ABSTRACT
A recent study from GLOBOCAN disclosed that during 2018 two million women worldwide had been diagnosed with breast cancer. Currently, mammography, magnetic resonance imaging, ultrasound, and biopsies are the main screening techniques, which require either, expensive devices or personal qualified; but some countries still lack access due to economic, social, or cultural issues. As an alternative diagnosis methodology for breast cancer, this study presents a computer-aided diagnosis system based on convolutional neural networks (CNN) using thermal images. We demonstrate that CNNs are faster, reliable and robust when compared with different techniques. We study the influence of data pre-processing, data augmentation and database size on several CAD models. Among the 57 patients database, our CNN models obtained a higher accuracy (92%) and F1-score (92%) that outperforms several state-of-the-art architectures such as ResNet50, SeResNet50, and Inception. This study exhibits that a CAD system that implements data-augmentation techniques reach identical performance metrics in comparison with a system that uses a bigger database (up to 33%) but without data-augmentation. Finally, this study proposes a computer-aided system for breast cancer diagnosis but also, it stands as baseline research on the influence of data-augmentation and database size for breast cancer diagnosis from thermal images with CNNs
Acknowledgments
JZ, ZA, KB, SM and NZ contributed to conception and design. JZ led the data pre-processing and algorithm conception. JZ, ZA, KB and SM contributed to analysis and discussion of the results. JZ contributed with the preparation of the manuscript. JZ, ZA, KB, SM and NZ contributed with to the reviews of the manuscript. All authors read and approved the final manuscript.
Article highlights
Efficiency and reliability for breast cancer diagnosis through thermography
CNNs performance enhancement with data augmentation techniques
Smaller and simpler CNNs architectures perform better than complex CNNs
Trade-off measurement between data augmentation and database size
Disclosure statement
The authors have stated that they have no conflicts of interest.
Additional information
Funding
Notes on contributors
J. Zuluaga-Gomez
Juan Pablo Zuluaga is a PhD student in Automatic Speech Recognition and Artificial Intelligence at The École Polytechnique Fédérale de Lausanne and the Idiap Research Institute in Switzerland. His research is focused on state-of-the-art techniques for speech recognition in technical areas such as air-traffic communications. He is also a member of two students' associations, the driverless sub-team of the EPFL Racing Team and Asclepios.
Z. Al Masry
Zeina Al Masry is an Associate Professor at the École Nationale Supérieure de Mécanique et des Microtechniques in Besançon, France. She is doing her research activities at the FEMTO-ST institute in the Prognostics and Health Management (PHM) research group. Her research works concern stochastic processes and applied statistics for PHM applications in medical and industrial fields.
K. Benaggoune
Khaled Benaggoune is a Ph.D. student in industrial computing at Batna University, Algeria. He had a Master's degree in industrial engineering from the industrial computing department, Batna 2 University. His works focus on artificial intelligence applications in the industrial and medical fields.
S. Meraghni
Safa MERAGHNI is a PhD student in Artificial Intelligence field at LINFI laboratory in Biskra University , Algeria. She had an engineering degree in computer science in 2011 from Ecole Supérieure d'Informatique (ESI), Algiers and a Master degree in Artificial intelligence in 2015 at university of Biskra. She is working on the use of artificial intelligence and Information technologies in industry and medical fields.
N. Zerhouni
Noureddine Zerhouni is a Full Professor at the École Nationale Supérieure de Mécanique et des Microtechniques in Besançon, France. He is a member of the Department of Automatic Control and Micro- Mechatronic Systems at the FEMTO-ST Institute. His current research interests include intelligent maintenance, and prognostics and health management.