Edge Computing Based Miniature Maps Using Embedded Webserver For Prediction of Malignancy

Cancer detection from histopathology images is based on extraction of spatial information from whole slide images(WSI) using image processing tools. Machine learning algorithms and state-of-the art convolution neural networks based prediction of malignancy requires cloud services which are often prone to network latency. We put forward a novel technique to classify histopathology images by creating a miniature probability maps to locate the presence or absence of cancer which can be used in any portable device connected to the local network. The shortcomings of loading the high resolution images in mobile devices is resolved as the extraction of patches from WSI and prediction is performed concurrently to minimize computational overhead. The miniature map indicating the presence or absence of cancer in pathology images can be viewed in mobile screen for the uploaded WSI from same device. A custom trained model is used to perform to reduce the processing time thereby faster predictions in end-devices deployed near the medical practitioner as assisting tool in disease diagnosis.