Introduction

Diagnosis and characterization of biliary strictures are challenging. The introduction of digital single-operator cholangioscopy (D-SOC) significantly improved the diagnostic yield of indeterminate biliary strictures. Pilot studies using artificial intelligence (AI) algorithms applied to D-SOC have shown promising results. We aimed to develop an AI system based on convolutional neural networks (CNN) for the automatic detection and characterization of malignant biliary strictures in D-SOC images.

Aims & Methods

Our group developed, trained, and validated a CNN based on D-SOC images from two centers in Portugal and Spain. Each frame was labeled as normal/benign findings or as a malignant lesion if histopathological evidence of biliary malignancy was available. Also, we evaluated the network's performance for detecting morphologic features, including tumor vessels, papillary projections, and nodules. The image dataset was split for the constitution of training and validation datasets. The performance of the CNN was measured by calculating the area under the curve (AUC), sensitivity, specificity, and positive and negative predictive values (PPV and NPV, respectively).

Results

A total of 84994 images from 130 D-SOC exams were included (44743 of malignant strictures and 40521 of benign findings). The model had an overall accuracy of 98.3%, a sensitivity of 98.3%, a specificity of 98.3%, and an AUROC of 1.00.
Our group evaluated the performance of the CNN for the detection of morphological characteristics associated with malignancy, particularly tumor vessels and papillary projection. The CNN detected tumor vessels with a sensitivity of 95.7%, a specificity of 88.6%, and an overall accuracy of 93.0%. Papillary projections were identified by the CNN with a sensitivity of 74.1%, a specificity of 94.5%, and an overall accuracy of 91.2%.
The CNN completed the reading of the 4250 batches of images (each batch comprising 128 images) in 23 minutes and 16 seconds, which can be translated to an approximate reading rate of 390 frames per second.

Conclusion

The developed deep learning algorithm accurately detected and differentiated malignant strictures from benign biliary conditions. Furthermore, the AI model provided accurate morphological characterization of the biliary strictures.
The potential of deep learning algorithms to impact the care of patients with suspected biliary malignancy is vast. This study assessed the performance of a CNN for detecting and differentiating malignant and benign biliary disorders using a large pool of D-SOC images. The excellent performance of this model lays the foundations for future exploration of AI technologies in this subset of patients, aiming to improve the clinical outcome of patients with suspected biliary malignancy.

References

1 Larghi A, Tringali A, Lecca PG, Giordano M, Costamagna G. Management of hilar biliary strictures. Am J Gastroenterol. 2008;103(2):458-73.
2 Tummala P, Munigala S, Eloubeidi MA, Agarwal B. Patients with obstructive jaundice and biliary stricture ± mass lesion on imaging: prevalence of malignancy and potential role of EUS-FNA. J Clin Gastroenterol. 2013;47(6):532-7.
3 Paranandi B, Oppong KW. Biliary strictures: endoscopic assessment and management. Frontline Gastroenterol. 2017;8(2):133-7.
4 Singhi AD, Slivka A. Evaluation of indeterminate biliary strictures: Is there life on MARS? Gastrointestinal Endoscopy. 2020;92(2):320-2.
5 Burnett AS, Calvert TJ, Chokshi RJ. Sensitivity of endoscopic retrograde cholangiopancreatography standard cytology: 10-y review of the literature. J Surg Res. 2013;184(1):304-11.
6 Navaneethan U, Njei B, Lourdusamy V, Konjeti R, Vargo JJ, Parsi MA. Comparative effectiveness of biliary brush cytology and intraductal biopsy for detection of malignant biliary strictures: a systematic review and meta-analysis. Gastrointest Endosc. 2015;81(1):168-76.
7 Pereira P, Vilas-Boas F, Peixoto A, Andrade P, Lopes J, Macedo G. How SpyGlass™ May Impact Endoscopic Retrograde Cholangiopancreatography Practice and Patient Management. GE Port J Gastroenterol. 2018;25(3):132-7.