Phát triển thuật toán trí tuệ nhân tạo trong phát hiện viêm thực quản trào ngược trên tập ảnh nội soi

0 Người đánh giá. Xếp hạng trung bình 0

Tác giả: Trí Thức Bùi, Việt Hằng Đào, Ngọc Hoa Lâm, Thị Ly Vũ

Ngôn ngữ: vie

Ký hiệu phân loại: 006.42 Optical pattern recognition

Thông tin xuất bản: Nghiên cứu y học (Đại học Y Hà Nội), 2023

Mô tả vật lý: 144-151

Bộ sưu tập: Metadata

ID: 502256

Thêm vào giỏ

Đánh giá độ chính xác của thuật toán AI trong phát hiện tổn thương viêm thực quản trào ngược, khảo sát yếu tố liên quan đến việc bỏ sót, nhận nhầm. Phương pháp nghiên cứu: Mô tả cắt ngang. Thuật toán được kiểm chứng trên tập ảnh tĩnh bao gồm 1000 ảnh với các chế độ ánh sáng bằng cách so sánh phần khoanh vùng chuẩn của chuyên gia. Độ chính xác được đánh giá bằng độ nhạy, độ đặc hiệu, giá trị dự báo dương tính, giá trị dự báo âm tính. So sánh các tỷ lệ - xét mối liên quan được sử dụng để khảo sát những yếu tố liên quan đến tỉ lệ bỏ sót và nhận nhầm. Kết quả: Độ chính xác là 81,7%. Số lượng và kích thước tổn thương liên quan đến tỉ lệ bỏ sót. Tổn thương kèm theo, độ sạch liên quan đến tỉ lệ nhận nhầm. Kết luận: Thuật toán Yolov8 có độ chính xác tốt, có tiềm năng phát triển theo hướng đồng kiểm với bác sĩ nội soi trong khi soi, hậu kiểm sau nội soi, tham gia đào tạo y khoa với dữ liệu lớn.The purpose of this study was to evaluate the accuracy of an AI algorithm in detecting reflux esophagitis and to investigate factors related to missed lesions and errors. Methods: A crosssectional descriptive study was conducted. The algorithm was tested on a dataset consisting of 1000 endoscopic images with various lighting modes, and the results were compared to the standard expert labeling. Accuracy was assessed using sensitivity, specificity, positive predictive value, and negative predictive value. Regression models were employed to investigate factors influencing the rate of missed and misdiagnosed lesions. Result: The algorithm achieved an accuracy rate of 81.7%. The analysis revealed that the number and size of lesions in the images were significantly associated with the rate of missed detections, while accompanying damage and image cleanliness levels were related to misdiagnosed lesions. Conclusion: The Yolo v8 algorithm demonstrates high accuracy and holds potential for further development, including real-time co-assistance during endoscopy, postendoscopic examination, and medical training using large datasets.

1. Đọc trực tuyến

1. Đọc trực tuyến

Tạo bộ sưu tập với mã QR