View article

Infrared thermography is gaining attention as an affordable, portable, privacy-sensitive, and radiation-free breast imaging modality. However, thermal images alone provide only two-dimensional (2D) information necessitates multiple images to visualize the complete 180° breast region, posing challenges in visually correlating these images. To address this issue, we propose 3D-BreastNet, a deep-learning architecture that reconstructs 3D thermal images from 2D thermal images. The proposed method employs a self-supervised learning strategy to learn the 3D silhouette and overlay observed temperatures onto the predicted 3D structure. To evaluate the performance of 3D-BreastNet, we compared the 2D projections of the predicted 3D breast silhouette at standard view angles with their corresponding input 2D silhouettes. The accuracy, dice index, jaccard index, and hausdorff distance were found to be 0.97, 0.97 …