Multiple Axial Spine Indices Estimation via Dense Enhancing Network With Cross-Space Distance-Preserving...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spine Biomechanics

Background:

• Manual measurement of axial spine indices is time-consuming, laborious, and prone to inaccuracies.
• Accurate spine indices are crucial for various computer-aided spine procedures, including diagnosis, treatment evaluation, and biomechanical modeling.

Purpose of the Study:

• To develop an automated method for estimating multiple axial spine indices from medical images.
• To improve the accuracy and efficiency of spine index measurement for clinical applications.

Main Methods:

• Proposed a novel deep learning model, the dense enhancing network (DE-Net), incorporating dense enhancing blocks (DEBs) with feature enhancement.
• Introduced cross-space distance-preserving regularization (CSDPR) to optimize the network's loss function.
• Trained and validated the model on 895 axial spine MRI images from 143 subjects, using manually measured indices as ground truth.

Main Results:

• The proposed DE-Net with CSDPR achieved the smallest prediction error among all evaluated deep learning models.
• All deep learning models demonstrated very small prediction errors compared to manual measurements.
• The method shows significant potential for enhancing computer-aided spine procedures.

Conclusions:

• The developed DE-Net with CSDPR offers a highly accurate and efficient automated solution for axial spine index estimation.
• This automated approach can overcome the limitations of manual measurements, benefiting clinical practice and research.
• The method holds great promise for advancing computer-aided diagnosis, treatment planning, and understanding of spinal pathologies.