研究目的
To compare two techniques for segmenting breast tumors from thermal images: marker-controlled level-set segmentation with anisotropic diffusion filtering versus with Gaussian filtering, aiming to improve computer-aided diagnosis of breast cancer by enhancing edge preservation and segmentation accuracy.
研究成果
The modified marker-controlled level-set segmentation, when combined with anisotropic diffusion filtering, outperforms Gaussian filtering in preserving edges and achieving higher accuracy, as evidenced by higher Jaccard index and Tanimoto values. This combination is more suitable for subsequent feature extraction in automated breast cancer diagnosis, offering a promising approach for low-cost, radiation-free screening.
研究不足
The study is limited to preprocessing and segmentation stages; it does not cover feature extraction or classification for full computer-aided diagnosis. The database used may have constraints in sample diversity or size. The marker placement might require manual adjustment, potentially introducing subjectivity. The techniques are evaluated only on thermal images, and their generalizability to other imaging modalities is not addressed.
1：Experimental Design and Method Selection:
The study compares two combinations of preprocessing and segmentation techniques: Gaussian filtering with marker-controlled level-set and anisotropic diffusion filtering with marker-controlled level-set. The rationale is to evaluate which preprocessing method better preserves edges for accurate segmentation in thermal images. The level-set method is chosen for its ability to handle complex shapes and edges, with a modified marker-controlled approach to automate initial contour placement.
2：Sample Selection and Data Sources:
Images were collected from an online database 'proeng' [9], which includes samples from women with varying breast characteristics. The selection criteria are not specified beyond using this database for the experiments.
3：List of Experimental Equipment and Materials:
The primary tool used is MATLAB 2017a for image processing and segmentation. No specific hardware or other materials are mentioned.
4：Experimental Procedures and Operational Workflow:
For each technique, the process involves: preprocessing the thermal image with either Gaussian or anisotropic diffusion filtering to remove noise, then applying marker-controlled level-set segmentation. Markers are placed on the left and right breasts as initial contours, and the level-set function evolves to segment the breast region. The workflow includes steps such as initial contour superimposition, evolution, and final segmentation.
5：Data Analysis Methods:
Accuracy is evaluated using Jaccard index and Tanimoto coefficient to measure similarity between segmented results and ground truth images. These statistical measures are computed to compare the effectiveness of the two techniques.
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
