Chapter 9 - Scrutinization of mammogram images using deep learning

Abstract

Progressive steps were taken by the deep learning model in a multitude of implementations. The main source of interest is the convolutional neural network (CNN), which is regarded as a prominent potent method for discovering incentive of images and other semantic information. Datasets include images taken using a innumerable imaging modalities, including MRIs, scans—CT and PET (Positron Emission Tomography), X-rays, ultrasounds, fluorescein angiography, and even photographs. This study examines several deep learning method design patterns and partially centers on the CNN, which provides an exceptional percentage of solutions when compared to other DNN (Deep Neural Network) architectures. CNN prioritizes image features and uses recognizable architectural styles. However, restrictions that go beyond DNN training and execution time will be discussed. Finally, a comparison of current software frameworks for deep learning method will be presented, along with future research prospects. The cornerstone of this exploration is to look at approaches to implement a deep learning model that can accurately identify breast cancer so that late medications can be mitigated because of inaccurate negatives and unnecessary medications can be avoided owing to wrongful convictions. A dataset of 1312 photos was used, which included photographs of both benign and malignant kinds. The models ResNet50, DenseNet201, AlexNet, and VGG16 are used. Without any transfer learning, we attain the best accuracy for our basic model. Augmentation is a technique for rescaling data and reducing overfitting. When validated on a well-known globally available set of data, the existing technique yields better output. Existing models are also helpful in medical image analysis, as they execute exactly with a low loss function.