LAUSR

Smart healthcare systems manage and deal with a large amount of data generated by the Internet of Medical Things (IOMT) devices. This big data processing requires efficient computer-aided diagnostic (CAD) systems. Recently, IOMT-based systems are in use for lung nodule detection especially for remote locations where experts are not available. Lung cancer is one of the critical types of cancer and requires an efficient CAD for early diagnosis of the disease. In the presented research, we proposed two modified UNET architectures based on enhanced receptive fields with the help of atrous convolutions for lung nodule segmentation that can be used with blockchain-assisted IOMT frameworks. Our first framework is based on dual branches and combined with deep residual learning to extract rich features. It employs different scales of pooling to capture information at both local and global levels of context. The second model utilizes naive inception blocks which consist of parallel convolutions in which different scales of kernels are used to bring out features of distinct size nodules. Moreover, the search space of both models is reduced to lung region of interest (ROI) by utilizing K-means clustering and morphological operators. Both proposed models use atrous convolutions to increase the filter’s field of views. The proposed models achieve 85% and 86.2% dice score on publicly available benchmark LIDC-IDRI dataset and show significant improvement over standard UNET and also outperform all existing published work.