EEF-Net: An Enhanced EfcientNet for breast tumor classifcation in mammograms Dishant Padalia, Kush Vora, Darshil Mehta and Ninad Mehendale ∗ ARTICLE INFO Keywords: Neural Networks Image Processing Breast Cancer Radiology ABSTRACT Breast cancer is one of the most common cancer types, and treatment largely depends on early de- tection. A person has a 13% (1 in 8 risks) of developing breast cancer at some time in their lives. Computer-aided diagnostic (CAD) systems powered by deep learning algorithms have enabled data analysis at high rates without compromising performance. Our work proposes an Enhanced Efcient- Net (EEF-Net) for determining the severity of breast cancer through mammograms. EEF-Net is built on top of EfcientNet and has been fne-tuned to classify mammograms into three classes: benign, malignant, and healthy. The architecture was trained using the publicly accessible MIAS dataset, and a sophisticated image pre-processing pipeline was used to remove noise and other artifacts from the mammograms. The model achieved state-of-the-art results in the classifcation of breast can- cer, achieving an accuracy of 97.14%, 98.67% sensitivity, 99.30% specifcity, and 98.30% precision. EEF-Net will assist radiologists in mass screening patients with high precision and will minimize the radiologists’ workload. 1. Introduction Cancer mayoclinic.org is a wide range of diseases char- acterized by the uncontrollable division of abnormal cells with the ability to interfere and destroy normal human tis- sue. Cells die as they age or become damaged, and new cells supplant the existing ones. However, when this well-ordered system breaks down, abnormal or damaged cells grow and start reproducing. As a result, these damaged cells unite and produce tumors, which are tissue lumps. These tumors can be benign or cancerous. Cancer has the ability to spread throughout the body and is the world’s second largest cause of mortality. Breast cancer is one such form of cancer which afects both men and women but is far more common in women. In the year 2020 itself, there were 2.3 million who.int new cases of breast cancer with over 680,000 deaths globally. As of the end of 2020, 7.8 million women had been diagnosed with breast cancer in the previous fve years, surpassing lung cancer and making it the world’s most commonly diagnosed malignancy. Currently, over 3.8 million cancer.net women in the United States have been diagnosed with breast cancer. In the United States in 2022, it is expected that 287,850 can- cer.net women will be diagnosed with invasive breast can- cer and 51,400 women will be diagnosed with non-invasive breast cancer. Men account for 0.5-1% of all breast cancer cases. In the year 2022, an estimated 2,710 men in the United States will be diagnosed with metastatic breast cancer. Breast cancer develops in the breast tissue and occurs when breast cells mutate (change) and multiply uncontrol- lably, resulting in a mass of tissue (tumor). Breast cancer, like other cancers, has the ability to infltrate and grow in the tissue around the breast. There are various forms of breast cancer, but ductal carcinoma is the most frequent, account- ∗ Corresponding author (ninad@somaiya.edu) dishant.padalia@somaiya.edu (D. Padalia); kush.v@somaiya.edu (K. Vora); darshil05@somaiya.edu (D. Mehta) ing for more than 80% of all occurrences. This cancer de- velops in the milk ducts and spreads to surrounding breast tissue after breaking through the duct wall. Lobular cancer begins in the breast lobules (where milk is generated) and spreads to the surrounding breast tissue. This form of can- cer accounts for 10% to 15% of all breast cancers. Infam- matory breast cancer is an uncommon and serious kind of cancer that seems to be an infection. Redness, swelling, pit- ting, and dimpling of the breast skin are common symptoms of infammatory breast cancer. Obstructive cancer cells in their skin’s lymph veins cause it. Breast cancers can form in other regions of the breast besides the milk ducts and lobules, however these types of cancer are less prevalent. Angiosar- coma, which develops in the cells that line blood or lymph vessels, and Phyllodes tumors, which begin in connective tissues and are usually benign, are two such examples. Figure 1: The mammogram above shows diferent types of artifacts, muscle/tissues, and tumor. On a mammogram, a lump or tumor will appear as a concentrated white area. Breast cancer diagnostic tests include mammography, positron emission tomography (PET) scans, ultrasonogra- phy, and magnetic resonance imaging (MRI). However, mammography III (1976) is the most common and preferred approach because it makes use of the fact that a large portion of the breast contains fat tissue, which is generally transpar- Page 1 of 12 This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4220435 Preprint not peer reviewed