Dataset Information

Deep-Learning for the Diagnosis of Esophageal Cancers and Precursor Lesions in Endoscopic Images: A Model Establishment and Nationwide Multicenter Performance Verification Study.

ABSTRACT:

Background

Suspicion of lesions and prediction of the histology of esophageal cancers or premalignant lesions in endoscopic images are not yet accurate. The local feature selection and optimization functions of the model enabled an accurate analysis of images in deep learning.

Objectives

To establish a deep-learning model to diagnose esophageal cancers, precursor lesions, and non-neoplasms using endoscopic images. Additionally, a nationwide prospective multicenter performance verification was conducted to confirm the possibility of real-clinic application.

Methods

A total of 5162 white-light endoscopic images were used for the training and internal test of the model classifying esophageal cancers, dysplasias, and non-neoplasms. A no-code deep-learning tool was used for the establishment of the deep-learning model. Prospective multicenter external tests using 836 novel images from five hospitals were conducted. The primary performance metric was the external-test accuracy. An attention map was generated and analyzed to gain the explainability.

Results

The established model reached 95.6% (95% confidence interval: 94.2-97.0%) internal-test accuracy (precision: 78.0%, recall: 93.9%, F1 score: 85.2%). Regarding the external tests, the accuracy ranged from 90.0% to 95.8% (overall accuracy: 93.9%). There was no statistical difference in the number of correctly identified the region of interest for the external tests between the expert endoscopist and the established model using attention map analysis (P = 0.11). In terms of the dysplasia subgroup, the number of correctly identified regions of interest was higher in the deep-learning model than in the endoscopist group, although statistically insignificant (P = 0.48).

Conclusions

We established a deep-learning model that accurately classifies esophageal cancers, precursor lesions, and non-neoplasms. This model confirmed the potential for generalizability through multicenter external tests and explainability through the attention map analysis.

SUBMITTER: Gong EJ

PROVIDER: S-EPMC9320232 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Background and study aims Traditionally in the case of a vascular interposition, endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) has been contraindicated. A transvascular route (TV) is feasible and probably a safe alternative approach in selected patients, but data are scarce. The primary aim of this study was to analyze the diagnostic yield and safety of EUS-TV-FNA in thoracic and abdominal lesions. Secondary aims included evaluation of the clinical impact and technical aspects. Patients and methods A retrospective multicenter study was conducted with inclusion of all consecutive patients that underwent EUS-TV-FNA from July 2007 to January 2020. Feasibility, cytopathology, procedure details, and safety were evaluated. Univariate analysis was performed to identify variables associated with incidents, cytopathological diagnosis, and clinical impact. Results Data were collected from a total of 49 cases and 50 EUS-TV-FNAs. The aorta (n = 19) and portal system (n = 17) were the most frequently punctured. The most frequent lesions were mediastinal lymph nodes (n = 13) and pancreatic tumors (n = 11). The diagnostic yield was 86 %, and there were nondiagnostic samples in seven cases. Overall sensitivity, specificity, and accuracy were 88 % (95 %CI,0.74-0.96), 100 % (95 %CI,0.59-1), and 90 % (95 %CI,0.78-0.96), respectively. Only three incidents were detected: two mural hematomas and a self-limited bleeding of gastroduodenal artery. In most patients, there was a significant impact on clinical management (88 %). Arterial vessel and ASA-III had a trend with incidents (both, P < 0.08). Rapid on-site evlauation was found to be an independent predictor for obtaining a conclusive sample (OR 6.2; 95 %CI, 1.06-36.73, P < 0.04). Conclusions EUS-TV-FNA is feasible, seems to be safe, and can be recommended when no other targets are available, and the information obtained would impact on the clinical plan.

Project description:BackgroundDietary patterns and symptoms research among Chinese with esophageal squamous cell carcinoma (ESCC) and its precursor lesions is limited, especially as it relates to multiple food consumption and multiple co-occurring symptoms. The aim of our study was to identify the dietary patterns and severity of symptom classes with the risk of esophageal squamous cell carcinoma and its histological precursor lesions, and develop a risk prediction model for different stages of esophageal disease.MethodsWe analyzed data from a multicenter cross-sectional study carried out in ESCC high incidence areas between 2017 and 2018, which included 34,707 individuals aged 40-69 years. Dietary patterns and severity of symptom classes were derived by applying a latent class analysis (LCA). A multiple logistic regression model was used to derive the odds ratio (ORs) and corresponding 95% confidence intervals (CIs) for ESCC and the different stages of esophageal disease according to the dietary patterns and severity of symptom classes identified. We built the risk prediction model by using a nomogram.ResultsWe identified five dietary patterns and three severity of symptom classes. The dietary patterns were classified as follows: "Healthy", "Western", "Lower consumers-combination", "Medium consumers-combination" and "Higher consumers-combination" patterns based on the intake of foods such as red meat, vegetables and fruits. The severity of symptoms was categorized into "Asymptomatic", "Mild symptoms" and "Overt symptoms" classes based on health-related symptoms reported by the participants. Compared to the "Healthy" pattern, the other four patterns were all associated with an increased risk of esophageal disease. Similarly, the other two symptom classes present different degrees of increased risk of esophageal disease compared to the "Asymptomatic". The nomograms reflect the good predictive ability of the model.ConclusionAmong individuals aged 40-69 years in high incidence regions of upper gastrointestinal cancer, the results supplied that subjects with diets rich in livestock and poultry meat and low in fruits and vegetables and subjects with typical symptoms were at increased ESCC risk. The findings highlight the importance of considering food and symptom combinations in cancer risk evaluation.