Automatic Spine Tissue Segmentation from MRI Data Based on Cascade of Boosted Classifiers and Active Appearance...

Area of Science:

• Medical Imaging
• Machine Learning
• Biomedical Engineering

Background:

• Accurate segmentation of vertebral column tissue from MRI is crucial for diagnosing spinal conditions.
• Manual segmentation is time-consuming and subject to inter-observer variability.
• Existing automated methods may lack the precision required for clinical applications.

Purpose of the Study:

• To develop and validate a novel, multi-stage Machine Learning method for automatic segmentation of vertebral column tissue in MRI images.
• To compare the performance of the automated method against manual segmentation by multiple physicians.
• To assess the algorithm's convergence, reliability, and generalization capabilities.

Main Methods:

• A multi-stage Machine Learning approach was employed, starting with vertebrae recognition using a Cascade Classifier.
• The primary segmentation was performed using a patch-based Active Appearance Model.
• Results were refined using centripetal Catmull-Rom splines and validated through physician-based manual segmentation and 10-fold cross-validation.

Main Results:

• The automated segmentation method achieved performance comparable to physicians (Fractional Figure-of-Merit [FF] = 90.19 ± 1.01%).
• The algorithm demonstrated proper convergence per iteration.
• Automated segmentation showed strong correlation with manual segmentation for both single ([Formula: see text]) and average ([Formula: see text]) measurements (p = 0.05).
• 10-fold cross-validation confirmed good model generalization (FF = 91.37 ± 1.13%).

Conclusions:

• The proposed Machine Learning method offers an accurate and reliable approach for automatic vertebral column tissue segmentation in MRI.
• The method's performance is comparable to expert manual segmentation, suggesting its potential for clinical use.
• The validated generalization indicates the method's robustness and practical applicability in analyzing spinal MRI data.