SVM based colon polyps classifier in a wireless active stereo endoscope.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Colon polyps require accurate classification to differentiate between hyperplastic and adenomatous types.
• Active stereo vision sensors offer potential for real-time 3D reconstruction of colon polyps.

Purpose of the Study:

• To develop and evaluate a system for the accurate 3D recognition and classification of colon polyps.
• To differentiate between hyperplastic and adenomatous polyps using advanced feature descriptors.

Main Methods:

• Utilized an active stereo vision sensor (Cyclope) for real-time 3D object reconstruction.
• Employed a Support Vector Machine (SVM) classifier trained on novel depth histogram features.
• Combined stereo measurements with SVM for polyp classification.

Main Results:

• Achieved a correct classification rate of approximately 97% for distinguishing between polyp types.
• Demonstrated the effectiveness of depth histogram features in polyp recognition.
• Provided detailed statistics on detection rates and computational complexity.

Conclusions:

• The developed algorithm shows high accuracy in classifying colon polyps.
• Depth histogram features combined with SVM offer a promising approach for polyp detection.
• The Cyclope sensor prototype shows potential for clinical applications in polyp analysis.