An Efficient and Optimal Deep Learning Architecture using Custom U-Net and Mask R-CNN Models for Kidney Tumor Semantic Segmentation

Today, kidney medical imaging has become the backbone for health professionals in diagnosing kidney disease and determining its severity. Physicians commonly use Computerized Tomography (CT) and Magnetic Resonance Imaging (MRI) scan models to obtain kidney disease information. The significance and impact of kidney tumor analysis drew researchers to semantic segmentation of kidney tumors. Traditional image processing methodologies, in general, require more computational power and manual assistance to analyze kidney medical images for tumor segmentation. Deep Learning advances are enabling less computational and automated models for kidney medical image analysis and tumor lineation. Blobs (regions of interest) detection from medical images is gaining popularity in kidney disease diagnosis and is used widely in detecting tumors, glomeruli, and cell nuclei, among other things. Kidney Tumor segmentation is challenging compared to other segmentation models due to morphological diversity, object overlapping, intensity variance, and integrated noise. In this paper, It have proposed a kidney tumor semantic segmentation model based on CU-Net and Mask R-CNN to extract kidney tumor information from abdominal MR images. Initially, It trained the Custom U-Net architecture on abdominal MR images with kidney masks for kidney image segmentation. The Mask R-CNN model is then used to lineate tumors from kidney images. Experiments on abdominal MR images using Python image processing libraries revealed that the proposed deep learning architecture segmented the kidney images and lined up the tumors with high accuracy.

@article{Parvathi2022,
  title     = {An Efficient and Optimal Deep Learning Architecture using Custom U-Net and Mask R-CNN Models for Kidney Tumor Semantic Segmentation},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {13},
  number    = {6},
  year      = {2022},
  publisher = {The Science and Information Organization},
  author    = {Sitanaboina S L Parvathi and Harikiran Jonnadula},
  doi       = {10.14569/IJACSA.2022.0130639},
  url       = {https://doi.org/10.14569/IJACSA.2022.0130639}
}