AUTHOR=Yang Yitong , Shah Zahraw , Jacob Athira J. , Hair Jackson , Chitiboi Teodora , Passerini Tiziano , Yerly Jerome , Di Sopra Lorenzo , Piccini Davide , Hosseini Zahra , Sharma Puneet , Sahu Anurag , Stuber Matthias , Oshinski John N. 

TITLE=Deep learning-based left ventricular segmentation demonstrates improved performance on respiratory motion-resolved whole-heart reconstructions

JOURNAL=Frontiers in Radiology

VOLUME=Volume 3 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/radiology/articles/10.3389/fradi.2023.1144004

DOI=10.3389/fradi.2023.1144004

ISSN=2673-8740

ABSTRACT=Deep learning (DL)-based segmentation has gained popularity for routine Cardiovascular Magnetic Resonance (CMR) image analysis and in particular, delineation of left ventricular (LV) borders for volume determination. Free-breathing, self-navigated, whole-heart CMR exams provide high-resolution, isotropic coverage of the heart for assessment of cardiac anatomy including LV volume. The combination of whole heart free-breathing CMR and DL-based LV segmentation has the potential to streamline the acquisition and analysis of clinical CMR exams. The purpose of this study was to compare the performance of a DL-based automatic LV segmentation network trained primarily on CT images in two whole-heart CMR reconstruction methods: 1) in-line, respiratory motion-corrected (Mcorr) reconstruction and 2) off-line, compressed sensing-based, multi-volume respiratory motion-resolved (Mres) reconstruction.  Given that Mres images were shown to have greater image quality from previous studies than Mcorr images, we hypothesized that the LV volumes segmented from Mres images are closer to the manually traced LV endocardial border than the Mcorr images. 

This retrospective study used 15 patients who underwent clinically indicated 1.5T CMR exams with a prototype, ECG-gated 3D radial phyllotaxis bSSFP sequence. For each reconstruction method, the absolute volume difference (AVD) of the automatically and manually segmented LV volumes was used as the primary quantity to investigate whether 3D DL-based LV segmentation generalized better on Mcorr or Mres 3D whole-heart images.  Additionally, we assessed the 3D Dice Similarity Coefficient (DSC) between manual and automatic LV masks and the sharpness of the LV myocardium-blood pool interface of each reconstructed 3D whole-heart images. A two-tail paired student t-test (alpha=0.05) was used to test the significance in this study.

The AVD was lower in the Mres reconstruction than in the Mcorr reconstruction, 7.73 ± 6.54 ml vs. 20.0 ± 22.4 ml, respectively (p=0.03). The 3D DSC was higher for Mres than the Mcorr images, 0.90±0.02 vs. 0.87±0.03 respectively, (p=0.02). Sharpness on the Mres images was higher than on Mcorr images, 0.15±0.05 vs. 0.12±0.04, respectively (p= 0.014). 

We conclude that the DL-based 3D automatic LV segmentation network trained on CT images and fine-tuned on MR images generalized better on Mres images compared to Mcorr images for quantifying LV volumes.