Multi-Dimensional Color Space Analysis with Deep Neural Network Architectures for Precision Breast Cancer Diagnosis

Early detection of cancer can significantly increase life expectancy. Accordingly, in recent years, there has been remarkable progress in the development of computer-aided diagnosis (CAD) systems to assist physicians and specialists. The use of histopathological images in breast cancer diagnosis is considered the gold standard; therefore, CAD systems that utilize histopathological images can be highly effective in supporting physicians’ diagnoses. In this study, we employed various pre-trained deep networks such as VGG, ResNet, Xception, and InceptionResNet to detect and classify microscopic images from the BACH breast cancer dataset. Our goal was to provide a comprehensive comparison of different methods based on evaluation metrics including accuracy, F۱-score, recall, and precision. The main innovation of this thesis lies in examining the effect of choosing different color spaces—such as RGB, YCbCr, and HSV—on the performance of the utilized networks. Our objective in this comparison is to select a color space that mimics the human eye (pathologist) as closely as possible. The findings of this study show that utilizing and integrating features from different layers of convolutional networks significantly improves network performance. Ultimately, for binary classification using the InceptionResNet network, feature fusion layer, and fully connected layer in the HSV color space, a classification accuracy of ۹۲% was achieved. For four-class classification using the Inception network, feature fusion layer, and fully connected layer in the HSV color space, a classification accuracy of ۸۶% was obtained.

Keywords:

Breast cancer detection , computer-aided diagnosis systems , deep networks , color space