Thermography based deep learning models for early breast cancer detection

Abstract

Breast cancer is one of the most common causes of death in women around the world. Researchers are actively seeking to develop early detection methods for breast cancer. Several treatment technologies contributed to the reduction in mortality rate from this disease, but early detection contributes the most to preventing disease spread, breast amputation and death. The problem, however, lies in the accuracy of early detection methods. Thermography is a promising technology for early diagnosis where thermal cameras employed are of high resolution and sensitivity. The combination of Artificial Intelligence (AI) with thermal images is an effective tool to detect early-stage breast cancer and is foreseen to provide impressive predictability levels. This thesis reviewed systematically the state-of-the art works employing thermography with AI, highlighted their contributions and drawbacks, and proposed open issues for research. Furthermore, the thesis has applied and investigated the behaviour of different recently introduced deep learning methods for identifying breast disorders and further proposed a modified method to suit the thesis goals. Inception MV4 achieved 7% faster classification response time compared to V4. The use of MV4 model is found to contribute to saving energy consumed and fluidity in arithmetic operations for the graphic processor. The results also indicate that increasing the number of layers may not necessarily be useful in improving the performance. Furthermore, the thesis develops a numerical simulation model to study the thermophysical properties of breast using COMSOL software. Topical Sito-Cooling on breast surface area was found to contribute to increasing thermal contrast in the simulated thermal images. The highest variations in skin temperatures between breasts with cancer and without cancers can scope from 0.274 to 2.58 C. Finally, the thesis introduced an application design in a graphical user interface and linked it with the AirDroid application to send thermal images from the smartphone to the cloud and then retrieve back the diagnostic result from the cloud to the smartphone app. The suggested framework novelty lies in its design to generate high-quality input video of thermal imagery of the patients’ breast region in real time, facilitating more accurate early breast cancer detection. The suggested structure was modelled in MATLAB 2019 and was compatible with majority of standard Desktop with thermal camera installed. It takes real time video stream of high-quality thermal imagery as input and produces defined video files with a binary classification characterizing normal or abnormal breasts with a recommended action for the patient. This is followed by a proposed thermal image acquisition procedure with set of recommendations for the development of a mobile app-based dataset. The thesis concludes that early breast cancer detection using smart apps is a valuable and reliable complementary tool for radiologists to aid the diagnosis process and reduce mortality rates.