Deep learning-based approach in surface thermography for inverse estimation of breast tumor size

Background and objective: In early breast cancer diagnosis, tumor size is key to improving the patient's survival chances. It helps doctors to determine the adequate treatment for each case. Surface thermography presents encouraging results to detect thermal patterns of breast tumor abnormality. However, the early and accurate estimation of tumor size based on temperature characteristics is quite challenging due to unavailability of labeled clinical data. This work proposes a Feed-Forward Deep Neural Network (FF-DNN) for an inverse estimation of breast tumor size using thermographic data. Methods: A 3D breast model was created by the COMSOL Multiphysics software incorporating tumors in the gland and covered by fat, muscle, and skin layers. Several tumor configurations are included to generate a large amount of training data. An initial thermal data analysis was performed to affirm the influence of breast tumor size on the skin surface temperature. Then, 1400 different cases were prepared, and the relevant features were extracted to train the deep learning model. The coefficient of determination (R2) and the Mean Square Error (MSE) were used as metrics for evaluating the predictive model. Results: The analysis of the normalized temperature variations demonstrated the influence of tumor size on the surface temperature of the breast. Thus, the comparison between the FF-DNN against the Convolutional Neural Network (CNN) showed the reliability of the proposed approach. As result, the prediction accuracy indicates the capacity of the proposed FF-DNN model to estimate tumor size from the provided relevant features with an MSE value of 0.194 and an R2 value of 0.998. Conclusion: The findings of this study indicate that surface thermography, when paired with deep learning, holds promise as a valuable diagnostic tool to improve the prognosis of breast cancer.