กระบวนการแยกส่วนประกอบของปอดจากภาพ Chest CT โดยอัตโนมัติ ด้วยวิธีการเรียนรู้เชิงลึก

Abstract

Lung cancer, characterized by uncontrollable cell division, stands as the leading cause of cancer-related deaths globally. The primary treatment approach typically focuses on radiation therapy. However, accessing this therapy requires radiation treatment planning, wherein radiation oncologists manually delineate tumor and Organ at Risk (OAR) components, notably the lungs, leading to time-intensive processes for each patient. This research aims to reduce time in treatment planning process by applying artificial intelligence technology to segment lung components alongside existing neural network architectures, including FCN, U-Net, and DeepLabV3+. The study utilizes image data from the Lung CT Segmentation Challenge 2017 (LCTSC) dataset, comprising Horizontal CT images from 60 patients, totaling 5,458 images. The workflow consists of three stages: data preparation, model training across all three mentioned model architectures, and model testing and performance evaluation based on Dice similarity coefficient (DSC) measurements. The results show that U-Net achieved the highest average DSC of 87.90%, ranging from 71.40% to 93.45% across different lung regions. This advancement promises to significantly expedite treatment planning processes, potentially enhancing patient care and outcomes.