Deep Learning for Resolving 3D Microstructural Changes in the Fibrotic Liver

Portal hypertension, a life-threatening complication of cirrhosis, is largely triggered by increased intrahepatic vascular resistance. Fibrosis, regenerative nodule formation, intrahepatic angiogenisis and sinusoidal remodelling are classical mechanisms that account for increased intrahepatic vascular resistance in cirrhosis. Our study leverages high-resolution 3D synchrotron radiation-based microtomography and a deep learning-based segmentation approach to investigate these microstructural changes in the liver. By employing a multi-planar U-Net model, trained using annotated tomographic slices sourced from our developed online learning tool, we effectively quantify critical vascular parameters such as sinusoid proportions, local thickness, and connectivity. These insights advance our understanding of liver microarchitecture and also allows correlating vascular parameters to inflammation and fibrosis severity. Understanding and quantifying these microstructural changes is essential to be able to predict the transition from seemingly benign conditions like steatosis or mild inflammation to severe fibrosis and cirrhosis.