Habitat radiomics plus 3D deep learning on noncontrast chest CT detects early esophageal cancer

A model integrating habitat radiomics, 3D deep learning, and clinical variables detected early-stage esophageal squamous cell carcinoma on noncontrast chest CT with AUCs of 0.895 (internal) and 0.847 (external). AI assistance improved radiologists' sensitivity and inter-reader a…

• The Combined model yielded AUCs of 0.895 (internal validation) and 0.847 (external test) for early-stage esophageal squamous cell carcinoma.
• AI assistance increased sensitivity for all four radiologists and improved inter-reader agreement in an exploratory reader study.
• The study was retrospective and case-control; prospective validation is needed to confirm real-world utility.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.