Deep Learning-based Angiogram Generation Model for Cerebral Angiography without Misregistration Artifacts | JoVE Visualize

Deep Learning-based Angiogram Generation Model for Cerebral Angiography without Misregistration Artifacts

Daiju Ueda¹, Yutaka Katayama¹, Akira Yamamoto¹

¹From the Departments of Diagnostic and Interventional Radiology (D.U., A.Y., S.L.W., H. Tatekawa, H. Takita, T.H., A.S., Y.M.), Neurosurgery (T. Ichinose, H.A., Y.W., T.G.), and Medical Statistics (D.K.), Graduate School of Medicine, Osaka City University, 1-4-3 Asahi-machi, Abeno-ku, Osaka 545-8585, Japan; and Department of Radiology, Osaka City University Hospital, 1-5-7 Asahi-machi, Abeno-ku, Osaka, 545-8586, Japan (Y.K., T. Ichida).

Radiology|March 31, 2021

View abstract on PubMed

Summary

A new deep learning (DL) model generates clear digital subtraction angiography (DSA)-like cerebral angiograms from dynamic angiograms, overcoming patient movement artifacts for improved procedural clarity.

Area of Science:

Medical Imaging
Artificial Intelligence in Medicine
Neuroradiology

Background:

Digital subtraction angiography (DSA) is crucial for visualizing cerebral vasculature.
Patient movement during DSA can cause misregistration artifacts, degrading image quality and interrupting procedures.
Developing automated methods to correct these artifacts is essential for reliable diagnosis.

Purpose of the Study:

To develop and validate a deep learning (DL) model capable of generating DSA-like cerebral angiograms directly from dynamic angiograms.
To quantitatively and visually assess the clinical utility of DL-generated angiograms, particularly in the presence of artifacts.

Main Methods:

A retrospective study utilized dynamic and DSA image pairs from 40 patients.
A DL model was trained on non-misregistered data and validated using separate datasets, including those with misregistration artifacts.
Quantitative metrics (PSNR, SSIM) and visual rating scales were employed for evaluation.

Main Results:

DL-generated angiograms demonstrated high quantitative agreement with DSA images (PSNR: 40.2 dB, SSIM: 0.97).
Visual evaluation revealed DL-generated angiograms were comparable or superior to original DSA images, even with misregistration artifacts.
The model successfully produced artifact-free, clinically useful cerebral angiograms.

Conclusions:

The developed DL model effectively generates high-quality, artifact-free cerebral angiograms from dynamic acquisitions.
This AI-driven approach offers a promising solution for improving DSA image quality and procedural efficiency in neuroradiology.
The model's ability to mitigate misregistration artifacts enhances diagnostic confidence and clinical workflow.