Breast Cancer Diagnosis Based on Feature Extraction Using Dynamic Models of Thermal Imaging and Deep Autoencoder Neural Networks
Hossein Ghayoumi Zadeh, Ali Fayazi, Bita Binazir, Mostafa Yargholi
Abstract
Abstract Breast cancer is the most commonly diagnosed cancer among women worldwide. Diagnosing breast cancer at its earliest stages increases the chance for treatment. Thermography offers a method to consider screening strategies for this type of cancer. Analysis of breast thermography images is very effective to extract local information that helps to identify abnormalities related to the breast region. In this paper, first, a semiautomatic approach is proposed for separating the breast region according to the different morphological structures of the breast tissue. The stages of breast segmentation are then performed based on the classification of the presented morphologies, which used a series of dynamic images of a patient. In the following, feature extraction is performed based on the proposed model. The eight statistical characteristics are obtained from a series of separated images of a person’s breasts. The autoencoder neural network, which is an unsupervised deep-learning algorithm, is used to classify the thermography images as healthy and unhealthy. To analyze the proposed model, the Database For Mastology Research is used. The number of people surveyed was 196, including 41 cases of cancer and 155 healthy cases. Each person had 10 thermography pictures, and the total number of analyzed images was 1,960. The accuracy and specificity of the proposed method were 94.87 and 96.77 %, respectively. The simulation results show that the proposed model is able to provide a significant response to different morphologies of the breast tissue.