Endoscopic video manifolds for targeted optical biopsy

Area of Science:

• Medical Imaging
• Computer Vision
• Gastroenterology

Background:

• Gastro-intestinal (GI) endoscopy is crucial for diagnosing digestive tract diseases.
• Current surveillance relies on visual assessment and tissue biopsies, which can be challenging for repeat examinations.
• Probe-based confocal laser endomicroscopy (pCLE) offers non-invasive optical biopsies, but identifying previous sites is difficult.

Purpose of the Study:

• To develop a computational framework for retargeting previous optical biopsy sites in surveillance upper GI-endoscopies.
• To introduce a novel representation for endoscopic videos called endoscopic video manifolds (EVMs).
• To facilitate accurate site recognition in non-invasive optical biopsy procedures.

Main Methods:

• A clustering and classification framework was developed.
• A new low-dimensional representation, endoscopic video manifolds (EVMs), was proposed for endoscopic videos.
• The EVM representation was adapted for clustering informative frames and patient-specific endoscopic segments.

Main Results:

• The proposed EVM representation demonstrated improved performance compared to original image representation and principal component analysis.
• The framework was validated on three in vivo patient datasets.
• Quantitative analysis confirmed the effectiveness of the EVM approach.

Conclusions:

• The developed framework shows feasibility in assisting endoscopists to retarget optical biopsy sites.
• Endoscopic video manifolds (EVMs) provide an effective representation for analyzing endoscopic video data.
• This method represents a promising advancement for improving surveillance in GI endoscopy.