Enhanced deep learning model for diagnosing breast cancer using thermal images

Breast cancer has emerged as one of the most prevalent malignancies affecting women today, underscoring the critical need for advanced diagnostic tools. Mammography has been the conventional method for early breast cancer detection, yet recent years have witnessed the emergence of thermal infrared scans, or thermographies, as a potential contender for diagnosing breast cancer, particularly in cases of dense breast tissue. Thermographic images reveal higher temperatures in regions containing tumors compared to healthy breast tissue, providing a promising avenue for early diagnosis. In parallel, the field of radiography has seen the advent of deep learning (DL) techniques, offering a computational approach to breast cancer identification. This study presents a novel approach, the Enhanced Deep learning-based Convolutional Neural Network (EDCNN), aimed at generating heatmaps from two-dimensional thermal breast images. These heatmaps are employed to quantitatively assess breast vascularity, yielding interpretable parameters for further analysis. In addition, the study proposes a classifier that predicts the likelihood of breast cancer purely based on these extracted parameters. To enhance the accuracy of this process, the algorithm combines Fuzzy C-means clustering with the Region of Interest (ROI) technique to effectively isolate the breast from surrounding body parts. The segmentation results are evaluated using temperature profiles, revealing substantial peaks in the patterns as indicators of ROIs. This identification of hot areas hints at the potential presence of a tumor. To validate the effectiveness of this approach, the study constructs DL models using convolutional neural networks, training them with thermal breast images from the Graphical DMR datasets. The results are compelling, with the EDCNN models outperforming alternative methods, achieving an impressive accuracy of 96.8% and a specificity rate of 93.7%. This research thus offers a robust, efficient, and reliable means of early breast cancer diagnosis, marking a significant advancement in the field. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.