A Fusion of Transfer Learning Features on Breast Cancer Classification

This paper presents a feature extraction fusion method that can help improve the performance of breast cancer classification. Specifically, we conducted three experiments to determine the performance of feature extraction performance, namely: feature extraction with one model and fusion of feature extraction on 2 and 3 transfer learning models. We combine features that are considered rich information that can capture the deep features of the image. The dataset used in this study is a private dataset collected from hospital patients. The data set is in the form of images divided into 2 modalities, namely ultrasound and MAMO. This research introduces a new approach to the incorporation of feature extraction to distinguish between benign and malignant classes. The results obtained from the mammography image yielded an accuracy of 93% and ultrasound yielded an accuracy of 98% when combining the three feature extractions InceptionV3, VGG19, and Xception, so this study shows the potential of combining three feature extractions on MAMO and ultrasound to improve the accuracy of breast cancer classification, and the experimental results show that the accuracy of combining 3 models is the highest.