[Paper Review] Visualizing intestines for diagnostic assistance of ileus based on intestinal region segmentation from 3D CT images

Hirohisa Oda, Kohei Nishio+8 more|arXiv (Cornell University)|Mar 3, 2020

Colorectal Cancer Screening and Detection8 references4 citations

TL;DR

This paper proposes a 3D U-Net-based visualization method for diagnosing ileus by segmenting small and large intestines from CT volumes using weakly annotated training data. The approach enables intuitive visualization of stenosed intestinal segments via distance-based coloring from endpoints, with clinicians using a manual selection interface to remove false positives, achieving a mean Dice score of 0.75 across seven ileus cases.

ABSTRACT

This paper presents a visualization method of intestine (the small and large intestines) regions and their stenosed parts caused by ileus from CT volumes. Since it is difficult for non-expert clinicians to find stenosed parts, the intestine and its stenosed parts should be visualized intuitively. Furthermore, the intestine regions of ileus cases are quite hard to be segmented. The proposed method segments intestine regions by 3D FCN (3D U-Net). Intestine regions are quite difficult to be segmented in ileus cases since the inside the intestine is filled with fluids. These fluids have similar intensities with intestinal wall on 3D CT volumes. We segment the intestine regions by using 3D U-Net trained by a weak annotation approach. Weak-annotation makes possible to train the 3D U-Net with small manually-traced label images of the intestine. This avoids us to prepare many annotation labels of the intestine that has long and winding shape. Each intestine segment is volume-rendered and colored based on the distance from its endpoint in volume rendering. Stenosed parts (disjoint points of an intestine segment) can be easily identified on such visualization. In the experiments, we showed that stenosed parts were intuitively visualized as endpoints of segmented regions, which are colored by red or blue.

Motivation & Objective

• To address the challenge of diagnosing ileus in non-expert clinicians due to the long, winding, and poorly contrasted small intestine in CT scans.
• To develop a robust intestinal segmentation method for ileus cases where fluid-filled intestines have low contrast with intestinal walls.
• To enable intuitive visualization of stenosed intestinal parts by coloring segments based on distance from endpoints.
• To reduce reliance on extensive manual labeling by using a weak annotation approach with sparse axial slice labels.
• To support rapid, accurate identification of stenosed regions through interactive manual selection of valid intestinal segments.

Proposed method

• 3D U-Net is trained using weak annotations—only a few axial slices with manually traced intestinal regions are labeled, avoiding full volumetric annotation.
• Patch cropping is applied to extract 3D volumes of size (288,192,16) around labeled slices for efficient training.
• False positive regions from 3D U-Net output are removed via a manual selection interface where clinicians click inside valid intestinal segments.
• Constrained distance transformation is used to identify endpoints of each intestinal segment by computing maximum distance from a selected seed point.
• Each intestinal segment is volume-rendered and colored based on distance from one endpoint, with endpoints highlighted in red and blue for visual clarity.
• Stenosed parts appear as disconnected segments with distinct colored endpoints, enabling quick identification by clinicians.

Experimental results

Research questions

• RQ1 Can a weakly supervised 3D U-Net segmentation method achieve accurate intestinal region detection in ileus CT scans with minimal manual labeling?
• RQ2 Can distance-based coloring from endpoints in volume rendering effectively highlight stenosed intestinal segments for clinical diagnosis?
• RQ3 How effective is a manual selection interface in removing false positives from 3D U-Net outputs in complex ileus cases?
• RQ4 To what extent does the proposed visualization method improve diagnostic clarity for non-expert clinicians in identifying ileus obstructions?
• RQ5 What is the quantitative performance of the method in terms of segmentation accuracy and stenosis visualization fidelity?

Key findings

• The proposed method achieved a mean Dice score of 0.75 across seven ileus cases, indicating strong segmentation performance.
• Case C achieved a high Dice score of 0.84 and successfully visualized the stenosed part as a distinct endpoint colored blue.
• Case A had the lowest Dice score (0.53), attributed to low contrast between intestinal walls and surrounding tissues, indicating limitations in low-contrast scenarios.
• The number of connections between intestinal segments averaged 0.29, indicating that segmentation results were mostly disconnected, which supports accurate stenosis detection.
• The manual selection interface effectively removed false positive regions, as seen in Case C where high responses in the lungs were eliminated from final visualization.
• The coloring scheme successfully highlighted endpoints of intestinal segments, enabling clinicians to intuitively identify stenosed parts as disjoint segments with colored ends.