Breast cancer constitutes a significant global health concern that impacts millions of women across the world. The diagnosis of breast cancer involves categorizing grades based on the histopathological characteristics of tumor cells. While histopathological assessment remains the established benchmark for breast cancer diagnosis, it is hampered by time-consuming procedures, subjectivity, and susceptibility to human errors. This study introduces a novel approach called ImageNet-VGG16 (IVNet) for the realtime diagnosis of breast cancer within a hospital environment. The research experiments are conducted using a benchmark dataset known as Jimma University Medical Center (JUMC) breast cancer grading. Advanced image processing techniques are applied to preprocess the data, enhancing performance. This preprocessing involves the utilization of Holistically Nested Edge Detection (HED) and Contrast Limited Adaptive Histogram Equalization (CLAHE) for transformation and stain normalization. We employ advanced neural network-based transfer learning techniques to analyze the preprocessed histopathological images and identify affected cells. Various pre-trained models are utilized, including convolutional neural networks (CNN) such as VGG16, ResNet50, InceptionNetv3, ImageNet, MobileNetv3, and EfficientNetV3, in a comparative framework. The principal objective of this research is the accurate classification of breast cancer images into Grade-1, Grade-2 and Grade-3. Through extensive experimental research, we achieved a commendable 97% correct classification rate by utilizing a hybrid of VGG16 and ImageNet as the proposed feature engineering method, IVNet. We also validate our proposed approach performance using other stateof-the-art study data and statistical t-test analysis. Furthermore, we develop a user-friendly Graphical User Interface (GUI) that facilitates real-time cell tracking in histopathological images. Our real-time diagnosing application offers valuable insights for treatment planning and assists medical professionals in making prognoses. Moreover, our approach can serve as a reliable decision support system for pathologists and clinicians, particularly in settings constrained by limited resources and restricted access to expertise and equipment.
