An Automated Deep Learning Pipeline for Non-Invasive Detection and Grading of Hepatic Steatosis from Ultrasound Imagery
- Version
- Download 9
- File Size 440.29 KB
- File Count 1
- Create Date 20 November 2025
- Last Updated 20 November 2025
An Automated Deep Learning Pipeline for Non-Invasive Detection and Grading of Hepatic Steatosis from Ultrasound Imagery
1st Arjun Saharawat
Department of Computer Science and Engineering Sharda University
Greater Noida, India arjunsaharawat314@gmail.com
3rd Keertan
Department of Computer Science and Engineering Sharda University
Greater Noida, India Keertancr7@gmail.com
2ndProf. (Dr.) V Sathyasuntharam
Department of Computer Science and Engineering Sharda University
Greater Noida, India sathiya4196@gmail.com
4th Asim Varshney
Department of Computer Science and
Engineering Sharda University
Greater Noida, India aseemvarshney00@gmail.com
Abstract—The global burden of Non-Alcoholic Fatty Liver Dis- ease (NAFLD) necessitates the development of accurate, scalable, and non-invasive diagnostic tools. The current gold standard, liver biopsy, is highly invasive and prone to sampling variability, while conventional B-mode ultrasonography suffers from signifi- cant operator dependency and diagnostic subjectivity. This paper introduces a unified, end-to-end deep learning pipeline for the automated detection and grading of hepatic steatosis severity (Normal, Mild, Moderate, Severe) from standard ultrasound images. The core of the system is a Convolutional Neural Network (CNN) based on the ResNet-50 architecture, utilizing transfer learning to extract highly discriminative features from complex echotextural patterns within the liver parenchyma. Evaluated on an independent test set of 1,000 images, the model achieved a high overall accuracy of 94.5% and a macro-average Area Under the Curve (AUC) of 0.98. These results demonstrate the system’s ro- bust capability to maintain high sensitivity and specificity across all disease grades, significantly surpassing the limitations of subjective manual assessment. This integrated solution enhances diagnostic consistency, improves clinical workflow efficiency, and offers a powerful, objective platform for widespread NAFLD screening and longitudinal disease monitoring.
Index Terms—Hepatic steatosis, non-alcoholic fatty liver dis- ease (NAFLD), deep learning, convolutional neural network (CNN), medical imaging, ultrasound, computer-aided diagnosis, ResNet, non-invasive diagnostics.