Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Influence of thermal contrast during dynamic thermography on a deep-learning-based estimation of breast tumour parameters
Submission Author:
Mateus Felipe Benicio Moraes , MA
Co-Authors:
Mateus Felipe Benicio Moraes, Tarcio Cardoso, Alisson Figueiredo
Presenter: Mateus Felipe Benicio Moraes
doi://10.26678/ABCM.COBEM2023.COB2023-1244
Abstract
Thermography is a low-cost and non-invasive imaging method that uses an infrared camera to detect and locate tumours by observing temperature anomalies on the breast skin surface. These anomalies can be caused by a tumour's high metabolic heat generation and blood perfusion rate. This technique can be used as an adjoint method to mammography in diagnosing tumours in dense breasts, common in women younger than 40 years. A parameter that influences the accuracy of the diagnosis is the thermal contrast since a decrease in its value makes it harder to observe the temperature anomalies. The purpose of this study is to investigate the thermal contrast’s impact on the prediction errors of a breast tumour’s size and location. A 3D hemispherical breast model with different tissue layers was built to compute the steady state and transient surface temperature profiles. A neural network was used to solve the inverse heat problem and estimate the size and location of the tumours. In those simulations, the location and size of the tumours were changed, with the 625 generated temperature profiles being used to train a neural network developed on MATLAB. The neural network was trained with surface temperature profiles obtained under passive and dynamic thermography conditions. A comparison between the estimates provided by each type of thermography was made. The estimates exhibited high correlation coefficients and strong linear relationships with the real values. An increase in thermal contrast improved the accuracy of all estimated parameters due to higher surface temperature variations. Deep-seated tumours showed no significant estimation error reduction in both types of thermography. The best estimates for parameters x, y and z were obtained by using dynamic thermography. The estimated radius showed the same error for both types of thermography, indicating that the increase in surface temperature variations was small in this case.
Keywords
breast cancer, Inverse Heat Condution Problem, Thermography, Deep learning
