Advances in the quality of clinical whole-slide images have enabled the systematic digital recognition of metastatic cancer from pathologic scans. To evaluate the accuracy and efficiency of digital assistance on identifying clinical-relevant samples at different severity, we conducted a study utilizing our deep learning algorithm for the detection of cancer metastasis in small image patches taken from larger digital pathology scans. 220,025 images were assessed in the training and evaluation set, and two models were used to predict the binary labels and compared: 4-layer Convolutional Neural Network (CNN) and ImageNet pre-trained model. This study demonstrates the potential of a deep learning algorithm to improve the computer-assisted diagnostic in pathology and clinical care.
Data resource: https://www.kaggle.com/c/histopathologic-cancer-detection