Deep Learning Assisted Kidney Organ Image Analysis for Assessing the Viability of Transplantation

Abstract

The kidney is a vital organ in humans that removes toxic waste from the body and maintains the balance between water, minerals, and salts. Malfunctioning of this vital organ has become one of the significant public health concerns in recent years. The most viable way to treat patients with acute kidney failure is via transplantation. A healthy substitute is required from a healthy donor, which goes through rigorous examination by experienced clinicians to ascertain its vitality. However, the whole procedure is time-consuming, not reliable, and has high intra-observer and inter-observer variations. For these reasons, we proposed a machine learning-based approach using photographic samples to assess the health of the donor organ. Deep learning models, VGG-16 and DenseNet121, were used for feature extraction from 120 organs labelled 1,2,3,4 and 5, where scores 1 and 2 are good, score 3 is fair (uncertain), and 4 and 5 as poor. Random Forest Regressor and Support Vector Regressor were trained and then used to predict the surgeon-derived score labels, determining whether an organ is transplantable or should be discarded. The results indicate an algorithm of this nature could go a long way show in deciding the transplantability of a kidney organ.