TLDR: A new study utilizes a deep learning model with MRI to quantify liver vessel volumes in chronic liver disease (CLD). Researchers found significant reductions in total and hepatic vein-to-volume ratios as CLD progresses, correlating with disease severity markers. This non-invasive method could serve as a valuable imaging biomarker for CLD staging.
The research paper, titled “MRI-derived quantification of hepatic vessel-to-volume ratios in chronic liver disease using a deep learning approach,” introduces a new method for understanding the progression of chronic liver disease (CLD). The study, conducted by a team of researchers including Alexander Herold, Daniel Sobotka, and Georg Langs, investigates how deep learning can be combined with Magnetic Resonance Imaging (MRI) to precisely measure changes in the liver’s blood vessels.
Chronic liver disease is a serious condition marked by ongoing inflammation and the formation of scar tissue (fibrosis), which can eventually lead to cirrhosis and complications such as portal hypertension. Existing diagnostic tools for liver fibrosis, like ultrasound and magnetic resonance elastography, often require specialized equipment and can be inconvenient for patients. Gadoxetic acid-enhanced MRI has emerged as a valuable imaging technique, providing both structural and functional insights into the liver. However, the detailed assessment of the liver’s complex vascular system in the context of CLD progression has not been extensively explored.
This study aimed to address this gap by employing a 3D U-Net deep learning model to analyze gadoxetic acid-enhanced MRI scans. The researchers quantified three key vessel-to-volume ratios: total vessel-to-volume ratio (TVVR), hepatic vein-to-volume ratio (HVVR), and portal vein-to-volume ratio (PVVR). Their primary objective was to determine if these ratios change as CLD advances, comparing healthy individuals, patients with non-advanced CLD (non-ACLD), and those with advanced CLD (ACLD). A secondary goal was to find correlations between these vessel ratios and established biomarkers that indicate liver function, fibrosis, and portal hypertension.
The study involved 197 participants: 35 healthy controls, 44 non-ACLD patients, and 118 ACLD patients. The results showed significant differences in liver vessel volumes across these groups. Both TVVR and HVVR were highest in healthy controls, intermediate in non-ACLD patients, and lowest in ACLD patients. Specifically, TVVR decreased by 28.2% between controls and non-ACLD patients, and by an additional 17.9% between non-ACLD and ACLD patients. HVVR showed even more pronounced reductions, with a 19.0% decrease from controls to non-ACLD, and a 41.2% decrease from non-ACLD to ACLD.
Interestingly, PVVR exhibited a significant 29.4% reduction between healthy controls and non-ACLD patients but remained stable when comparing non-ACLD and ACLD groups. This suggests that the hepatic and portal venous systems undergo different patterns of vascular changes as the disease progresses. The researchers propose that the initial drop in PVVR might be due to early compression of smaller portal branches by developing fibrosis, while its subsequent stability could be a result of opposing forces: continued compression of small branches versus the compensatory widening of larger portal veins due to increasing portal pressure.
Furthermore, the study identified correlations between these vessel-to-volume ratios and various clinical indicators of liver health. HVVR, in particular, correlated more strongly and consistently with biomarkers that signal worsening liver function (such as lower albumin, higher bilirubin, and elevated ALBI and MELD-Na scores) and increased fibrosis/portal hypertension (such as higher FIB-4 score, liver stiffness measurement, and spleen volume). These correlations suggest that these vessel ratios, derived using deep learning, could serve as non-invasive imaging biomarkers to aid in the diagnosis and staging of CLD.
The deep learning model, a 3D U-Net architecture, was trained using manual segmentations from a smaller subset of 24 patients. It achieved good accuracy for liver and spleen segmentation and moderate accuracy for liver vessel segmentation, performing better on larger vessels. The model’s ability to generalize its findings was confirmed by its consistent performance across healthy controls, non-ACLD, and ACLD patient populations.
Also Read:
- Precise Heart Phase Detection Through Motion Analysis in CMR Images
- New AI Approach Improves Detection of Epilepsy-Related Brain Lesions in MRI
While acknowledging certain limitations, such as the retrospective nature of the study and the relatively small dataset used for training the deep learning model, the authors highlight the significant potential of this approach. They conclude that the quantitative measurement of hepatic vessel volumes using deep learning-based MRI analysis can non-invasively detect vascular changes that accompany different stages of chronic liver disease, offering a promising new imaging biomarker for CLD staging. You can read the full research paper here: MRI-derived quantification of hepatic vessel-to-volume ratios in chronic liver disease using a deep learning approach.
