PROJECT TITLE :

Deep Neural Networks Improve Radiologists Performance in Breast Cancer Screening

ABSTRACT:

To classify mammograms for breast cancer screening, we developed a deep Convolutional Neural Network that was trained and assessed on over 2,00000 tests (over 1000000 images). On the screening population, our network gets an AUC of 0.895 in predicting breast cancer. The great level of precision can be attributed to a few technological advancements. Using a high-capacity patch-level network to learn from pixel-level labels while simultaneously using a network to learn from breast-level labels, our network's revolutionary two-stage architecture and training technique. Our model is built around an unique ResNet-based network that is designed for high-resolution medical images in terms of depth and width. To prepare the network for this duty, it is necessary to train it on the screening of the BI-RADS categorization. 4) Selecting the best combination of input views from a variety of options. A reader study with 14 readers, each reading 720 screening mammography tests, proved that our algorithm is as accurate as experienced radiologists when confronted with the same data.. Both radiologists and our neural network can accurately identify malignancy, but when combined, our hybrid model is more accurate than each of them alone. We conduct a thorough examination of our network's performance on different subpopulations of the screening population, the model's design, training technique, errors, and the features of its internal representations in order to better understand our results. https://github.com/nyukat/breast cancer classifier is where you can find our best models