Deep learning for automatic vertebra analysis: A methodological survey of recent advances.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spine Surgery

Background:

• Automated vertebra analysis (AVA) is crucial for computer-aided diagnosis, surgical planning, and postoperative evaluation in spine workflows.
• Current AVA methods face challenges including field of view variations, complex morphology, limited annotated data, and domain shift performance degradation.

Purpose of the Study:

• To provide an up-to-date, systematic review of recent deep learning (DL) advancements in automated vertebra analysis (AVA).
• To address the knowledge gap caused by the rapid evolution of DL methods in AVA.

Main Methods:

• Consolidated publicly available datasets and evaluation metrics for standardized benchmarking.
• Analyzed recent DL-based AVA approaches focusing on network architecture improvements and learning strategy designs.
• Examined persistent technical barriers and emerging clinical needs shaping future research.

Main Results:

• Identified key challenges in AVA, including data scarcity and domain generalization.
• Reviewed DL advancements in network architectures and learning paradigms for AVA.
• Highlighted emerging research directions such as multimodal learning and foundation models.

Conclusions:

• This review offers a comprehensive synthesis of current DL-based AVA research.
• It guides future research towards developing robust, generalizable, and clinically deployable AVA systems.
• Emphasizes the integration of advanced techniques like multimodal learning and foundation models for intelligent medical imaging.