ABSTRACT

A major concern for women’s health in today’s age is breast cancer. Thermography is an upcoming technology that is painless, private and relatively cheap to screen breast health. The presence of asymmetric hot blood vessel patterns in the breast thermogram portrays an abnormality. Proper extraction of these hotspots from the breast can help build a reliable breast cancer detection system and play a critical role in knowing the extent of spread of the cancer. In this work, segmentation of mammary thermograms is performed to extract the hottest blood vessel patterns using five state-of-the-art image segmentation methods. The proposed work is tested on the benchmark breast thermogram public dataset available at the Visual Lab. The most vascularised areas of each breast are extracted, and their areas are matched with the patches in the ground truth images. Based on metrics like DICE similarity coefficient and Jaccard index, it is concluded that particle swarm optimisation (PSO) algorithm and multi-seed region-growing technique provide the best segmentation results that are closer to the ground truth images. This indicates that infrared imaging is a promising tool that can act as a catalyst in predicting breast anomalies.

摘要

在当今时代，女性健康的一个主要问题是乳腺癌。热成像是一项即将到来的技术，它可以无痛、私密且相对便宜，可用于筛查乳房健康。乳房热像图中存在不对称的热血管模式描绘了异常。从乳房中正确提取这些热点有助于建立可靠的乳腺癌检测系统，并在了解癌症扩散程度方面发挥关键作用。在这项工作中，使用五种最先进的图像分割方法对乳腺热图进行分割，以提取最热的血管模式。拟议的工作在 Visual Lab 提供的基准乳房热图公共数据集上进行了测试。提取每个乳房血管最多的区域，并且它们的区域与地面实况图像中的补丁相匹配。基于 DICE 相似系数和 Jaccard 指数等指标，得出粒子群优化 (PSO) 算法和多种子区域增长技术提供的最佳分割结果更接近地面真实图像。这表明红外成像是一种很有前途的工具，可以作为预测乳房异常的催化剂。