An Automated Deep Learning Approach for Spine Segmentation and Vertebrae Recognition Using Computed Tomography Images

Area of Science:

• Medical imaging
• Artificial intelligence
• Spine analysis

Background:

• Accurate spine image analysis requires precise segmentation and vertebrae recognition.
• Current deep learning models for this task are computationally intensive.

Purpose of the Study:

• To introduce a novel, computationally efficient deep learning model for spine segmentation and vertebrae recognition using CT images.
• To improve accuracy in spine analysis compared to existing methods.

Main Methods:

• A two-step approach using a cascaded hierarchical atrous spatial pyramid pooling residual attention U-Net (CHASPPRAU-Net) for spine segmentation.
• A 3D mobile residual U-Net (MRU-Net) incorporating MobileNetv2, residual, and attention modules for vertebrae recognition from multi-view 3D spine images.
• Validation on the VerSe 20 and VerSe 19 datasets.

Main Results:

• The proposed model demonstrated higher accuracy in both spine segmentation and vertebrae recognition.
• Achieved superior performance over current state-of-the-art methods on benchmark datasets.

Conclusions:

• The novel deep learning model offers an effective and accurate solution for spine segmentation and vertebrae recognition.
• This approach presents a computationally efficient advancement in medical image analysis for spinal disorders.