研究目的
To develop a fully automatic algorithm for segmenting esophagus layers from OCT images to address the time-consuming and subjective nature of manual labeling, using Fast Marching Method and Fourth-Order Runge-Kutta method.
研究成果
The proposed automatic segmentation algorithm achieves results comparable to manual segmentations, with mean absolute error thickness difference less than 6 pixels and Dice's similarity coefficient of 0.8587, indicating good accuracy and potential for clinical application, though further improvements with machine learning are suggested for future work.
研究不足
Accurate region positioning requires prior knowledge, and if the organizational structure is too tight, segmentation may fail due to errors in area restriction.
1：Experimental Design and Method Selection:
The algorithm uses Fast Marching Method (FMM) to calculate weighted geodesic distance with a velocity function combining vertical gradient, horizontal gradient, and curvature, and Fourth-Order Runge-Kutta method (RK4) to find the shortest path for boundary segmentation. Preprocessing includes median filtering for noise reduction, image enhancement for contrast improvement, and adaptive search area limitation.
2：Sample Selection and Data Sources:
400 healthy guinea pig esophagus B-scan OCT images are used, obtained from prior studies. Manual labelings are created using ITK-SNAP software by two observers.
3：List of Experimental Equipment and Materials:
OCT imaging system (specific model not mentioned), computer for processing, ITK-SNAP software for manual segmentation.
4：Experimental Procedures and Operational Workflow:
Steps include noise removal with median filter, plastic sheath removal, image flattening, contrast enhancement, boundary segmentation in a specific order (B1, B3, B2, B4, B6, B5) using FMM and RK4 with adaptive region limitation, and comparison with manual segmentations.
5：Data Analysis Methods:
Evaluation using Average Thickness (AT), Mean Absolute Error Thickness Difference (MAE), and Dice’s Similarity Coefficient (DSC) metrics.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
