研究目的
To determine the appropriate ROI size and location to classify fatty and dense mammograms for early breast cancer detection using computer-aided diagnosis systems.
研究成果
The study concludes that an ROI size of 200×200 pixels from the central breast region provides the highest classification accuracy (96.1%) for distinguishing between dense and fatty mammograms. The use of texture features and feature selection with FDR enhances performance, reducing computation time. This approach aids in early breast cancer detection by efficiently classifying breast tissue density, with implications for improving CAD systems.
研究不足
The study is limited to the MIAS database, which may not represent all mammographic variations. Only linear SVM is used for classification; other classifiers or non-linear methods could be explored. ROI sizes are fixed and manually selected; automated or adaptive sizing might improve results. The research focuses on texture features; other feature types could be considered.
1：Experimental Design and Method Selection:
The study uses a multi-size ROI analysis technique with different ROI sizes (256×256, 200×200, 175×175, 150×150, 128×128, 64×64 pixels) extracted from the central region of mammograms. Texture features (Haralick, Fractal, Fourier Power Spectrum) are extracted, and Fisher's Discriminant Ratio (FDR) is used for feature selection. A linear SVM classifier is employed for classification.
2：Sample Selection and Data Sources:
Digital mammograms from the MIAS (Mammographic Image Analysis Society) database are used, specifically fatty and dense glandular mammograms. The database includes 322 mammograms of size 1024×1024 pixels.
3：List of Experimental Equipment and Materials:
Software tools include Matlab and Matrox imaging for implementation. No specific hardware is mentioned.
4：Experimental Procedures and Operational Workflow:
ROIs are extracted, pre-processed with 2D median filtering and unsharp masking for noise reduction and edge enhancement. Texture features are calculated, features are selected using FDR, and classification is performed with SVM. Performance is evaluated using sensitivity, specificity, and accuracy metrics.
5：Data Analysis Methods:
Statistical analysis involves calculating sensitivity, specificity, and accuracy from classification results. The SVM classifier is used with linear kernel.
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
