Computer assisted intervention to improve diagnostic yield in capsule endoscopy post ingestion.

Area of Science:

• Medical Imaging
• Gastroenterology
• Computer Science

Background:

• Wireless capsule endoscopy (WCE) quality is often compromised by visibility issues like debris or bubbles.
• Current methods assess WCE quality retrospectively, limiting real-time corrective actions.
• Early detection of non-diagnostic examinations is crucial for timely intervention.

Purpose of the Study:

• To develop a feasibility-aware computer-assisted intervention (CAI) framework for real-time WCE quality management.
• To reformulate WCE quality assessment as a sequential decision problem during acquisition.
• To enable early, anatomically and temporally consistent intervention recommendations.

Main Methods:

• The CAI framework analyzes early WCE frames (10-25%) to assess visibility impairment.
• Key metrics include mean impairment severity, temporal instability, and degradation mechanisms.
• Logistic normalization estimates downstream non-diagnostic risk, incorporating feasibility constraints like transit time.

Main Results:

• The framework demonstrated non-trivial discrimination of non-diagnostic outcomes (AUROC ≈ 0.78) using early visibility data.
• While 'worst-window collapse' showed higher discrimination (AUROC 0.89), the CAI risk offered stable performance.
• CAI preserved mechanism identifiability and decision stability for intervention selection.

Conclusions:

• WCE quality management can be effectively addressed as a sequential, feasibility-aware CAI problem.
• The proposed framework facilitates early, mechanism-consistent, and anatomically feasible interventions.
• This approach shifts from retrospective assessment to proactive quality control in WCE.