Performance evaluation of breast cancer classification using traditional and modern techniques: A comparative study
Yuanquan Xie, Junyi Zhu · 2021
In recent years, the incidence of breast cancer has increased year by year, posing a great threat to women's health. There are two types of breast cancer, non-invasive and invasive, and most patients get invasive breast cancer, which is fatal. Therefore, it's meaningful and necessary for breast cancer diagnosis at early stage. Breast lumps are the most common manifestations and they are detectable on different formats of medical images. However, traditional human-based diagnosis and image analysis place high demands on doctors’ professional skills and energy. Fortunately, with the development of artificial intelligence technology, computer-aided diagnosis (CAD) is gradually replacing the traditional mammogram diagnosis and analysis. In this study, we developed a non-invasive CAD system on the digital mammograms from the Minimammographic dataset by using discrete wavelet transform (DWT) and higher order gradient (HoG) image decomposition methods for features extraction, and principle component analysis (PCA) for dimensionality reduction. Then we compared the classification effects of traditional machine learning algorithms and modern deep learning based algorithms on the decomposited images. The traditional supported vector machine (SVM) achieved an accuracy of 79.66% and the modern convolutional neuron network (CNN) achieved an accuracy of 83.05%. They showed a slight difference. Considering the computational cost of the CAD system, the traditional SVM seems to be more suitable. In addition, we also tried to use a breast lump index (BLI) to quantify the patients’ diagnosis results.