Wavelet method for CT colonography computer-aided polyp detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• Colorectal cancer (CRC) incidence is reduced by detecting and removing colon polyps.
• Computed tomographic colonography (CTC) computer-aided detection (CAD) effectively identifies large polyps (≥1 cm).
• CTC CAD exhibits lower sensitivity and higher false positive rates for medium-sized polyps (6-9 mm), crucial for patient management.

Purpose of the Study:

• To develop and evaluate a wavelet-based postprocessor for reducing false positives in CTC CAD for medium-sized polyps.
• To assess the impact of the postprocessor on detection sensitivity for polyps in the 6-9 mm size range.

Main Methods:

• A wavelet-based postprocessor was developed to refine CTC CAD findings.
• The postprocessor was applied to CTC CAD data from 44 patients with 45 confirmed medium-sized polyps (6-9 mm).
• Performance was evaluated using fourfold cross-validation with 5000 bootstraps, comparing pre- and post-processing metrics.

Main Results:

• The wavelet-based postprocessor reduced false positives by 56.61% (from 12.4 to 5.38 per patient, p <0.001).
• Sensitivity for detecting 6-9 mm polyps remained high, with only a minor, non-significant decrease (73.33% to 71.11%, p=0.1713).
• The 95% confidence interval for sensitivity was [66.39%, 75.74%].

Conclusions:

• The wavelet-based postprocessor substantially reduces false positives in CTC CAD for medium-sized colon polyps.
• This technique improves the clinical utility of CTC CAD by enhancing accuracy for clinically important polyp sizes.
• The method offers a promising approach to optimize colon polyp detection and reduce unnecessary follow-ups.