|
Abstract:
|
Pathological examination has been done manually by visual inspection of hematoxylin and eosin (H&E)-stained images. However, this process is labor intensive, prone to large variations, and lacking reproducibility in the diagnosis of a tumor. We aim to develop an automatic workflow to classify different cells found in cancerous tumors portrayed in digital renderings of the H&E-stained images. For a given image, we propose a segmentation technique based on a deep learning convolutional neural network. The network is trained to classify each pixel in a given image. In such a network, a mix of convolutional and pooling layers is used to combine and summarize the data obtained from the RGB coordinates. Fully-connected layers are employed, and a softmax loss function is used to obtain a pixelwise classification map. A probability map indicating nucleus membership is computed based on the classification result. An active contour based method, along with morphological operations and watershed transform, is applied on the probability map to identify the cell nuclei. The classification results can be used to analyze the morphological properties of the tumor cells and predict cancer progress.
|
