Scholarlink

Breast cancer detection via mammograms is enhanced by identifying microcalcifications. This study develops a Convolutional Neural Network (CNN) for detecting and localizing microcalcifications in digital mammograms. Using mammograms from VinDr-Mammo (844 images), INbreast (334 images), and MIAS (106 images), images were preprocessed to improve contrast and remove artifacts. The CNN architecture includes three convolutional layers with Prewitt, GLCM, and Gabor filters, followed by ReLU activation and max-pooling layers. The model was trained with the Adam optimizer and binary cross-entropy loss function. The model achieved 73.00% accuracy, 70.17% precision, 71.42% sensitivity, and 66.66% specificity in classifying mammograms. For localization, the system reached 73.00% accuracy, 71.92% precision, 78.84% sensitivity, and 66.66% specificity. Despite a 26.00% rate of false negatives in classification, the CNN effectively detected microcalcifications down to 1 mm. Further optimization could improve reliability. In conclusion, this CNN-based system shows potential for early detection of microcalcifications, critical for timely breast cancer diagnosis.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Mammogram Analysis for Microcalcification Identification Using Convolutional Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • A. Hernández-Vázquez,
  • Y. Hernández-Rodríguez,
  • F. Cortes-Rojas,
  • R. Bayareh-Mancilla,
  • O. Cigarroa-Mayorga

摘要

Breast cancer detection via mammograms is enhanced by identifying microcalcifications. This study develops a Convolutional Neural Network (CNN) for detecting and localizing microcalcifications in digital mammograms. Using mammograms from VinDr-Mammo (844 images), INbreast (334 images), and MIAS (106 images), images were preprocessed to improve contrast and remove artifacts. The CNN architecture includes three convolutional layers with Prewitt, GLCM, and Gabor filters, followed by ReLU activation and max-pooling layers. The model was trained with the Adam optimizer and binary cross-entropy loss function. The model achieved 73.00% accuracy, 70.17% precision, 71.42% sensitivity, and 66.66% specificity in classifying mammograms. For localization, the system reached 73.00% accuracy, 71.92% precision, 78.84% sensitivity, and 66.66% specificity. Despite a 26.00% rate of false negatives in classification, the CNN effectively detected microcalcifications down to 1 mm. Further optimization could improve reliability. In conclusion, this CNN-based system shows potential for early detection of microcalcifications, critical for timely breast cancer diagnosis.