Intraoperative integration of structured light scanning for automatic tissue classification: a feasibility study.

Area of Science:

• Medical Imaging
• Machine Learning in Surgery
• Computational Anatomy

Background:

• Structured light scanning (SLS) offers an inexpensive and accurate method for intraoperative imaging.
• Integrating SLS into surgical workflows can facilitate rapid registration and augmentation of preoperative imaging.
• Previous work established the feasibility of SLS for capturing high-accuracy anatomical surface data in real-time.

Purpose of the Study:

• To investigate the feasibility of using machine learning to automatically characterize anatomical tissues.
• To analyze textural and spatial information from SLS scans for tissue identification.
• To enable effective integration of SLS technology into surgical workflows through automated scan analysis.

Main Methods:

• Collected 3D surface scans and tissue features (cartilage, bone, ligament) from seven total knee arthroplasty patients.
• Utilized three fresh-frozen knee cadaver specimens for detailed dissection and annotation.
• Trained and evaluated machine learning models (random forest, neural networks) using annotated textural and spatial data.

Main Results:

• Machine learning models achieved approximately 80% accuracy in classifying bone, cartilage, and ligament.
• Accuracy reached nearly 90% when classifying only bone and cartilage.
• Average accuracy for two- to four-class tissue separation in cadaver data ranged from 76% to 82%.

Conclusions:

• Machine learning methods are feasible for accurate classification of anatomical tissues from intraoperative surface models.
• Automated tissue characterization streamlines the use of structured light scanners in surgery.
• This facilitates applications like 3D surgical documentation and enhanced image registration.