the-art techniques, and diverse applications, while identifying persistent challenges such as data privacy, domain shift, annotation burden, and the lack of standardized benchmarks.The authors discuss GANs, diffusion models, CycleGAN, and SAM-based approaches, along with few-shot and universal-model architectures, highlighting their complementary roles in improving MIS generalization and efficiency.They also outline practical barriers to clinical adoption and propose directions for future research, including integration across paradigms and the development of privacy-preserving, standardized evaluation frameworks.A GitHub repository is maintained to track ongoing innovations, reflecting the dynamic evolution of MIS methods.

Abstract

Medical imaging is a cornerstone of modern healthcare, driving advancements in diagnosis, treatment planning, and patient care. Among its various tasks, segmentation remains one of the most challenging problem due to factors such as data accessibility, annotation complexity, structural variability, variation in medical imaging modalities, and privacy constraints. Despite recent progress, achieving robust generalization and domain adaptation remains a significant hurdle, particularly given the resource-intensive nature of some proposed models and their reliance on domain expertise. This survey explores cutting-edge advancements in medical image segmentation, focusing on methodologies such as Generative AI, Few-Shot Learning, Foundation Models, and Universal Models. These approaches offer promising solutions to longstanding challenges. We provide a comprehensive overview of the theoretical foundations, state-of-the-art techniques, and recent applications of these methods. Finally, we discuss inherent limitations, unresolved issues, and future research directions aimed at enhancing the practicality and accessibility of segmentation models in medical imaging. We are maintaining a \href{https://github.com/faresbougourzi/Awesome-DL-for-Medical-Imaging-Segmentation}{GitHub Repository} to continue tracking and updating innovations in this field.

Figures (10)

• Figure 1: Examples of segmentation modalities and tasks in medical imaging. The figure shows the diversity of medical imaging modalities, each with distinct appearances and properties including varying shapes, weak boundaries, and low contrast.
• Figure 2: General overview of MIS challenges. This survey focuses on the latest developments in generative AI models, few-shot segmentation techniques, foundation models, and universal models. These approaches have demonstrated their effectiveness in addressing key challenges in medical image segmentation, including both data-related and method-specific challenges.
• Figure 3: Illustration of generative adversarial models, including: (a) Vanilla GAN goodfellow2014generative, (b) cGAN mirza2014conditionalgenerativeadversarialnets, (c) ACGAN odena2017conditional and (d) CycleGAN Karras2018ASG.
• Figure 4: General overview of the diffusion model, which consists of two main phases: the diffusion phase, where noise is gradually added, and the denoising phase, where noise is progressively removed using a U-Net-like architecture.
• Figure 5: Examples of generative models in MIS. a. DAN zhang2017deep. b. 3D-Cyc-Seg zhang2018translating, $i_{path}$ and $j_{path}$ represent the cyclic translation and segmentation process from domain A to B and from domain B to A, respectively. c. ADV-SSVS ma2021self, X and Y represent the segmentation map and coronary angiogram image, receptively. d. DARL kim2024c, a hybrid GAN-Diffusion Model. e. MedSegDiff wu2024medsegdiff. Better view in Colors.