Computer-aided diagnosis for optical diagnosis of diminutive colorectal polyps including sessile serrated lesions: a real-time comparison with screening endoscopists

Abstract

BACKGROUND : We aimed to compare the accuracy of the optical diagnosis of diminutive colorectal polyps, including sessile serrated lesions (SSLs), between a computer-aided diagnosis (CADx) system and endoscopists during real-time colonoscopy. METHODS : We developed the POLyp Artificial Recognition (POLAR) system, which was capable of performing real-time characterization of diminutive colorectal polyps. For pretraining, the Microsoft-COCO dataset with over 300 000 nonpolyp object images was used. For training, eight hospitals prospectively collected 2637 annotated images from 1339 polyps (i. e. publicly available online POLAR database). For clinical validation, POLAR was tested during colonoscopy in patients with a positive fecal immunochemical test (FIT), and compared with the performance of 20 endoscopists from eight hospitals. Endoscopists were blinded to the POLAR output. Primary outcome was the comparison of accuracy of the optical diagnosis of diminutive colorectal polyps between POLAR and endoscopists (neoplastic [adenomas and SSLs] versus non-neoplastic [hyperplastic polyps]). Histopathology served as the reference standard. RESULTS : During clinical validation, 423 diminutive polyps detected in 194 FIT-positive individuals were included for analysis (300 adenomas, 41 SSLs, 82 hyperplastic polyps). POLAR distinguished neoplastic from non-neoplastic lesions with 79 % accuracy, 89 % sensitivity, and 38 % specificity. The endoscopists achieved 83 % accuracy, 92 % sensitivity, and 44 % specificity. The optical diagnosis accuracy between POLAR and endoscopists was not significantly different (P = 0.10). The proportion of polyps in which POLAR was able to provide an optical diagnosis was 98 % (i. e. success rate). CONCLUSIONS : We developed a CADx system that differentiated neoplastic from non-neoplastic diminutive polyps during endoscopy, with an accuracy comparable to that of screening endoscopists and near-perfect success rate.

Trial registration: ClinicalTrials.gov NCT03822390.

Fig. 1

POLAR system design and user protocol during clinical validation. When using the POLAR system during clinical validation, the endoscopist has to take a maximum of three nonmagnified images using narrow-band imaging. After one image of the lesion is taken, it is processed by the POLAR system. If the system is able to provide a high-confidence diagnosis (a green mark is shown on the monitor), the endoscopist can continue with the procedure (i. e. resect the lesion). If the system is not able to provide a high-confidence diagnosis, the system provides feedback to the endoscopist on why this was not possible (e. g. not able to localize the lesion [red mark], not of sufficient quality [red mark], or only able to perform an optical diagnosis with low confidence [orange mark]). If the system is not able to provide a high-confidence diagnosis, the endoscopist has to take another image, up to a maximum of three per lesion. If the system is still not able to provide a high-confidence diagnosis after three images, the endoscopists can stop taking images, and proceed with the procedure. The low-confidence diagnosis with the highest prediction score is used as the final diagnosis of the system. If the system is not able to provide a low-confidence diagnosis after three images, this is considered as a failure of the system. NBI, narrow-band imaging.

Fig. 2

Examples of polyp images with the POLAR system output. a The POLAR system is not confident enough about the localization (8.5 %). b, c The area within the bounding box is not sharp enough or is overexposed. d The POLAR system predicts the histology of the polyp with only low confidence. e, f The POLAR system predicts the histology of the polyps with high confidence (blinded prediction [ e ], visible prediction [ f ]).

Fig. 3

Study flow chart. CADx, computer-aided diagnosis; SSL, sessile serrated lesion; HPP, hyperplastic polyp.

Fig. 4

Diagnostic performance of both the computer-aided diagnosis system and endoscopists for differentiating neoplastic from non-neoplastic diminutive lesions. a Diminutive lesions overall. b Diminutive lesions located in the rectosigmoid. CADx, computer-aided diagnosis; PPV, positive predictive value; NPV, negative predictive value.