Noninvasive Diagnosis of the Type of Breast Tumor through Artificial Neural Networks

Abstract

Different changes such as developing benign and malignant lesions in tissues lead to specific variations in their macroscopic and microscopic structure, which are associated with the alteration of their mechanical properties. In the present study, the mechanical parameters of different breast tissue lesions were noninvasively estimated with high precision based on the displacement data by using the powerful neural network method in order to detect the type of tumor in the breast tissue. The displacement data of various breast tissues, as well as the corresponding mechanical properties were acquired to develop and train the neural network models. For simulating breast tissue behavior and extracting the relevant displacement data to train the neural networks, the finite element modeling was applied using Abaqus software. Ogden and Yeoh hyperelastic models which are precise for expressing the hyperelastic behavior of soft tissues, specifically the breast, were used to create the finite element model for tumor-containing breast tissue. With the aim of obtaining a robust neural network model, the displacement data extracted from the finite element model and white noise summated to simulate laboratory conditions while deriving tissue data from finite element model. As indicated by the results, the trained neural network models represent high precision and efficiency in appraising the mechanical parameters of various breast tissues according to the displacement data, which promises its use for carefully diagnosing the type of breast lesion.