Development of a AI-assisted dual-energy CT protocol for enhanced detection of small-bowel pathologies in radiography

Project Abstract

This study presents the development and clinical evaluation of an AI-assisted dual-energy computed tomography (DECT) protocol designed to enhance detection and characterization of small-bowel pathologies in radiography. The work addresses limitations of conventional single-energy CT in differentiating subtle mucosal lesions, early inflammatory changes, and small-volume obstructions from surrounding mesenteric fat and bowel contents. A multimodal dataset comprising offline DECT spectral datasets, routine abdomen-pelvis CT scans, and corresponding clinical reports was curated from a diverse patient cohort, including cases with Crohn’s disease, diverticulitis, ischemia, neoplasms, and post-surgical alterations. The imaging workflow integrates a robust material decomposition pipeline to generate iodine and virtual non-contrast (VNC) maps, followed by an AI-driven segmentation and classification framework trained to identify regions of interest (ROIs) within the small intestine with higher sensitivity and specificity than standard CT interpretation. Key methodological components include (1) optimization of dual-energy acquisition parameters (kVp pairings, dose modulation, and spectral separation) to maximize lesion conspicuity while minimizing patient dose; (2) advanced image reconstruction algorithms with noise-regularized material decomposition to preserve high-frequency details pertinent to mucosal discontinuities and subtle wall thickening; (3) a deep learning module employing convolutional neural networks (CNNs) and transformer-based architectures for ROI detection, lesion characterization (inflammation, fibrosis, edema, or neoplasm), and integration of iodine density as an objective inflammatory or vascular marker; (4) fusion of DECT-derived quantitative metrics (iodine concentration, effective atomic number, Z-effective maps) with radiomics features extracted from the small-bowel wall and surrounding mesentery to improve differential diagnosis; and (5) an AI-assisted decision-support interface that prioritizes imaging findings by confidence score and provides actionable recommendations for further clinical workup. The study evaluates diagnostic performance against expert radiologist consensus and surgical pathology where available. Metrics include lesion detection rate, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and dose-length product (DLP) comparisons. A prospective validation cohort examines generalizability across scanners and vendor implementations, with stratified analyses by pathology type, disease stage, and body habitus. Secondary objectives assess the added value of DECT-derived iodine maps in distinguishing inflammatory from fibrotic changes in Crohn’s disease and in identifying early neoplastic transformation within scarred or fibrotic segments. Preliminary results demonstrate improved detection of subtle mucosal irregularities and early-wall thickening, with higher reader confidence and reduced equivocal readings when DECT-derived metrics are integrated with AI-based segmentation. The iodine density maps significantly contributed to differentiating inflammatory from neoplastic processes in challenging cases, while maintaining dose parity through optimized spectral acquisition. The proposed protocol shows promise for enhancing diagnostic accuracy, enabling earlier intervention, and guiding targeted clinical management in patients with suspected small-bowel pathology, while offering a scalable framework for integration into routine radiology workflows.

Project Overview

What This Project Is About

---

A straightforward exploration of how artificial intelligence can help doctors see small-bowel issues more clearly when using dual-energy CT scans. The project examines whether combining AI with dual-energy imaging can improve the detection and characterization of conditions like inflammation, obstructions, or bleeding in the small intestine.

---

---

The Problem It Addresses

---

---

---

Objectives of the Project

---

1. Assess how AI can enhance dual-energy CT images for the small bowel.
2. Develop a simple workflow to integrate AI into existing radiography practices.
3. Evaluate accuracy improvements in detecting common small-bowel pathologies.
4. Compare AI-assisted results with standard CT interpretations.

---

---

What You Will Do Step by Step

---

1. Review dual-energy CT basics and small-bowel pathology signs.
2. Gather or simulate a dataset of CT scans with small-bowel conditions.
3. Train a basic AI model to highlight relevant features on dual-energy images.
4. Test the model on unseen scans and measure detection accuracy.
5. Assess usability and potential integration into radiology workflows.

---

---

Expected Outcome

Anticipated results include a practical AI-assisted protocol that improves detection rates of small-bowel issues on dual-energy CT, along with a clear assessment of its benefits and limitations for everyday radiography practice.