AI bests non-expert endoscopists in Barrett’s neoplasia detection
Researchers developed a deep-learning, computer aided system that detected neoplasia in patients with Barrett’s esophagus at a better rate than nonspecialized endoscopists, according to study results.
J.J. Bergman, MD, PhD, and colleagues wrote that technological improvements have aided the detection of neoplasia, but detection failure is still not uncommon, particularly among some groups of clinicians.
“This problem arises because general endoscopists are infrequently confronted with early neoplasia in BE, since progression to neoplasm is rare,” they wrote. “General endoscopists thus have limited familiarity with the endoscopic appearances of early BE neoplasia.”
Investigators sought to develop a computer-aided detection (CAD) system that could be used in real-time, clinical practice and improve detection of neoplasia in patients with BE. They built their system using five independent endoscopy datasets and pre-trained it using nearly 500,000 labelled endoscopic images collected from all intestinal segments.
This training included more than 1,700 unique esophageal images collected from 669 patients. The developers tested the systems performance using datasets 4 and 5, which was also scored by 53 general endoscopists from four countries.
Bergman and colleagues found that in dataset 4, their CAD system was able to classify images containing neoplasms with 89% accuracy, 90% sensitivity and 88% specificity. In dataset 5, the CAD system bested the endoscopists in accuracy (88% vs. 73%), sensitivity (93% vs. 72%) and specificity (83% vs. 74%).
Additionally, the CAD system had better accuracy than any individual endoscopist and did a comparable job at delineation.
“The system detected neoplasia with high accuracy and near-perfect localization. The CAD system achieved higher accuracy than a panel of non-expert endoscopist assessors, strongly suggesting that CAD may improve the accuracy of surveillance-detected early BE neoplasia by general endoscopists,” Bergman and colleagues wrote. “We anticipate that the system’s performance will further improve by expanding the number of images in the different database layers of our deep learning system and by including video recordings.” – by Alex Young
Disclosures: Bergman reports receiving research support from Fujifilm and NinePoint Medical, as well as speaking fees from Fujifilm. Please see the full study for all other authors’ relevant financial disclosures.