Computer-aided prediction of polyp histology on white light colonoscopy using surface pattern analysis.

Area of Science:

• Gastroenterology
• Medical Imaging
• Computer Science

Background:

• Colorectal polyps require accurate histological assessment to determine malignancy risk.
• Distinguishing between dysplastic and nondysplastic polyps is crucial for patient management.
• Current diagnostic methods rely on pathological examination and endoscopist interpretation, which can be subjective.

Purpose of the Study:

• To evaluate a novel computational histology prediction system for colorectal polyps.
• To assess the system's accuracy in classifying polyp histology based on textural surface patterns.
• To compare the system's performance against pathological diagnosis and expert endoscopist assessments.

Main Methods:

• A computer-aided diagnosis (CAD) system was developed to analyze textural elements (textons) of colorectal polyps from high-definition white light images.
• Textural features such as contrast, shape, and bifurcation of surface patterns were characterized.
• The CAD system's predictions were compared with pathological diagnoses and classifications by endoscopists using established criteria (Kudo and NBI International Colorectal Endoscopic classifications).

Main Results:

• The CAD system correctly classified 91.1% of 225 evaluated polyps (142 dysplastic, 83 nondysplastic).
• Accuracy for dysplastic polyps was 92.3%, and for nondysplastic polyps, it was 89.2%.
• In a subgroup of 100 diminutive polyps (≤5mm), the system achieved 87.0% accuracy, with no significant difference compared to endoscopist assessments.

Conclusions:

• A computer vision system utilizing polyp surface characterization in white light images accurately predicts colorectal polyp histology.
• The developed system demonstrates high diagnostic performance, comparable to experienced endoscopists.
• This technology holds potential for improving the accuracy and efficiency of colorectal polyp diagnosis.