RESUMO
BACKGROUND: We aimed to develop a convolutional neural network (CNN) model for detecting neoplastic lesions during real-time digital single-operator cholangioscopy (DSOC) and to clinically validate the model through comparisons with DSOC expert and nonexpert endoscopists. METHODS: In this two-stage study, we first developed and validated CNN1. Then, we performed a multicenter diagnostic trial to compare four DSOC experts and nonexperts against an improved model (CNN2). Lesions were classified into neoplastic and non-neoplastic in accordance with Carlos Robles-Medranda (CRM) and Mendoza disaggregated criteria. The final diagnosis of neoplasia was based on histopathology and 12-month follow-up outcomes. RESULTS: In stage I, CNN2 achieved a mean average precision of 0.88, an intersection over the union value of 83.24â%, and a total loss of 0.0975. For clinical validation, a total of 170 videos from newly included patients were analyzed with the CNN2. Half of cases (50â%) had neoplastic lesions. This model achieved significant accuracy values for neoplastic diagnosis, with a 90.5â% sensitivity, 68.2â% specificity, and 74.0â% and 87.8â% positive and negative predictive values, respectively. The CNN2 model outperformed nonexpert #2 (area under the receiver operating characteristic curve [AUC]-CRM 0.657 vs. AUC-CNN2 0.794, Pâ<â0.05; AUC-Mendoza 0.582 vs. AUC-CNN2 0.794, Pâ<â0.05), nonexpert #4 (AUC-CRM 0.683 vs. AUC-CNN2 0.791, Pâ<â0.05), and expert #4 (AUC-CRM 0.755 vs. AUC-CNN2 0.848, Pâ<â0.05; AUC-Mendoza 0.753 vs. AUC-CNN2 0.848, Pâ<â0.05). CONCLUSIONS: The proposed CNN model distinguished neoplastic bile duct lesions with good accuracy and outperformed two nonexpert and one expert endoscopist.
Assuntos
Inteligência Artificial , Neoplasias , Humanos , Redes Neurais de Computação , Curva ROC , Valor Preditivo dos TestesRESUMO
BACKGROUND: Currently, there is insufficient data about the accuracy in the diagnosing of pancreatic cystic lesions (PCLs), especially with novel endoscopic techniques such as with direct intracystic micro-forceps biopsy (mFB) and needle-based confocal laser-endomicroscopy (nCLE). AIM: To compare the accuracy of endoscopic ultrasound (EUS) and associated techniques for the detection of potentially malignant PCLs: EUS-guided fine needle aspiration (EUS-FNA), contrast-enhanced EUS (CE-EUS), EUS-guided fiberoptic probe cystoscopy (cystoscopy), mFB, and nCLE. METHODS: This was a single-center, retrospective study. We identified patients who had undergone EUS, with or without additional diagnostic techniques, and had been diagnosed with PCLs. We determined agreement among malignancy after 24-mo follow-up findings with detection of potentially malignant PCLs via the EUS-guided techniques and/or EUS-guided biopsy when available (EUS malignancy detection). RESULTS: A total of 129 patients were included, with EUS performed alone in 47/129. In 82/129 patients, EUS procedures were performed with additional EUS-FNA (21/82), CE-EUS (20/82), cystoscopy (27/82), mFB (36/82), nCLE (44/82). Agreement between EUS malignancy detection and the 24-mo follow-up findings was higher when associated with additional diagnostic techniques than EUS alone [62/82 (75.6%) vs 8/47 (17%); OR 4.35, 95%CI: 2.70-7.37; P < 0.001]. The highest malignancy detection accuracy was reached when nCLE and direct intracystic mFB were both performed, with a sensitivity, specificity, positive predictive value, negative predictive value and observed agreement of 100%, 89.4%, 77.8%, 100% and 92.3%, respectively (P < 0.001 compared with EUS-alone). CONCLUSION: The combined use of EUS-guided mFB and nCLE improves detection of potentially malignant PCLs compared with EUS-alone, EUS-FNA, CE-EUS or cystoscopy.