Title: Harnessing the power of machine learning for breast anomaly prediction using thermograms

Authors: Aayesha Hakim; R.N. Awale

Addresses: Veermata Jijabai Technological Institute, H.R. Mahajani Marg, Matunga, Mumbai, Maharashtra 400019, India ' Veermata Jijabai Technological Institute, H.R. Mahajani Marg, Matunga, Mumbai, Maharashtra 400019, India

Abstract: Breast cancer is the most fatal cancer among women globally. Thermography provides an early sign of a developing abnormality based on the temperature changes in breasts. In this work, statistical features extracted from the segmented breast region are used for breast cancer prognosis. Machine learning algorithms like support vector machine (SVM), k-nearest neighbourhood (kNN), naïve Bayes and logistic regression without and with principal component analysis (PCA) as a pre-cursor are applied to the extracted data to classify thermograms as malignant or benign. Classification was also performed using tree-based classifiers, namely, decision tree and random forest. This work indicates that thermal imaging is capable of predicting breast pathologies coupled with machine learning algorithms. The PCA-SVM model has the best accuracy, sensitivity, specificity and AUROC of 92.74%, 77.77%, 95.83% and 0.8699 respectively. Among tree-based classifiers, random forest classifier has the best accuracy, sensitivity, specificity and AUROC of 94.4%, 97.5%, 78.72% and 0.97961 respectively with five-fold cross validation. Our study produced competitive results when compared to other studies in the literature.

Keywords: breast cancer; breast thermography; infrared imaging; thermal imaging; machine learning.

DOI: 10.1504/IJMEI.2023.127250

International Journal of Medical Engineering and Informatics, 2023 Vol.15 No.1, pp.1 - 22

Received: 12 Aug 2020
Accepted: 15 Jan 2021
Published online: 30 Nov 2022 *