[Paper Review] A Deep learning Approach to Generate Contrast-Enhanced Computerised Tomography Angiography without the Use of Intravenous Contrast Agents

Anirudh Chandrashekar, Ashok Handa+4 more|arXiv (Cornell University)|Mar 2, 2020

Advanced X-ray and CT Imaging4 citations

TL;DR

This study proposes a deep learning approach using a Cycle Generative Adversarial Network (CycleGAN) to generate contrast-enhanced CT angiography (CTA) images from non-contrast CT scans, eliminating the need for intravenous contrast agents. The method successfully differentiates blood, thrombus, and vessel wall tissues by leveraging intrinsic Hounsfield unit differences in non-contrast data, producing CTA-like images that closely resemble ground-truth contrast-enhanced scans with high visual fidelity.

ABSTRACT

Contrast-enhanced computed tomography angiograms (CTAs) are widely used in cardiovascular imaging to obtain a non-invasive view of arterial structures. However, contrast agents are associated with complications at the injection site as well as renal toxicity leading to contrast-induced nephropathy (CIN) and renal failure. We hypothesised that the raw data acquired from a non-contrast CT contains sufficient information to differentiate blood and other soft tissue components. We utilised deep learning methods to define the subtleties between soft tissue components in order to simulate contrast enhanced CTAs without contrast agents. Twenty-six patients with paired non-contrast and CTA images were randomly selected from an approved clinical study. Non-contrast axial slices within the AAA from 10 patients (n = 100) were sampled for the underlying Hounsfield unit (HU) distribution at the lumen, intra-luminal thrombus and interface locations. Sampling of HUs in these regions revealed significant differences between all regions (p<0.001 for all comparisons), confirming the intrinsic differences in the radiomic signatures between these regions. To generate a large training dataset, paired axial slices from the training set (n=13) were augmented to produce a total of 23,551 2-D images. We trained a 2-D Cycle Generative Adversarial Network (cycleGAN) for this non-contrast to contrast (NC2C) transformation task. The accuracy of the cycleGAN output was assessed by comparison to the contrast image. This pipeline is able to differentiate between visually incoherent soft tissue regions in non-contrast CT images. The CTAs generated from the non-contrast images bear strong resemblance to the ground truth. Here we describe a novel application of Generative Adversarial Network for CT image processing. This is poised to disrupt clinical pathways requiring contrast enhanced CT imaging.

Motivation & Objective

• To develop a method that generates diagnostic-quality contrast-enhanced CT angiography (CTA) without intravenous contrast agents.
• To overcome the limitations of contrast-induced nephropathy and patient discomfort associated with iodinated contrast agents.
• To utilize intrinsic radiomic differences in non-contrast CT data to simulate contrast enhancement.
• To enable retrospective analysis of historical non-contrast CT scans for vascular morphology and aneurysm progression modeling.
• To explore the feasibility of applying this method to other vascular and soft-tissue structures.

Proposed method

• A 2D Cycle Generative Adversarial Network (CycleGAN) was trained to perform non-contrast to contrast (NC2C) image translation.
• Paired non-contrast and contrast-enhanced CT axial slices from 13 patients were used as training data, augmented to 23,551 images.
• Hounsfield Unit (HU) distributions were analyzed in the lumen, intra-luminal thrombus (ILT), and interface regions, revealing statistically significant differences (p < 0.001).
• The generator network learned to synthesize contrast-enhanced appearance by mapping HU patterns in non-contrast images to corresponding contrast-enhanced features.
• The discriminator network was trained to distinguish real contrast-enhanced images from generated ones, enforcing realism in the output.
• The model was evaluated by comparing generated CTA images to ground-truth contrast-enhanced scans for visual and structural similarity.

Experimental results

Research questions

• RQ1 Can deep learning methods accurately simulate contrast enhancement in CT angiography using only non-contrast CT data?
• RQ2 Do intrinsic Hounsfield Unit differences in non-contrast CT scans provide sufficient information to differentiate blood, thrombus, and vessel wall components?
• RQ3 Can a CycleGAN-based approach generate diagnostically useful CTA-like images without paired training data?
• RQ4 To what extent can this method preserve anatomical details such as small vessels (e.g., renal arteries) and complex interfaces like blood-thrombus boundaries?
• RQ5 Can this approach be extended to other vascular and soft-tissue structures beyond the aorta?

Key findings

• Significant Hounsfield Unit (HU) differences were found between the lumen, intra-luminal thrombus (ILT), and interface regions (p < 0.001 for all comparisons), confirming intrinsic radiomic differentiation.
• The CycleGAN model successfully generated CTA-like images from non-contrast CT scans with strong visual resemblance to ground-truth contrast-enhanced images.
• The model demonstrated the ability to differentiate visually incoherent soft tissue regions such as blood, thrombus, and vessel wall in non-contrast scans.
• Generated images preserved fine anatomical details, including small arterial branches such as renal and vertebral arteries.
• The method enables the potential reuse of historical non-contrast CT scans for advanced vascular morphological analysis, including 3D geometric and volumetric modeling of abdominal aortic aneurysms.
• The approach may reduce reliance on intravenous contrast agents, thereby mitigating risks of contrast-induced nephropathy and injection-related complications.