研究目的
Comparing feature-based CAD and CNN-based CAD for breast cancer classification in DBT images to determine which method provides better accuracy.
研究成果
The CNN-based CAD significantly outperformed the feature-based CAD in breast cancer classification from DBT images, with higher accuracy, precision, and recall. These results suggest that CNN-based CAD can be a valuable tool in clinical medicine for assisting radiologists in breast cancer identification. Future studies should evaluate the proposed strategy on more images and larger CNN architectures, as well as include other image modalities to enhance representation.
研究不足
The study was limited to a small dataset of 20 cases from a single institution, which may not represent the broader population. Additionally, the study focused on comparing two specific CAD methods without exploring other potential approaches or larger CNN architectures.
1：Experimental Design and Method Selection:
The study involved comparing two CAD systems for breast cancer classification in DBT images: feature-based CAD and CNN-based CAD. The methods included image preprocessing, candidate tumor identification, 3D feature generation, classification, image cropping, augmentation, CNN model design, and deep learning.
2：Sample Selection and Data Sources:
DBT images of 20 practical cases were obtained from TVGH in Taiwan, involving Asian females of varying ages.
3：List of Experimental Equipment and Materials:
The study utilized DBT images and developed CAD systems for analysis.
4：Experimental Procedures and Operational Workflow:
The process included reading DBT images, preprocessing (noise removal, intensity conversion, morphological treatment, breast ROI extraction), candidate tumor determination, feature generation for feature-based CAD, and CNN model application for CNN-based CAD.
5：Data Analysis Methods:
Performance was evaluated using accuracy, precision, and recall, with statistical analysis to compare the two CAD methods.
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
