Faster Region-Based Convolutional Neural Networks for Tumor Localization in Breast Thermograms: A Novel Approach

Overview

One in eight women in the United States will be diagnosed with breast cancer in her lifetime. Early diagnosis of breast cancer is key to effective treatment and an increased survival rate. Mammography, the current gold standard of diagnosis, exhibits a low sensitivity in younger women and women with dense breast tissue and has a high false positive rate. Alternate forms of screening include breast thermography, which does not require compression of the breast and is equally accurate regardless of age or tissue density. However, thermography has a much higher false positive rate and a much lower true positive rate than mammography. Computer Aided Diagnosis (CAD) systems can help reduce the false positive rate and increase the sensitivity of thermography while retaining its benefits over mammography. To date, there have been no CAD systems developed to identify tumors in breast thermograms. In this study, we aim to develop a CAD system to detect tumors in thermograms with a comparable sensitivity and specificity to that of radiologists. We develop a Faster Region-Based Convolutional Neural Network with a sensitivity of 0.906, a false positive rate of 0.232, and an AUC of 0.89. This system not only outperforms radiologists, but also is more sensitive than other algorithms designed to detect tumors in mammograms, ultrasound, or other modalities. This paper demonstrates that in conjunction with this CAD model, breast thermography can be used as a reliable method of screening and is a step forward in the field of computer aided diagnostics.

Awards (1)

• 1st Place Mathematics & Computer Science

Competition history

• JSHS 2020 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.