Image Processing Pipeline for Liver Fibrosis Classification Using Ultrasound Shear Wave Elastography

Area of Science:

• Medical Imaging
• Hepatology
• Machine Learning

Background:

• Liver fibrosis staging is crucial for patient management.
• Current manual assessment of ultrasound shear wave elastography (SWE) can be time-consuming and subjective.
• There is a need for more accurate and efficient methods for fibrosis classification.

Purpose of the Study:

• To develop and evaluate an automated method for classifying liver fibrosis stage ≥F2 using SWE.
• To compare the performance of the automated method against a reference manual approach.

Main Methods:

• An automated system integrating image quality assessment, region of interest selection, and machine learning-based classification was developed.
• The method utilized a database of 527 subjects with 5526 SWE images and pathologist-scored biopsies.
• Performance was evaluated using area under the receiver operating characteristic curve (AUROC), specificity at 95% sensitivity, and the number of SWE images required.

Main Results:

• The automated method achieved an AUROC of 0.93, significantly outperforming the reference method's AUROC of 0.69.
• Specificity at 95% sensitivity was 71% for the automated method versus 5% for the reference method.
• The automated method required only four images per decision, compared to eight or more for the reference method.

Conclusions:

• The developed automated SWE classification method significantly enhances accuracy for detecting liver fibrosis stage ≥F2.
• This automated approach reduces the number of required SWE images, potentially improving clinical workflow efficiency.
• The findings suggest a promising tool for objective and rapid liver fibrosis assessment.