Genetic algorithms automate breast ROI extraction in thermography

A new paper on arXiv (arXiv:2605.22899) presents a fully automated method for Region of Interest (ROI) extraction in thermographic breast images using Genetic Algorithms (GA). The authors—LC Mendes, EO Rodrigues, Sandro C Izidoro, Aura Conci, and Panos Liatsis—leverage color information and a cardioid-shaped fitness function to distinguish breast tissue from the background. Unlike prior techniques that require manual seed point selection, this GA-based approach runs autonomously, processing each image by evolving candidate ROIs over generations. The method successfully isolated the breast region in 52 out of 58 test images, achieving an 89.6% success rate. This work is the first in the literature to combine GA with cardioids for this specific application.

By automating ROI extraction, the method reduces human variability and could standardize acquisition protocols for thermographic breast cancer screening. Accurate ROI extraction is critical because it directly impacts subsequent analysis—false edges or background inclusion can skew thermal readings and lead to misdiagnosis. The paper, originally presented at IWSSIP 2020, demonstrates that evolutionary computation can reliably handle the challenging task of segmenting irregular breast boundaries in infrared images. For clinical adoption, further validation on larger datasets and comparison with deep learning approaches would be valuable, but this purely algorithmic solution offers a lightweight, interpretable alternative that requires no training data.