ARTIFICIAL NEURAL NETWORK FOR BREAST CANCER DETECTION IN SCREENING MAMMOGRAPHY

Breast cancer is the leading cause of cancer death in women in Brazil and has a high survival rate when detected early. Mammography is the only exam whose application in screening programs has proven efficacy in reducing mortality from this pathology. Artificial intelligence has shown significant potential to aid in imaging diagnosis, especially the technique of convolutional neural networks (CNN), a deep learning model that receives an input image, assigns importance to it – learnable parameters and biases – to various elements/aspects in the image to be able to differentiate one from the other. This study aims to present a machine learning system capable of analyzing mammography images to identify potential positive breast cancer cases to supplement human-AI cooperation and improve diagnostic accuracy. The method for developing the classifier is based on the use of a CNN model with supervised training. In this study, three publicly available datasets of mammography images were used: CBIS-DDSM, INbreast and MIAS. To evaluate the proposed framework, a 50:50 approach was chosen, which means that 50% of the images from each dataset are consumed for training and the rest for testing. The programming language used to develop the algorithm was Python, with TensorFlow and Keras libraries, both version 2.6, and the Keras Tuner API, version 1.0.4. The model reached scores of 96.07% specificity and 94.73% sensitivity, hitting 968 positive and 926 negative images, among the 2000 constituents of the test dataset. The results are consistent with the literature and demonstrate the great potential of the system to assist radiologists in differentiating benign and malignant findings in mammograms and may also reduce invasive diagnostic procedures in benign lesions and assist in the early detection of breast cancer. This model can be implemented in places where there is a shortage of subspecialized breast radiologists, supplementing health promotion and prevention for those who lack qualified care.