Real-time Use of Computer-aided Diagnosis in the Optical Diagnosis of Gastric Neoplasia: A Multicenter Randomized Controlled Trial.

Abstract

Background & aims: The multicenter, randomized, control trial was conducted to evaluate whether computer-aided diagnosis (CADx) improves the optical diagnosis of gastric neoplasia in nonexpert endoscopists.

Methods: Patients undergoing endoscopy prior to endoscopic resection (ER) of gastric neoplasia or for surveillance after ER were randomized to the CADx-assisted and non-CADx-assisted groups. Endoscopy was performed by a nonexpert endoscopist blinded to the patient information. The CADx system provides a confidence level for gastric neoplasia on a still white light image (WLI) captured during endoscopy and indicates neoplasia if the level is ≥ a 60% cutoff. The targets were lesions diagnosed as neoplasia and those diagnosed as non-neoplasia but requiring biopsy. Endoscopists performed WLI diagnoses to differentiate between neoplasia and non-neoplasia. The accuracy, sensitivity, and specificity in nonexperts have been evaluated using pathology as the gold standard.

Results: A total of 312 patients with 265 gastric neoplasias and 164 non-neoplasias were enrolled. The accuracy, sensitivity, and specificity among nonexperts were 65.3% vs 59.9% (P = .24), 68.6% vs 63.9% (P = .42), and 60.8% vs 53.3% (P = .34) in the CADx-assisted and non-CADx-assisted groups, respectively. Flat or superficially depressed small neoplasias tended to be missed by the CADx.

Conclusions: The present CADx assistance did not improve the nonexpert optical diagnosis of gastric neoplasia. Further improvements of the CADx with WLI are needed by changing the confidence level cutoff and additional training on easily missed neoplasias. Another option is CADx with magnifying narrow band imaging. (Japan Registry of Clinical Trials, number jRCTs032210171).