Automated Computed Tomography-based Liver Steatosis Risk Stratification of Deceased Organ Donors Using Real-world...

Area of Science:

• Radiology
• Organ Transplantation
• Medical Imaging Analysis

Background:

• Macrovesicular steatosis in liver donors increases transplant risks.
• Prerecovery computed tomography (CT) assessment is variable and requires expertise.
• Automating CT-based steatosis evaluation can improve efficiency.

Purpose of the Study:

• To develop and validate automated methods for assessing macrovesicular steatosis in liver donor CT scans.
• To compare the performance of different predictive models, including clinical features, univariable metrics, radiomics, and deep learning.

Main Methods:

• Analysis of 147 liver donor CT scans with biopsy reports.
• Automatic segmentation of liver and spleen.
• Development of predictive models: clinical features, univariable metrics (liver attenuation, liver-to-spleen ratio, liver-spleen difference), radiomics, a 2-step radiomic approach, and a 3D convolutional neural network.
• Performance evaluation using area under the receiver operating characteristic (AUROC) with repeated cross-validation.

Main Results:

• The 2-step automated approach achieved a high AUROC of 0.87 (IQR, 0.81-0.92).
• Univariable models showed strong performance: liver attenuation (AUROC 0.81), liver-to-spleen ratio (AUROC 0.84), and liver-spleen difference (AUROC 0.83).
• The 2-step model significantly outperformed the clinical-only model (AUROC 0.65).

Conclusions:

• An automated 2-step radiomic model effectively predicts macrovesicular steatosis in liver donors.
• This automated method can support clinical decision-making in the donor-offer process.
• Streamlining donor evaluation can expedite the process and potentially improve transplant outcomes.