Detection of Breast Abnormality using Machine Learning and Deep Learning Methods

Cancer is regarded as the fatal disease worldwide especially breast cancer, a prevalent and potentially significant factor affecting mortality among women, that originates in the breast cells grows abnormally and invade the neighboring cells of the human body. The early diagnosis and management of cancer with suitable treatments would significantly improve the human survival rates. Among the various modalities available for cancer detection mammography is regarded as the most accepted and effective gold standard imaging modality for the detection of breast cancer. Current research explores the use of convolutional neural networks (CNNs) with Model Architecture (ResNet101) to enhance breast cancer detection processes, specifically targeting invasive ductal carcinoma regions within whole-slide images (WSIs). Further, several CNN architectures were evaluated for automated breast cancer identification and their performances are compared against traditional machine learning (ML) techniques. Our comprehensive dataset contains mammography with benign and malignant masses. This dataset basically consists of 2347 masses images being extracted from INbreast, MIAS, and DDSM dataset. Then data augmentation and contrast-limited adaptive histogram equalization were utilized to preprocess our images, where upon data augmentation around 24576 images were generated utilizing the given dataset. In addition, we also integrated IN breast, MIAS, DDSM data’s together, then all the images were resized to $227*227$ pixels. Furtherly, this model was trained and tested for accuracy and precision. Thus, this paper would cover various types of breast cancer imaging modalities along with the research related to implementation of CNN model for breast mammogram classification and for developing comprehensive dataset with augmented images thus improving 3D architecture performance.