Automatic renal perfusion estimation on postoperative PCASL MRI based on deep learning image analysis and segmentation

Anne Oyarzun-Domeño, Izaskun Cía, Rebeca Echeverria-Chasco, María A Fernández-Seara, Paloma L. Martin-Moreno, Nuria Garcia-Fernandez, Gorka Bastarrika, Javier Navallas, Arantxa Villanueva

Keywords: Renal perfusion, PCASL MRI, segmentation

TL;DR: Automatic segmentation of kidney, cortex, and medulla on postoperative PCASL MRI images enables rapid and accurate renal perfusion estimation.

Abstract: Non-invasive arterial spin labeling is a magnetic resonance imaging technique that can be used for kidney transplant evaluation and perfusion estimation. This work proposes an automatic workflow for renal segmentation and perfusion estimation based on a deep learning approach and image analysis, for the postoperative evaluation of the allograft. Our method outperforms state-of-the-art results in terms of multiclass segmentation on low spatial resolution and low signal-to-noise-ratio data. Similarity coefficients above 90% are achieved for kidney, cortex, and medulla segmentation results and perfusion values within the acceptable ranges are obtained.