Optimizing and Visualizing Deep Learning for Benign/Malignant Classification in Breast Tumors

TL;DR

This study demonstrates that deep learning, specifically CNN ensembles, can effectively classify benign and malignant breast tumors from mammograms, outperforming traditional methods and providing interpretability aligned with clinical features.

Contribution

The paper introduces a robust CNN ensemble approach for mammogram classification and a novel visualization method linking neural features to radiological markers.

Findings

Achieved 85% accuracy and 0.91 ROC AUC with CNNs.

Ensemble of GoogLeNet architectures improved performance.

Visualization correlates neural features with clinical markers.

Abstract

Breast cancer has the highest incidence and second highest mortality rate for women in the US. Our study aims to utilize deep learning for benign/malignant classification of mammogram tumors using a subset of cases from the Digital Database of Screening Mammography (DDSM). Though it was a small dataset from the view of Deep Learning (about 1000 patients), we show that currently state of the art architectures of deep learning can find a robust signal, even when trained from scratch. Using convolutional neural networks (CNNs), we are able to achieve an accuracy of 85% and an ROC AUC of 0.91, while leading hand-crafted feature based methods are only able to achieve an accuracy of 71%. We investigate an amalgamation of architectures to show that our best result is reached with an ensemble of the lightweight GoogLe Nets tasked with interpreting both the coronal caudal view and the mediolateral oblique view, simply averaging the probability scores of both views to make the final prediction. In addition, we have created a novel method to visualize what features the neural network detects for the benign/malignant classification, and have correlated those features with well known radiological features, such as spiculation. Our algorithm significantly improves existing classification methods for mammography lesions and identifies features that correlate with established clinical markers.