Early Breast Cancer Screening from Thermography via Deep Pre-Trained Edge Detection with Extreme Gradient Boosting

Abstract

Breast cancer is the most common type of cancer affecting women worldwide. Early diagnosis is crucial for successful treatment, recovery, and mortality reduction. Therefore, various non-invasive, minimally invasive, and invasive approaches have been developed for breast cancer detection. Breast thermography is a promising non-invasive, low-cost, and radiation-free imaging technique that can provide information about the physiological variations of breast tissue. It can be considered as a supplement to X-ray mammography or even a primary early cancer detection tool, particularly for women with dense breast tissue or those who are not suitable candidates for mammography. This study proposes a fully automated system for thermography-based early breast cancer detection. The characteristic features of the breast are extracted based on the deep pre-trained holistically-nested edge detector. Such features are used to diagnostically classify breast tissue into normal or malignant classes using the machine learning algorithm extreme gradient boosting. The classification improved to some extent by incorporating the Canny edge detector. The developed system achieved an accuracy of 97.4%, with a precision of 95%, a recall of 100%, and an F1 score of 97%.