Sickle Cell Disease Severity Prediction from Percoll Gradient Images Using Graph Convolutional Networks

Sickle cell disease (SCD) is a severe genetic hemoglobin disorder that results in premature destruction of red blood cells. Assessment the severity challenging task clinical routine, since causes broad variance SCD manifestation despite common cause remain unclear. Identification biomarkers would predict grade importance for prognosis and assessment responsiveness patients to therapy. Detection changes (RBC) density by means separation Percoll gradients could be such marker as it allows resolve intercellular differences follow most damaged dense cells prone vasoocclusion. Quantification interpretation images obtained from distribution RBCs an important prerequisite establishment this approach. Here, we propose novel approach combining graph convolutional network, neural fast Fourier transform, recursive feature elimination directly image. Two but expensive laboratory test parameters are used training network. To make model independent tests during prediction, these two estimated network image directly. On cohort 216 subjects, achieve prediction performance only slightly below where groundtruth measurements used. Our proposed method first computational difficult prediction. The two-step relies solely on inexpensive simple analysis tools can have significant impact patients’ survival low resource regions access medical instruments doctors limited.