ICUnet++: an Inception-CBAM network based on Unet++ for MR spine image segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spine Surgery

Background:

• Spinal diseases necessitate accurate medical image segmentation for diagnosis and treatment.
• Traditional segmentation methods are time-consuming and labor-intensive.
• Automated segmentation of vertebrae and intervertebral discs is crucial for efficient clinical evaluation.

Purpose of the Study:

• To develop an efficient and novel automatic segmentation network model for MR spine images.
• To improve the accuracy and speed of spinal image segmentation compared to traditional methods.
• To introduce the Inception-CBAM Unet++ (ICUnet++) model for enhanced feature extraction and attention-based refinement.

Main Methods:

• Designed the Inception-CBAM Unet++ (ICUnet++) model, integrating Inception modules for multi-receptive field feature extraction.
• Incorporated Attention Gate and CBAM modules to enhance local feature highlighting.
• Evaluated the model using the SpineSagT2Wdataset3 spinal MRI dataset.
• Utilized Intersection over Union (IoU), Dice Similarity Coefficient (DSC), True Positive Rate (TPR), and Positive Predictive Value (PPV) for performance assessment.

Main Results:

• The ICUnet++ model achieved high segmentation performance.
• Achieved IoU of 83.16%, DSC of 90.32%, TPR of 90.40%, and PPV of 90.52%.
• Demonstrated significant improvements in segmentation indicators compared to existing methods.

Conclusions:

• The proposed ICUnet++ model is effective for automatic segmentation of MR spine images.
• The model enhances feature extraction and attention mechanisms for improved accuracy.
• This advancement facilitates quicker and more accurate clinical evaluation of spinal diseases.