Breast Lesion Classification From Digital Breast Tomosynthesis Images Using Convolutional Neural Networks

Jose Zarate      Kevin Xu      Joshua Tennyson

jhz7@duke.edu     kx33@duke.edu     joshua.tennyson@duke.edu

Paper

k

Breast cancer is a leading cause of cancer-related deaths among women worldwide. Early detection through digital breast tomosynthesis (DBT) screening, a 3D mammography technology, is crucial for improving patient outcomes. We developed 3 machine learning models based on convolutional neural networks with varying architecture and degrees of complexity to classify DBT images as normal, benign, or cancerous. Our “Original CNN” model performed the best and was able to accurately detect cancerous lesions around 50% accuracy even for data sets with a low resolution physical layer applied to it.

Paper: Paper

Code and Data: Codebase Image Database