Using Deep Neural Network Approach for Multiple-Class Assessment of Digital Mammography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Breast cancer affects over ten thousand women annually in Taiwan.
• Mammography is a key diagnostic tool but suffers from operator variability and subjective interpretation, leading to inconsistent results.
• There is a need for objective and reliable methods for mammography image analysis.

Purpose of the Study:

• To develop and evaluate a deep neural network algorithm for classifying mammography images.
• To improve the consistency and accuracy of breast cancer detection using artificial intelligence.
• To explore the effectiveness of a fully convolutional dense connection network (FC-DCN) for this task.

Main Methods:

• A retrospective study collected clinical mammography images.
• Images were classified using the Breast Imaging Reporting and Data-Analyzing System (BI-RADS).
• A fully convolutional dense connection network (FC-DCN) was employed, utilizing image preprocessing, data augmentation, and transfer learning.

Main Results:

• The developed model achieved an accuracy of 86.37%.
• Sensitivity was recorded at 100%, and specificity at 72.73%.
• The FC-DCN framework effectively reduced training parameters while yielding a reasonable classification model.

Conclusions:

• The deep neural network, based on the FC-DCN model, demonstrates significant potential for accurate mammography image classification.
• This AI-driven approach can help overcome the limitations of traditional mammography interpretation.
• The study highlights the feasibility of using advanced deep learning techniques for improving breast cancer diagnosis.