Comparing Raman Spectroscopy-Based Artificial Intelligence to High-Definition White Light Endoscopy for Endoscopic Diagnosis of Gastric Neoplasia: A Feasibility Proof-of-Concept Study.

BACKGROUND

Endoscopic assessment for the diagnosis of gastric cancer is limited by interoperator variability and lack of real-time capability. Recently, Raman spectroscopy-based artificial intelligence (AI) has been proposed as a solution to overcome these limitations.

OBJECTIVE

To compare the performance of the AI-enabled Raman spectroscopy with that of high-definition white light endoscopy (HD-WLE) for the risk classification of gastric lesions.

METHODS

This was a randomized double-arm feasibility proof-of-concept trial in which participants with suspected gastric neoplasia underwent endoscopic assessment using either the Raman spectroscopy-based AI (SPECTRA IMDx™) or HD-WLE performed by expert endoscopists. Identified lesions were classified in real time as having either low or high risk for neoplasia. Diagnostic outcomes were compared between the two groups using histopathology as the reference.

RESULTS

A total of 20 patients with 25 lesions were included in the study. SPECTRA, in real-time, performed at a statistically similar level to that of HD-WLE performed by expert endoscopists, achieving an overall sensitivity, specificity, and accuracy of 100%, 80%, and 89.0%, respectively, by patient; and 100%, 80%, and 92%, respectively, by lesion, while expert endoscopists using HD-WLE attained a sensitivity, specificity, and accuracy of 100%, 80%, and 90%, respectively, by patient; and 100%, 83.3%, and 91.7%, respectively, by lesion, in differentiating high-risk from low-risk gastric lesions.

CONCLUSIONS

The SPECTRA's comparable performance with that of HD-WLE suggests that it can potentially be a valuable adjunct for less experienced endoscopists to attain accurate and real-time diagnoses of gastric lesions. Larger-scale prospective randomized trials are recommended to validate these promising results further.