Training robust T1-weighted magnetic resonance imaging liver segmentation models using ensembles of datasets with different contrast protocols and liver disease etiologies.

Image segmentation of the liver is an important step in treatment planning for liver cancer. However, manual segmentation at a large scale is not practical, leading to increasing reliance on deep learning models to automatically segment the liver. This manuscript develops a generalizable deep learning model to segment the liver on T1-weighted MR images. In particular, three distinct deep learning architectures (nnUNet, PocketNet, Swin UNETR) were considered using data gathered from six geographically different institutions. A total of 819 T1-weighted MR images were gathered from both public and internal sources. Our experiments compared each architecture's testing performance when trained both intra-institutionally and inter-institutionally. Models trained using nnUNet and its PocketNet variant achieved mean Dice-Sorensen similarity coefficients>0.9 on both intra- and inter-institutional test set data. The performance of these models suggests that nnUNet and PocketNet liver segmentation models trained on a large and diverse collection of T1-weighted MR images would on average achieve good intra-institutional segmentation performance.

Metabolomics biomarkers of hepatocellular carcinoma in a prospective cohort of patients with cirrhosis.

Background & Aims: The effectiveness of surveillance for hepatocellular carcinoma (HCC) in patients with cirrhosis is limited, due to inadequate risk stratification and suboptimal performance of current screening modalities. Methods: We developed a multicenter prospective cohort of patients with cirrhosis undergoing surveillance with MRI and applied global untargeted metabolomics to 612 longitudinal serum samples from 203 patients. Among them, 37 developed HCC during follow-up. Results: We identified 150 metabolites with significant abundance changes in samples collected prior to HCC (Cases) compared to samples from patients who did not develop HCC (Controls). Tauro-conjugated bile acids and gamma-glutamyl amino acids were increased, while acyl-cholines and deoxycholate derivatives were decreased. Seven amino acids including serine and alanine had strong associations with HCC risk, while strong protective effects were observed for N-acetylglycine and glycerophosphorylcholine. Machine learning using the 150 metabolites, age, gender, and PNPLA3 and TMS6SF2 single nucleotide polymorphisms, identified 15 variables giving optimal performance. Among them, N-acetylglycine had the highest AUC in discriminating Cases and Controls. When restricting Cases to samples collected within 1 year prior to HCC (Cases-12M), additional metabolites including microbiota-derived metabolites were identified. The combination of the top six variables identified by machine learning (alpha-fetoprotein, 6-bromotryptophan, N-acetylglycine, salicyluric glucuronide, testosterone sulfate and age) had good performance in discriminating Cases-12M from Controls (AUC 0.88, 95% CI 0.83-0.93). Finally, 23 metabolites distinguished Cases with LI-RADS-3 lesions from Controls with LI-RADS-3 lesions, with reduced abundance of acyl-cholines and glycerophosphorylcholine-related lysophospholipids in Cases. Conclusions: This study identified N-acetylglycine, amino acids, bile acids and choline-derived metabolites as biomarkers of HCC risk, and microbiota-derived metabolites as contributors to HCC development. Impact and implications: The effectiveness of surveillance for hepatocellular carcinoma (HCC) in patients with cirrhosis is limited. There is an urgent need for improvement in risk stratification and new screening modalities, particularly blood biomarkers. Longitudinal collection of paired blood samples and MRI images from patients with cirrhosis is particularly valuable in assessing how early blood and imaging markers become positive during the period when lesions are observed to obtain a diagnosis of HCC. We generated a multicenter prospective cohort of patients with cirrhosis under surveillance with contrast MRI, applied untargeted metabolomics on 612 serum samples from 203 patients and identified metabolites associated with risk of HCC development. Such biomarkers may significantly improve early-stage HCC detection for patients with cirrhosis undergoing HCC surveillance, a critical step to increasing curative treatment opportunities and reducing mortality.

Training Robust T1-Weighted Magnetic Resonance Imaging Liver Segmentation Models Using Ensembles of Datasets with Different Contrast Protocols and Liver Disease Etiologies.

Image segmentation of the liver is an important step in several treatments for liver cancer. However, manual segmentation at a large scale is not practical, leading to increasing reliance on deep learning models to automatically segment the liver. This manuscript develops a deep learning model to segment the liver on T1w MR images. We sought to determine the best architecture by training, validating, and testing three different deep learning architectures using a total of 819 T1w MR images gathered from six different datasets, both publicly and internally available. Our experiments compared each architecture's testing performance when trained on data from the same dataset via 5-fold cross validation to its testing performance when trained on all other datasets. Models trained using nnUNet achieved mean Dice-Sorensen similarity coefficients > 90% when tested on each of the six datasets individually. The performance of these models suggests that an nnUNet liver segmentation model trained on a large and diverse collection of T1w MR images would be robust to potential changes in contrast protocol and disease etiology.