Automated Joint Space Detection Improves Bone Segmentation Accuracy

作者： H. Mark Kenney1,2,3, Daniel Lichau4, Rémi Blanc4, Lindsay Schnur1, Richard D. Bell5, Christopher T. Ritchlin1,3, Edward M. Schwarz1,2,3,6,7,8, Hani A. Awad1,7,8, Ronald W. Wood1,6,9,10 1Center for Musculoskeletal Research,University of Rochester Medical Center, 2Department of Pathology & Laboratory Medicine,University of Rochester Medical Center, 3Department of Medicine,University of Rochester Medical Center, 4Thermo Fisher Scientific, 5Arthritis and Tissue Degeneration Program, Department of Research,Hospital for Special Surgery, 6Department of Urology,University of Rochester Medical Center, 7Department of Biomedical Engineering,University of Rochester Medical Center, 8Department of Orthopaedics,University of Rochester Medical Center, 9Department of Obstetrics and Gynecology,University of Rochester Medical Center, 10Department of Neuroscience,University of Rochester Medical Center

简介：

Overview

This study presents an automated joint space detection workflow that achieves high-throughput segmentation of murine hindpaw bones with over 98% accuracy. The method is adaptable for use in forepaws and paws affected by inflammatory-erosive arthritis, although performance may require further optimization.

Key Study Components

Area of Science

• Neuroscience
• Image Processing
• Bone Health Assessment

Background

• High throughput image processing algorithms have improved bone segmentation.
• Prior segmentation strategies face challenges in reproducibility and manual correction.
• Automated image analysis is needed for quantitative bone health assessment.
• Micro computed tomography datasets can distinguish bone and soft tissue effectively.

Purpose of Study

• To develop a reproducible automated image analysis workflow.
• To enable quantitative assessment of bone health and osteoarthritis progression.
• To minimize manual labor in image segmentation processes.

Methods Used

• Utilization of a supervised watershed-based segmentation method.
• Application of a pre-trained deep learning joint space prediction model.
• Integration with Amira software for image processing.
• Evaluation of segmentation accuracy improvements.

Main Results

• Achieved over 98% accuracy in hindpaw bone segmentation.
• Identified limitations in segmentation performance for paws with arthritis.
• Demonstrated the effectiveness of combining deep learning with traditional methods.
• Highlighted areas for future optimization in segmentation workflows.

Conclusions

• The automated workflow significantly enhances bone segmentation accuracy.
• Further optimization is needed for applications in inflammatory conditions.
• This approach can facilitate better assessment of osteoarthritis progression.

Frequently Asked Questions