Open research questions in Medical Image Segmentation Techniques

What the literature leaves open

• Domain shift and entropy discrepancy between different imaging modalities. Limited availability of labeled data in the target domain. Need for robust and efficient methods for domain adaptation in medical imaging.

  TCSA-UDA: Text-driven cross-semantic alignment for unsupervised domain adaptation in medical image segmentation · 2026 · DOI

• To evaluate the proposed method on other medical imaging datasets. To explore the application of the proposed method to other domains, such as natural language processing and computer vision. To develop more robust and efficient methods for domain adaptation in medical imaging.

  TCSA-UDA: Text-driven cross-semantic alignment for unsupervised domain adaptation in medical image segmentation · 2026 · DOI

• Future work will build upon the foundation of Med-TDA, exploring new TDA approaches for medical imaging.

  Med-TDA: Medical Imaging Topological Data Analysis Tool · 2026 · DOI

• There are no standardized definitions or pipelines to benchmark model performance in TDA for medical imaging. Current TDA tools are mainly developed for general machine learning applications, not specifically for medical imaging.

  Med-TDA: Medical Imaging Topological Data Analysis Tool · 2026 · DOI

• Lastly, while Monte Carlo sampling is used to measure uncertainty, it is also important to note that utilizing Monte Carlo sampling will add approx- imately a unique amount of computational burden to methodology; therefore, future investigations should explore adaptive routing regularization, domain adaptation methods and lightweight approximations to distributional uncertainty in order to further enhance the generalizability and efficiency for method deployments. One limitation is that evaluation was conducted using carefully curated challenge datasets instead of evaluating using com- pletely heterogeneous multi-institutional clinical cohorts which typically exhibit much larger variability in scanners.

  Dynamic expert routing with boundary-aware decoding for accurate brain tumor segmentation from incomplete MRI data · 2026 · DOI

• Nobre Menezes et al. (17) 416 XCA images EfficientUNet++ • Tested only on diseased segments • Annotation quality dependency Tao et al. (18) 134 + 150 images Bottleneck Residual U-Net (BRU-Net) • Background segmentation errors • Challenges with tiny vessels • Dependence on image quality Hao et al. (19) 332 XCA images SVS-Net (Sequential Vessel Segmentation Network) • Challenges in segmenting small and thin vessels • Lack of temporal–spatial consistency in vessel masks • Limited handling of background artifacts Cervantes Sanchez et al. 130 XCA images Multi-layer perceptron with multi-scale filtering • Limited vessel detection due to single-scale filtering (20) (Gaussian + Gabor) • Issues with uneven lighting and low contrast Jiang et al.

  Angio-fusion net: dual-stream enhanced VGG16 attention U-Net for vessel morphology preservation in XCA segmentation · 2026 · DOI

• Coronary vessel segmentation in XCA is challenging due to poor image quality, including low contrast, noise, and artifacts, as well as complex visualization of vessel structures with curves and bifurcations. Traditional preprocessing methods, such as G-channel thresholding, have feature significant masking. Background interference from pulmonary tissues, bones, catheters, and cardiac motion artifacts further complicates accurate vessel delineation, particularly for low- (4–7). The contrast vessels with unclear boundaries advancement of deep learning has significantly enhanced medical vessel segmentation. Modern Convolutional Neural Network (CNN) architectures, such as U-Net (8) and DeepLabV3+ (9), have advanced feature information. Feature Pyramid Networks (FPNs) (10) have improved detail lesion retention, particularly segmentation. Hybrid models, such as SegFormer and Swin- U-Net, have feature representation and contextual understanding (11). Attention mechanisms have also contributed by optimizing feature selection and reducing background noise. However, U-Net in handling complex and multi-class segmentation, emphasizing the necessity for continued research in this domain. segmentation by using multi-scale faces difficulties further refined particularly coronary imaging, skin in in Several models have addressed challenges in coronary vessel segmentation. Notably, Fuzzy Attention (FA)-SegNet (12), TABLE 1 Comparison of segmentation methods and constraints for XCA images.

  Angio-fusion net: dual-stream enhanced VGG16 attention U-Net for vessel morphology preservation in XCA segmentation · 2026 · DOI

• Further research can be conducted to improve the efficiency and accuracy of the proposed model. The model can be applied to other image registration tasks.

  A multiscale variational framework for joint diffeomorphic image registration and bilateral intensity correction · 2026 · DOI

• Existing studies do not consider the scenario where the target image also suffers from intensity inhomogeneity. The proposed model addresses this gap.

  A multiscale variational framework for joint diffeomorphic image registration and bilateral intensity correction · 2026 · DOI

• The existing algorithms suffer from under and over-segmentation. The existing algorithms have high computational complexity in computing image and region entropy.

  Unsupervised Optimization of Boundary Information Based on the Coefficient of Variation to Improve Image Segmentation · 2026 · DOI

• There is a lack of studies on automatic intelligence-assisted segmentation and quantification of mouse cardiac slice images. Existing cardiac segmentation models mainly focus on human cardiac studies. The measurement of mouse cardiac slice images is challenging due to the complex and irregularly U-shaped structure of infarcted areas.

  Dynamic U-shaped convolutional network for mouse cardiac image segmentation and quantification · 2026 · DOI

• Existing methods often struggle to accurately capture heart motion - Conventional image registration methods rely on intensity-based image registration similarity losses - Deep learning-based image registration methods have limitations, such as relying on intensity-based image registration similarity losses

  CardioMorphNet: Cardiac motion prediction using a shape-guided Bayesian recurrent deep network · 2026 · DOI

• Single-slice CT interpretation remains challenging due to low contrast, blurred boundaries, and interference from skull-related structures. Existing methods may not effectively address these challenges and may not provide robust and accurate classification performance.

  Robust slice-level stroke classification in non-contrast head CT via structural consistency regularization and counterfactual suppression · 2026 · DOI

• Developing segmentation models that remain reliable across diverse medical imaging domains is a persistent challenge. Variations in imaging modalities, scanners, and acquisition settings introduce significant domain shifts. Existing approaches have limited generalization and reduced robustness in real-world clinical scenarios.

  HyperSeg-DG: Multi-Scale Hyper Feature Context for Domain Generalized Medical image Segmentation · 2026 · DOI

• Ongoing academic research is needed to refine and extend Discrete Wavelet Analysis methodologies for image segmentation. Addressing the challenges and limitations of the technique, including computational complexity and noise sensitivity, is a future research direction.

  Retraction Notice: Discrete Wavelet Analysis: A Mighty Approach for Image Segmentation · 2026 · DOI

• The gap in image segmentation techniques that can effectively handle multiscale analysis and localize features within images. The need for refining and extending Discrete Wavelet Analysis methodologies for image segmentation.

  Retraction Notice: Discrete Wavelet Analysis: A Mighty Approach for Image Segmentation · 2026 · DOI

• The threshold method is mainly for gray information. The region growing method needs human participation to select the appropriate seed points for each region. The algorithm is sensitive to noise.

  Retraction Notice: Applications of Image Segmentation Techniques in Medical Images · 2026 · DOI

• Existing learning-based methods encounter difficulties in achieving high registration accuracy under large anatomical deformations. Traditional registration methods optimize deformations iteratively for each image pair, which is computationally intensive and highly sensitive to hyperparameter tuning.

  CRR-Net: a correlation reconstruction and refinement network for deformable medical image registration · 2026 · DOI

• Investigating the performance of PlantFormer in scenarios with high-density, early-stage disease outbreaks. Exploring the application of PlantFormer in other datasets or environments.

  PlantFormer: a precise plant disease segmentation network with interactive backbone and global-anisotropic context aggregation · 2026 · DOI

• The lack of effective models for precise plant disease segmentation in complex field environments. The need for a network that can address the challenges of anisotropic spread of lesions and blurred biological boundaries.

  PlantFormer: a precise plant disease segmentation network with interactive backbone and global-anisotropic context aggregation · 2026 · DOI

• Future research should investigate the application of deep learning methods to other medical image registration tasks. Future research should explore the use of deformation regularization in other image registration applications.

  Beyond the LUMIR challenge: The pathway to foundational registration models · 2026 · DOI

• Previous challenges have relied upon anatomical label maps. There is a need for a next-generation benchmark for unsupervised brain MRI registration.

  Beyond the LUMIR challenge: The pathway to foundational registration models · 2026 · DOI

• Developing methods that can handle tissue tears in spatial transcriptomics registration. Addressing batch effects and cross-donor integration in ST analysis. Improving the accuracy and robustness of supervised methods for ST registration.

  Tissue tearing degrades optimal-transport and diffeomorphic registration of spatial transcriptomics beyond displacement magnitude: a multi-seed deformation benchmark and a supervised graph cross-attention proof-of-concept. · 2026 · DOI

• The behavior of spatial transcriptomics registration methods under tissue tears has not been systematically characterized. There is a need for methods that can handle tissue tears in spatial transcriptomics registration.

  Tissue tearing degrades optimal-transport and diffeomorphic registration of spatial transcriptomics beyond displacement magnitude: a multi-seed deformation benchmark and a supervised graph cross-attention proof-of-concept. · 2026 · DOI

• Further evaluation of the proposed Pairwise Surface DSC method in different clinical contexts. Investigation of other surface-based metrics for clinical acceptability. Development of more reliable liver segmentation models.

  A Real-World Evaluation of Failure Detection for Liver CT Segmentation · 2026 · DOI

Most-cited papers in Medical Image Segmentation Techniques

• EMCAD: Efficient Multi-Scale Convolutional Attention Decoding for Medical Image Segmentation · 2024 · 420 citations
• A Review of Medical Image Segmentation Algorithms · EAI Endorsed Transactions on Pervasive Health and Technology · 2021 · 233 citations
• Image Data Augmentation Approaches: A Comprehensive Survey and Future Directions · IEEE Access · 2024 · 217 citations
• Layered deep learning for automatic mandibular segmentation in cone-beam computed tomography · Journal of Dentistry · 2021 · 147 citations
• SegFormer3D: an Efficient Transformer for 3D Medical Image Segmentation · 2024 · 123 citations
• Recursive Deformable Pyramid Network for Unsupervised Medical Image Registration · IEEE Transactions on Medical Imaging · 2024 · 114 citations
• Refined tooth and pulp segmentation using U-Net in CBCT image · Dentomaxillofacial Radiology · 2021 · 110 citations
• MONAI Label: A framework for AI-assisted interactive labeling of 3D medical images · Medical Image Analysis · 2024 · 91 citations
• Modality-Agnostic Domain Generalizable Medical Image Segmentation by Multi-Frequency in Multi-Scale Attention · 2024 · 90 citations
• Adaptive Bidirectional Displacement for Semi-Supervised Medical Image Segmentation · 2024 · 90 citations

Most recent work

• LDM-Morph: Latent diffusion model guided deformable image registration · Pattern Recognition · 2026
• Dynamic expert routing with boundary-aware decoding for accurate brain tumor segmentation from incomplete MRI data · Scientific Reports · 2026
• BMSR: A multimodal image registration method for addressing modal differences and spatial misalignment · Infrared Physics & Technology · 2026
• Conditional Diffusion Models for CT Image Synthesis from CBCT: A Systematic Review · Tomography · 2026
• Generalizable CT Vision-Language Modeling for Population Health and Disease Risk · medRxiv · 2026
• From Classical Pipelines to Promptable Foundation Models: A Cross-Domain Survey of Thin-Object Segmentation for Power Lines, Cracks, and Retinal Vessels · IEEE Internet of Things Journal · 2026
• NeuroSeg-MF: robust neuron segmentation in two-photon Ca ²⁺ imaging using multi-feature fusion and detection-guided SAM · Biomedical Optics Express · 2026
• A Gaussian mixture model for combining single threshold and adaptive threshold segmentation of bone microstructure · Biomedical Physics & Engineering Express · 2026
• Decipher-MR: a vision-language foundation model for 3D MRI representations · npj Digital Medicine · 2026
• Phase-constrained zero-shot self-supervised learning for BLADE liver MRI reconstruction · Magnetic Resonance Materials in Physics, Biology and Medicine · 2026

Related topics in Computer Science