Medical Image Segmentation Using Deep Learning: A Survey (2009.13120v3)

Abstract: Deep learning has been widely used for medical image segmentation and a large number of papers has been presented recording the success of deep learning in the field. In this paper, we present a comprehensive thematic survey on medical image segmentation using deep learning techniques. This paper makes two original contributions. Firstly, compared to traditional surveys that directly divide literatures of deep learning on medical image segmentation into many groups and introduce literatures in detail for each group, we classify currently popular literatures according to a multi-level structure from coarse to fine. Secondly, this paper focuses on supervised and weakly supervised learning approaches, without including unsupervised approaches since they have been introduced in many old surveys and they are not popular currently. For supervised learning approaches, we analyze literatures in three aspects: the selection of backbone networks, the design of network blocks, and the improvement of loss functions. For weakly supervised learning approaches, we investigate literature according to data augmentation, transfer learning, and interactive segmentation, separately. Compared to existing surveys, this survey classifies the literatures very differently from before and is more convenient for readers to understand the relevant rationale and will guide them to think of appropriate improvements in medical image segmentation based on deep learning approaches.

• The paper presents a comprehensive survey detailing advancements in supervised deep learning, particularly U-Net and its variants, to improve segmentation accuracy.
• It demonstrates how refined loss functions and network components address class imbalance and enhance feature extraction for precise medical segmentation.
• The survey also explores weakly supervised methods and future directions such as NAS and GCNs to achieve efficient and interpretable medical image analysis.

Medical Image Segmentation Using Deep Learning: An Examination

The paper "Medical Image Segmentation Using Deep Learning: A Survey" provides an extensive review of recent advancements in applying deep learning techniques to medical image segmentation. This survey explicitly organizes the literature into a hierarchical structure from broad to specific categories, contrasting with conventional grouping methods. It mainly focuses on supervised and weakly supervised learning, omitted unsupervised approaches due to their diminished popularity.

Overview of Methodologies

The survey delineates several key focus areas in the development of deep learning methods for medical image segmentation:

1. Supervised Learning Approaches:
   • Backbone Networks: The survey highlights the prominence of encoder-decoder architectures like U-Net and its variants. These architectures prominently feature skip connections to fuse low-level and high-level image features for enhanced segmentation accuracy.
   • Network Function Blocks: The role of components such as dense connections, inception modules, attention mechanisms, and multi-scale information fusion is thoroughly analyzed. The survey emphasizes the impact of these components in refining feature extraction and representation to improve segmentation performance.
   • Loss Functions: The paper discusses adjustments to traditional loss functions, such as cross-entropy and Dice loss, and introduces more sophisticated versions like Tversky and boundary loss to mitigate class imbalance issues, a frequent challenge in medical imaging.
2. Weakly Supervised Learning:
   • The survey explores data augmentation using GANs, transfer learning to leverage pre-trained models, and interactive segmentation approaches allowing clinician input. These methods address scenarios where large, high-quality labeled datasets are impractical.

Implications and Challenges

The comprehensive review indicates that while significant progress has been made, challenges remain, particularly in achieving high segmentation accuracy for small structures within medical images due to class imbalance. The survey suggests that the integration of more advanced loss functions and leveraging domain knowledge can potentially lead to improved outcomes.

Moreover, the survey highlights the potential of network architecture search (NAS) and graph convolutional neural networks (GCNs) as emerging methodologies to optimize model design and efficiency. The discussion around explainability in deep learning models indicates a growing interest in ensuring both clinical acceptance and regulatory compliance of AI-based tools in healthcare settings.

Future Directions

The survey suggests several avenues for future research and improvement in medical image segmentation:

• Further development of NAS and GCNs for automatic and efficient model design tailored to specific medical imaging tasks.
• Exploration of hybrid models that integrate convolutional and transformer-based networks to capture both local and global image contexts.
• Addressing the limitations of current weakly supervised methods towards achieving results on par with fully supervised approaches through innovative strategies.
• Continued focus on developing interpretable AI models that provide insights into the decision-making process, thereby facilitating their integration into clinical workflows.

Conclusion

This survey represents a substantive resource for researchers in the domain of medical image analysis, providing a structured insight into current methodologies and highlighting both their strengths and limitations. It serves as a foundation upon which upcoming research can build towards developing more sophisticated, efficient, and interpretable deep learning models for medical image segmentation.