Improved accuracy for myocardial blood flow mapping with deep learning-enabled CMR arterial spin labeling (DeepMASL): Validation by microsphere in vivo

TL;DR

A new deep learning method called DeepMASL improves the accuracy of measuring blood flow in the heart, validated against a gold-standard technique in a canine model.

Contribution

Introduces DeepMASL, a deep learning-enabled arterial spin labeling technique that significantly improves myocardial blood flow quantification accuracy.

Findings

DeepMASL reduced hyperemic MBF underestimation from 33–49% to less than 10% error.

Strong correlations (r = 0.85–0.86) were found between DeepMASL and microsphere MBF measurements.

Bland-Altman analysis showed minimal bias and narrow limits of agreement in both normal and stenotic dogs.

Abstract

Current myocardial arterial spin labeling (ASL) methods are sensitive to noise (background and physiology), which limits the accuracy of myocardial blood flow (MBF) measurement. In this study, we demonstrated a new deep learning-enabled myocardial ASL approach (DeepMASL) and evaluated its accuracy to quantify MBF in a canine model of coronary arterial disease in vivo. The reference method was invasive microsphere measurements.

Eighteen mongrel dogs were divided into two groups: healthy (n = 9) and coronary stenosis (n = 9). The latter was induced in an open-chest model with 3 types of stenosis: 50% (n= 3), 70% (n = 3), and 90% (n = 3). Each dog received pharmaceutically induced hyperemia, by the infusion of either dipyridamole or dobutamine to induce different levels of MBF. Microsphere measurements were performed at rest and during the hyperemia. A cardiac ASL sequence was employed to acquire ASL signals at the mid-section of the heart, at rest and during the hyperemia. A physics-based deep learning network (DeepMASL) was developed using synthetic ASL signals with different levels of background noise. Segmented MBF values produced by both non-DeepMASL and DeepMASL methods were measured in all dogs to compare with segmented microsphere MBF values.

While the non-DeepMASL method severely underestimated hyperemic MBF by 33–49%, the DeepMASL approach dramatically improved the accuracy to obtain error less than 10%. There were strong correlations (r = 0.85 – 0.86) in segmented MBF values between measurements by DeepMASL and microsphere methods in either normal or ischemic dogs with varying degrees of coronary artery stenosis. The Bland-Altman analysis revealed mild to moderate variations of DeepMASL (95% confident interval: −1.3 to 1.5 ml/min/g in normal dogs and −1.8 to 1.3 ml/min/g in stenotic dogs) and almost zero bias.

The novel DeepMASL demonstrates much improved accuracy in the quantification of regional MBF at varying levels of coronary artery stenosis, which is correlated strongly with microsphere MBF values. The validated data indicates the potential for this DeepMASL technique to be translated for noncontrast diagnosis of myocardial perfusion deficit in a clinical setting.

ga1

Figures