Using SVM and PSO-NN Models to Predict Breast Cancer

Abstract

Breast cancer has become one of the most common cancers among modern women. With the increase of breast cancer patients, the diagnosis data of relevant images, pathology and test reports have also increased dramatically. However, with the increasing workload of doctors, they are also faced with the problems of higher misdiagnosis rate. The imaging and pathological features of breast masses are widely used in the diagnosis and risk prediction of breast cancer. Artificial intelligence can predict the outcomes of breast cancer through the features that are computed from digitized images, which has a good clinical prospect of application. In this paper, Support Vector Machine (SVM) and Particle Swarm Optimization-Neural Network (PSO-NN) were used to establish two different breast cancer prediction models combined with the same Wisconsin Diagnostic Breast Cancer (WDBC) dataset (569 samples, including 357 benign tumors and 212 malignant tumors). The established SVM and PSO-NN models achieved a sensitivity of 95.28% and 95.75%, a specificity of 98.58% and 99.44%, an accuracy of 97.36% and 98.07%, and an Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) of 0.9592 and 0.9873 for breast cancer prediction, respectively. The experimental results show that SVM and PSO-NN prediction models both have high classification accuracy, which can provide new tools for the diagnosis of breast cancer.