MH Jalandhar, India
Posters & Accepted Abstracts: Arch Clin Microbiol
Evaluation of histopathologic slides for recognition of foci of cancer is a labour intensive and time consuming process. We evaluated two different machine learning models, both deep convolutional neural networks, built on ResNet 50 architecture, in recognition of foci of breast cancer from histopathological images. The first model was pre-trained with the ImageNet dataset, the second was trained only on training set with an optimized learning rate. Whole mounted slide images (WSI) data set prepared by Cruz Roa et.al., was used for this study. 198, 738 images of benign foci and 78, 786 malignant foci were extracted from the WSI images. We split the data set randomly in training and evaluation set. After completion of training, both models could detect foci of cancer with 88% sensitivity. We conclude that pretrained ResNet models have the potential to become an effective screening tool for histologic diagnosis of breast cancer.