Non-Contrast CT Esophageal Varices Grading through Clinical Prior-Enhanced Multi-Organ Analysis

TL;DR

This study tackles the non-invasive grading of esophageal varices by leveraging non-contrast CT scans through a novel multi-organ framework, MOON++. By fusing esophagus, liver, and spleen imaging with clinical priors via ORI, DCCA, and ordinal learning, MOON++ achieves superior EV severity classification over single-organ and radiologist baselines. The approach demonstrates robust gains on a large NCCT dataset, with strong validation, external generalization, and a reader study confirming clinical relevance. The work highlights the potential for NCCT-based, multi-organ assessment to reduce invasive procedures while maintaining diagnostic accuracy, particularly for severe EV (G3).

Abstract

Esophageal varices (EV) represent a critical complication of portal hypertension, affecting approximately 60% of cirrhosis patients with a significant bleeding risk of ~30%. While traditionally diagnosed through invasive endoscopy, non-contrast computed tomography (NCCT) presents a potential non-invasive alternative that has yet to be fully utilized in clinical practice. We present Multi-Organ-COhesion Network++ (MOON++), a novel multimodal framework that enhances EV assessment through comprehensive analysis of NCCT scans. Inspired by clinical evidence correlating organ volumetric relationships with liver disease severity, MOON++ synthesizes imaging characteristics of the esophagus, liver, and spleen through multimodal learning. We evaluated our approach using 1,631 patients, those with endoscopically confirmed EV were classified into four severity grades. Validation in 239 patient cases and independent testing in 289 cases demonstrate superior performance compared to conventional single organ methods, achieving an AUC of 0.894 versus 0.803 for the severe grade EV classification (G3 versus <G3) and 0.921 versus 0.793 for the differentiation of moderate to severe grades (>=G2 versus <G2). We conducted a reader study involving experienced radiologists to further validate the performance of MOON++. To our knowledge, MOON++ represents the first comprehensive multi-organ NCCT analysis framework incorporating clinical knowledge priors for EV assessment, potentially offering a promising non-invasive diagnostic alternative.

Figures (14)

• Figure 1: Demonstration of esophageal varices grading on non-contrast and contrast-enhanced CT scans with registration tian2023same. (a) Grade 0: No EV present, showing normal esophageal thickness without dilation. (b) Grade 1: Mild EV, not visible on non-contrast CT but apparent on venous phase CT post-contrast, displaying mild thickening and mildly tortuous vascular shadows. (c) Grade 3: Severe EV with dilated variceal clusters compressing the stomach fundus, showing esophageal narrowing and heterogeneous enhancement.
• Figure 2: Comparison of organ anatomy in EV: (a) Non-EV subject displaying normal liver and spleen volume. (b) Subject with severe EV (Grade 3), commonly associated with decompensated cirrhosis. In the spleen, enlargement often occurs due to portal vein hypertension. In the liver, several changes can be observed, e.g., alterations in liver density resulting from abnormal perfusion and the presence of hepatic ascites surrounding the abdominal cavity. Clinically, LSVR is a more effective factor in evaluating liver fibrosis.
• Figure 3: Overview of the Multi-Organ Cohesion Network++. (a) Organ representation interaction. (b) Adaptor. (c) Classifier.
• Figure 4: 3D visualization and projection of the liver, spleen volume and LSVR.
• Figure 5: Challenges exist when using well pre-trained models, e.g., TotalSegmentator wasserthal2023totalsegmentator for organ segmentation, as multiple failure cases underscore the difficulty in generating accurate liver and spleen masks in subjects with liver cirrhosis.