Intelligent diagnosis of breast cancer using IR thermography, thermal modeling and AI

Abstract

There is a high chance of developing breast cancer in women and it remains to be one of the prevalent diseases from which women suffer. However, early detection and appropriate measures increase the chance of surviving. Specialists can overlook 10-30% of signs of breast cancer in mammograms. In this paper we suggest using machine learning (ML) algorithms with deep learning, namely, Bayesian Network (BN). In this paper 2 BN models designed to predict breast cancer based on medical information and thermography are presented. The data from "Multifunctional Medical Center" of the Nur-Sultan city were obtained, which includes 40 patients. The models are trained through Tree Augmented Naive Bayes algorithm, and robustness is verified through k-folds method. The models show high performance despite small sample size, with low amount of misdiagnosed cases. Overall accuracy for the two models is 85% and 95%. Mean temperature and standard deviation were found to be main predictors, as well as a predictor variable obtained from a CNN model. Future works are aimed at enlarging the dataset and improving the balance between the number of positive and negative diagnoses to improve the model's performance.