A comparison between two semantic deep learning frameworks for the autosomal dominant polycystic kidney disease segmentation based on magnetic resonance images

摘要: The automatic segmentation of kidneys in medical images is not a trivial task when the subjects undergoing examination are affected by Autosomal Dominant Polycystic Kidney Disease (ADPKD). Several works dealing with Computed Tomography from pathological were proposed, showing high invasiveness or requiring interaction user for performing images. In this work, we propose fully-automated approach Magnetic Resonance images, both reducing acquisition device and any users Two different approaches proposed based on Deep Learning architectures using Convolutional Neural Networks (CNN) semantic without needing to extract hand-crafted features. details, first performs procedure pre-processing input. Conversely, second two-steps classification strategy: CNN automatically detects Regions Of Interest (ROIs); subsequent classifier ROIs previously extracted. Results show that even though detection shows an overall number false positives, extracted allows achieving performance terms mean Accuracy. However, entire input network remains most accurate reliable better than previous approach. obtained results investigated polycystic since strategies reach Accuracy higher 85%. Also, methodologies performances comparable consistent other found literature working sources, analyses needed task.