Binary Classification of Mammograms Using Horizontal Visibility Graph

Home Intelligent Systems and Applications Conference paper Binary Classification of Mammograms Using Horizontal Visibility Graph Anirban Ghosh, Priya Ranjan, Naga Srinivasarao Chilamkurthy, Richa Gulati, Rajiv Janardhanan & Pooja Ramakant Conference paper First Online: 01 January 2023 110 Accesses Part of the Lecture Notes in Electrical Engineering book series (LNEE,volume 959) Abstract Breast carcinoma, the most common cancer in women across the world now accounts for almost 30% of new malignant tumor cases. Despite the high incidence rate, breast cancer mortality has been maintained under control thanks to recent advances in molecular biology technology and an enhanced level of complete diagnosis and standard therapy. The method strives to overcome the clinical dilemma of undetected and misdiagnosed breast cancer, resulting in a poor clinical prognosis. Early computer-aided detection by mammography is an important aspect of the plan. In most of the diagnostic strategies currently in vogue, undue importance has been given to one of the performance metrics instead of a more balanced result. In our present study, we aim to resolve this dogma by first converting the mammograms into their equivalent graphical representation and then finding the network similarity between two such generated graphs. Subsequently, we will also elaborate on the use of horizontal visibility graph (HVG) representation to classify images and use Hamming-Ipsen-Mikhailov (HIM) network similarity (distance) metric to develop novel triage mammograms according to the severity of the disease. Our HVG-HIM metric-based classification of mammograms had an accuracy of 88.37%, specificity of 92%, and sensitivity of 83.33%. We also clearly highlight the trade off between performance and processing time.