Comparison of Deep Learning Architectures For Pre-Screening of Breast Cancer Thermograms

被引：22

作者：

Carlos Torres-Galvan, Juan ^{[1
,2
]}

Guevara, Edgar ^{[1
,2
,3
]}

Javier Gonzalez, Francisco ^{[1
,2
]}

机构：

[1] Univ Autonoma San Luis Potosi, Terahertz Sci & Technol Ctr C2T2, San Luis Potosi, San Luis Potosi, Mexico

[2] Univ Autonoma San Luis Potosi, Sci & Technol Natl Lab LANCyTT, San Luis Potosi, San Luis Potosi, Mexico

[3] Univ Autonoma San Luis Potosi, CONACYT, San Luis Potosi, San Luis Potosi, Mexico

来源：

2019 PHOTONICS NORTH (PN) | 2019年

关键词：

Deep Learning; Thermography; Breast Cancer;

D O I：

10.1109/pn.2019.8819587

中图分类号：

TB3 [工程材料学];

学科分类号：

0805 ; 080502 ;

摘要：

Infrared thermography can be used for pre-screening of breast cancer but the results of this technique depend on the experience of the human expert. We propose an automated analysis approach to assess the capabilities of deep neural networks to classify breast thermograms. The dataset consisted of 173 images and we compared seven deep learning architectures. VGG-16 convolutional neural network outperformed with a sensitivity of 100%, specificity of 82.35% and balanced accuracy of 91.18%. Such results indicate that deep neural networks can be used in the analysis of thermal images for breast cancer pre-screening.

引用

页数：2