A robust and efficient approach to detect 3D rectal tubes from CT colonography

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiology

Background:

• Rectal tubes (RTs) are a frequent cause of false positives (FPs) in CT colonography computer-aided detection (CAD) systems.
• Improving RT detection enhances CAD performance and radiologist confidence.

Purpose of the Study:

• To develop a novel, robust, and efficient method for detecting rectal tubes (RTs) in CT colonography.
• To reduce false positives in CAD systems by accurately identifying RTs.

Main Methods:

• A bottom-up approach utilizing probabilistic models with kernel density estimation for local RT detection.
• A RANSAC-based shape model for inferring the global RT path.
• Subimage projection into a subspace and quadratic discriminant analysis (QDA) for RT classification.
• Bottom-top clustering to determine the RT tip position.

Main Results:

• The method achieved a 99.5% detection rate for annotated RT paths on a diverse dataset.
• When integrated with CAD, the method removed 98.4% of RT-induced FPs.
• No loss of sensitivity was observed during CAD evaluation.

Conclusions:

• The proposed method effectively detects rectal tube paths in CT colonography.
• RT detection significantly reduces CAD-generated false positives without compromising sensitivity.