Assessment of body composition and prediction of infectious pancreatic necrosis via non-contrast CT radiomics and deep learning.

Abstract

<h4>Aim</h4>The current study aims to delineate subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), the sacrospinalis muscle, and all abdominal musculature at the L3-L5 vertebral level from non-contrast computed tomography (CT) imagery using deep learning algorithms. Subsequently, radiomic features are collected from these segmented images and subjected to medical interpretation.<h4>Materials and methods</h4>This retrospective analysis includes a cohort of 315 patients diagnosed with acute necrotizing pancreatitis (ANP) who had undergone comprehensive whole-abdomen CT scans. The no new net (nnU-Net) architecture was adopted for the imagery segmentation, while Python scripts were employed to derive radiomic features from the segmented non-contrast CT images. In light of the intrinsic medical relevance of specific features, two categories were selected for analysis: first-order statistics and morphological characteristics. A correlation analysis was conducted, and statistically significant features were subjected to medical scrutiny.<h4>Results</h4>With respect to VAT, skewness (<i>p</i> = 0.004) and uniformity (<i>p</i> = 0.036) emerged as statistically significant; for SAT, significant features included skewness (<i>p</i> = 0.023), maximum two-dimensional (2D) diameter slice (<i>p</i> = 0.020), and maximum three-dimensional (3D) diameter (<i>p</i> = 0.044); for the abdominal muscles, statistically significant metrics were the interquartile range (IQR; <i>p</i> = 0.023), mean absolute deviation (<i>p</i> = 0.039), robust mean absolute deviation (<i>p</i> = 0.015), elongation (<i>p</i> = 0.025), sphericity (<i>p</i> = 0.010), and surface volume ratio (<i>p</i> = 0.014); and for the sacrospinalis muscle, significant indices comprised the IQR (<i>p</i> = 0.018), mean absolute deviation (<i>p</i> = 0.049), robust mean absolute deviation (<i>p</i> = 0.025), skewness (<i>p</i> = 0.008), maximum 2D diameter slice (<i>p</i> = 0.008), maximum 3D diameter (<i>p</i> = 0.005), sphericity (<i>p</i> = 0.011), and surface volume ratio (<i>p</i> = 0.005).<h4>Conclusion</h4>Diminished localized deposition of VAT and SAT, homogeneity in the VAT and SAT density, augmented SAT volume, and a dispersed and heterogeneous distribution of abdominal muscle density are identified as risk factors for infectious pancreatic necrosis (IPN).