Deep Learning Techniques for Breast Cancer Analysis: A Review Subuhana N Center for Artificial Intelligence TKM College of Engineering Kollam Kerala, India 20075@tkmce.ac.in Aysha Rega S Center for Artificial Intelligence TKM College of Engineering Kollam Kerala, India 20139@tkmce.ac.in Sumod Sundar Center for Artificial Intelligence TKM College of Engineering Kollam Kerala, India sumodsundar@tkmce.ac.in Abstract—Breast Cancer is one of the most frequent cancer among females worldwide. Despite considering medical advance- ments, Breast Cancer remains the world’s second-largest cause of death; hence, early detection of this disease significantly impacts mortality reduction. Breast abnormalities, on the other hand, are complicated to diagnose. Deep learning is the most widely employed technique for accurate diagnosis. Breast Cancer screen- ing technologies such as mammography, ultrasound, and MRI are used extensively. Using image processing and deep learning techniques, the computer-assisted diagnosis help radiologists in identifying problems more quickly. Deep learning algorithms exhibit the best outcomes since they extract the features of images deeply. Furthermore, radiomics analysis has the advantage of being used as a non-invasive method of characterizing tumours directly from clinical medical pictures. For cancer researchers, forecasting the survival rate of Breast Cancer patients is a severe challenge. The efficiency of Deep learning techniques has obtained much attention to provide reliable findings. We have done a brief review on current trends in Breast Cancer analytics using Deep learning techniques. The results are presented in tables to show how different strategies and their outcomes have changed over time. Index Terms—Breast Cancer, CNN, Deep learning, Radiomics, ResNet I. I NTRODUCTION Breast Cancer is becoming more common around the world every year. Breast Cancer is a severe threat to women’s health; it is the leading cause of death from cancer in women. According to the World Health Organization, this disease claimed the lives of around 627,000 women in 2018 [1]. Breast Cancer develops when the tissues of the breast begin to abnormal and uncontrollably divided. Breast Cancer can be controlled if detected early. Many cases may be addressed with early detection, which reduces the death rate. One of the most used imaging methods for identifying Breast Can- cer is mammography. Mammography, on the other hand, is ineffective in dense breasts due to sensitivity and ionization restrictions. Other modalities such as ultrasonography and Magnetic Resonance Imaging (MRI) are frequently used. This heterogeneous disease is marked by a wide range of molecular characteristics, clinical behaviour, morphological appearance, and therapeutic responses. Invasive breast cancer is also becoming increasingly difficult to anticipate and treat due to its complexity and wide range of clinical outcomes [7]. Invasive Ductal Carcinoma (IDC) is the most common subtype of breast cancer, accounting for over 80% of all cases [10]. The tissue regions having invasive regions must be recognized from non-invasive tissues to detect IDC’s presence. Identifying and classifying invasive and non-invasive tissues is a critical clinical activity, and automated technologies can help doctors diagnose patients faster, eliminate errors, and provide better care and treatment. As a result, the capacity to more correctly forecast the prognosis of Breast Cancer inpatients and determining their life expectancy by helping doctors to make informed decisions and guiding appropriate therapy. Recent breakthroughs in medical imaging, particularly in arti- ficial intelligence applied to image processing, show promise in solving clinical constraints such as cancer detection, therapy response assessment, and disease progression tracking. Many medical imaging processing and analysis tasks have also been successfully implemented using Deep learning. Deep learning would be the best alternative for gene profiling which is much faster and cheaper. Deep learning-based solutions, in general, require a large dataset to achieve decent results. Radiomics is used for efficient feature extraction, and it can be used in classification techniques that work well with data on a smaller scale. Radiomics is a generic method rather than traditional statistical approaches (which frequently focus on a small number of variables or features), as it may use a large number of features as input to develop a predictive model. For reducing false-positive rates, precisely identifying the malignancy of lesions discovered in a screening scan is crucial. By extracting and analyzing many quantitative picture features, radiomics can distinguish malignant from benign tumours. Here we aim to highlight the most cutting-edge deep learning techniques in Breast Cancer techniques, including systems for diagnosis, segmentation, classification and prognosis of Breast Cancer. This paper is organized as follows: Section II briefly discusses different techniques used to diagnose, diagnose, diagnose, and classify breast tumours. Section III presents the various Breast Cancer datasets and it is dimensionality features. An exhaustive discussion on the advantages and limitations of various automated techniques in Breast Cancer diagnosis is carried out and summarized in Section IV. 978-1-6654-4885-7/21/$31.00 ©2021 IEEE. 2021 Fourth International Conference on Microelectronics, Signals & Systems (ICMSS) | 978-1-6654-4885-7/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICMSS53060.2021.9673651 Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on April 03,2022 at 18:22:46 UTC from IEEE Xplore. Restrictions apply.