Details

Autor(en) / Beteiligte

• Bhuma, Chandra Mohan

Titel

Three Class Classification of Digital Mammograms using Chebyshev Moments and Convolutional Neural Network

Ist Teil von

• 2020 5th International Conference on Computing, Communication and Security (ICCCS), 2020, p.1-5

Ort / Verlag

IEEE

Erscheinungsjahr

2020

Quelle

IEEE Electronic Library Online

Beschreibungen/Notizen

• Two approaches for effective classification of digital mammograms for early detection of breast cancer are proposed in this work. Detection of breast cancer at an early stage is an effective way for reducing the mortality. Biopsy is the ultimate solution for detection of breast cancer. However, it is invasive and costly too. Digital mammograms are non invasive and assist the radiologist for a preliminary assessment of the disease at an early stage. In the literature most of the works on digital mammogram analysis are based on two class classifications, i.e., whether the image is a normal image or abnormal image. Further, if it is abnormal, the tumor present is malignant or benign. So, the classification is easy as the problem is divided into two binary classification problems. However, there are some works on three class classification of digital mammograms. In this work, publicly available mini MIAS database is used to assess the performance of the proposed algorithms. The first work uses Chebyshev moments for calculating the features and Support Vector Machine with fine tuned hyper parameters is used as a classifier. The second work proposes a simple and compact convolution neural network for the classification. 10 fold cross validation is used in both the approaches for the model preparation. The MIAS dataset is highly imbalanced and hence a misclassification cost is assigned for the underrepresented classes. Performance measures i.e, accuracy, and balanced accuracy are used to demonstrate the usefulness of the proposed algorithms. Results are compared with the state of the art existing works in the literature and superiority of the proposed approaches is justified.