Dysplasia discrimination in intestinal-type neoplasia of the esophagus and colon via digital image analysis.

Determining gastrointestinal tract dysplasia level is clinically important but can be difficult, and given this challenge, we investigated colonic and esophageal dysplastic progression using digital image analysis (IA). Whole slide images were obtained for colonic normal mucosa (NCM), hyperplastic polyps (HP), conventional tubular adenomas (TA), and adenomas with high-grade dysplasia (HGD), and esophageal intestinal metaplasia negative for dysplasia (IM), indefinite for dysplasia (IFD), low-grade dysplasia (LGD), and HGD. Characteristic nuclei were circumscribed, and parameters discriminating groups included nuclear circumference (μm), area (μm(2)), and 15 positive pixel count (PPC) algorithm IA measurements. In colon polyps and esophageal lesions, average nuclear area and circumference ranged 30-108.6 μm(2) and 27.5-48.9 μm, respectively. Differences for average nuclear area and circumference met statistical significance (p < 0.05) between diagnostic groups in the esophagus and colon, except for IM versus IFD nuclear area. Pixel intensity (brightness) separated lesions within both groups with statistical significance except for colonic TAs versus HPs and esophageal LGD versus IM. HGD nuclei in both groups demonstrated more pixel staining heterogeneity than other lesions. Hierarchical clustering and principal component analysis demonstrated that lesions with similar diagnoses tended to cluster together on a low- to high-grade spectrum. Our results confirm that quantitative IA is an effective adjunct reflecting dysplasia in colon polyps and Barrett esophagus lesions. Nuclear area, circumference, and PPC algorithm findings distinguished lesions in a statistically significant manner. This suggests utility for future studies on similar methods, which may provide an adjunctive ancillary technique for pathologists and enhance patient care.