# PatchDSA: improving digital subtraction angiography with patch-based phase-matching in natural breathing scenarios

## TL;DR

This paper introduces PatchDSA, a new method that improves digital subtraction angiography by reducing motion artifacts during natural breathing.

## Contribution

The novel patch-based phase-matching technique enables better alignment of mask and contrast images in dynamic DSA.

## Key findings

- PatchDSA effectively reduces motion artifacts in abdominal DSA images.
- The method outperformed existing phase-matching techniques in all tested cases.
- It suppresses artifacts from physiological motions like peristalsis and cardiac pulsation.

## Abstract

Digital subtraction angiography (DSA) is used to visualize blood vessels by subtracting pre-contrast (mask) images from contrast images; sequential mask and contrast images are used to generate dynamic DSA images that allow observation of blood flow and organ movements. However, misalignment between mask and contrast images can cause motion artifacts, which not only obscure the appearance of enhanced structures but also lead to the misidentification of patterns as vascular structures. In this study, we proposed a new method for generating abdominal sequential DSA images using a patch-based phase-matching technique between mask and contrast images acquired under natural breathing conditions. Our method divides mask and contrast images into small patches and selects the mask image patch most structurally similar to each patch in the target contrast image. Furthermore, the selected mask image patch is refined by searching for the subpixel-level region that most closely matches the target contrast image patch. The proposed method was evaluated using 20 abdominal angiogram cases, and its performance was compared with an existing phase matching–based method. Our experimental results showed that the proposed method effectively reduced motion artifacts and outperformed the comparison method in all cases. We demonstrated that our method successfully identified the optimal mask image for each contrast image on a patch-by-patch basis, allowing it to suppress artifacts caused by physiological motions such as peristalsis and cardiac pulsation, thereby generating higher-quality DSA images.