Application of Quantitative Ultrasonography and Artificial Intelligence for Assessing Severity of Fatty Liver: A Pilot Study

Hyuksool Kwon, Myeong-Gee Kim, Seok-Hwan Oh, Youngmin Kim, Guil Jung, Hyeon-Jik Lee, Sang-Yun Kim, Hyeon‐Min BaeDiagnostics 2024 · DOI

TL;DR

This pilot study introduces an AI-enhanced quantitative ultrasound (QUS) algorithm that significantly improves the accuracy of assessing non-alcoholic fatty liver disease (NAFLD) severity. By leveraging an AI model to analyze ultrasound signals, this method achieves a strong correlation with MRI-PDFF, a gold standard, and demonstrates superior reliability compared to existing ultrasound techniques.

What Problem Does This Paper Address?

Non-alcoholic fatty liver disease (NAFLD) is a growing global health concern, often linked to obesity and diabetes. Early and accurate detection is crucial to prevent severe complications like cirrhosis and liver cancer. While liver biopsy is the definitive diagnostic method, its invasive nature, cost, and risks limit its use for widespread screening. Traditional ultrasound methods, though non-invasive, often lack the precision and accuracy needed for reliable NAFLD assessment, leading to inter-observer variability and inconsistent results. This paper addresses the need for a more accurate, reliable, and non-invasive diagnostic tool for NAFLD severity assessment.

Methods

The core of this research is an AI-enhanced quantitative ultrasound (QUS) algorithm developed by Barreleye Inc. This algorithm focuses on estimating the attenuation coefficient (AC) of liver parenchyma from ultrasound signals. The AI model utilizes a deep neural network architecture that includes:

  • Encoders: To extract time-varying intensity information from pulse-echo data.
  • B-mode Guided Adaptive Denormalization (BGN): A novel component that uses B-mode ultrasound images to adaptively normalize the encoded features. This helps to focus on AC-relevant information by incorporating morphological context from the liver.
  • Convolutional Layers: To further process the normalized features and derive the final AC value.

This AI-based QUS-AC method was compared against two conventional ultrasound techniques:

  • Canon Attenuation Imaging (ATI): An FDA-approved technique that uses real-time color mapping.
  • FibroTouch Ultrasound Attenuation Parameter (UAP): A device that measures both liver stiffness and attenuation.

All ultrasound measurements were validated against Magnetic Resonance Imaging Proton Density Fat Fraction (MRI-PDFF), considered the reference standard for liver fat quantification. Inter-observer reliability was also assessed using the intraclass correlation coefficient (ICC).

Datasets

The study utilized a synthetic dataset for training and validation of the AI model. This dataset was generated using a custom ultrasound computer simulator based on the k-wave simulation tool in MATLAB. The simulation environment mimicked the Vantage 64LE ultrasound system.

  • Training: 8000 image datasets
  • Validation: 1000 image datasets
  • Testing: 1000 image datasets

The clinical validation was performed on data from 35 participants recruited at a single center.

Key Insights and Ideas

  • AI's Role in Standardizing Ultrasound: The AI algorithm's ability to adaptively select parenchymal areas and derive AC values addresses the inherent limitations of manual settings and operator subjectivity in traditional QUS, leading to more consistent and reproducible measurements.
  • B-mode Guidance is Key: The integration of B-mode images into the AI model (BGN) is a crucial innovation. It allows the algorithm to leverage morphological information to improve the accuracy of AC estimation, suggesting that combining structural and functional ultrasound data can yield better results.
  • AI-QUS as a Superior Alternative: The study strongly suggests that AI-enhanced QUS-AC is not just comparable but potentially superior to existing non-invasive ultrasound methods (ATI and UAP) for assessing NAFLD severity, especially in terms of correlation with MRI-PDFF and inter-observer reliability.
  • Cost-Effectiveness Potential: While not explicitly quantified in this pilot, the paper highlights that AI-enhanced QUS offers a more accessible and potentially cost-effective alternative to MRI and liver biopsies, which could be transformative for widespread NAFLD screening.

Key Results

  • Strong Correlation with MRI-PDFF: The AI-enhanced QUS-AC showed a very strong correlation with MRI-PDFF (r = 0.95, R-squared = 0.8386).
  • Outperforming Traditional Methods:
    • ATI showed a moderate correlation with MRI-PDFF (r = 0.73, R-squared = 0.6537).
  • UAP showed a weaker correlation with MRI-PDFF (r = 0.51, R-squared = 0.0131).
  • Excellent Inter-Observer Reliability: The AI-QUS-AC achieved an ICC of 0.983 for individual observations, indicating excellent reliability.
  • ATI had an ICC of 0.70 for individual observations.
  • UAP had an ICC of 0.30 for individual observations.
  • High Diagnostic Accuracy: The AI-QUS-AC demonstrated high diagnostic accuracy across different grades of NAFLD, with AUROC values ranging from 0.93 to 0.99.

Limitations and Future Work

  • Pilot Study Limitations: The authors acknowledge the limitations of a small sample size (n=35) and a single-center design, which may affect the generalizability of the findings to diverse populations.
  • Homogeneity of Study Population: The homogeneity of the study population raises questions about the algorithm's performance across different demographic and metabolic profiles.
  • Training Data Bias: The efficacy of AI is contingent on the quality and diversity of training data. Biases in the training data could lead to skewed assessments.

Future Work:

  • Expanding Capabilities: The authors suggest expanding the AI-QUS algorithm's capabilities to assess other hepatic conditions and organ systems.
  • Integration with Other Modalities: Integrating AI-enhanced QUS with other diagnostic tools could create a more comprehensive liver disease diagnostic toolkit.
  • Continuous Learning: Further development of AI algorithms could focus on improving adaptability to new data and learning from ongoing clinical use.
  • Larger, Multi-center Studies: Future research should involve larger, multi-center studies to validate these findings and confirm the algorithm's generalizability.

Summary updated

Abstract

Non-alcoholic fatty liver disease (NAFLD), prevalent among conditions like obesity and diabetes, is globally significant. Existing ultrasound diagnosis methods, despite their use, often lack accuracy and precision, necessitating innovative solutions like AI. This study aims to validate an AI-enhanced quantitative ultrasound (QUS) algorithm for NAFLD severity assessment and compare its performance with Magnetic Resonance Imaging Proton Density Fat Fraction (MRI-PDFF), a conventional diagnostic tool. A single-center cross-sectional pilot study was conducted. Liver fat content was estimated using an AI-enhanced quantitative ultrasound attenuation coefficient (QUS-AC) of Barreleye Inc. with an AI-based QUS algorithm and two conventional ultrasound techniques, FibroTouch Ultrasound Attenuation Parameter (UAP) and Canon Attenuation Imaging (ATI). The results were compared with MRI-PDFF values. The intraclass correlation coefficient (ICC) was also assessed. Significant correlation was found between the QUS-AC and the MRI-PDFF, reflected by an R value of 0.95. On other hand, ATI and UAP displayed lower correlations with MRI-PDFF, yielding R values of 0.73 and 0.51, respectively. In addition, ICC for QUS-AC was 0.983 for individual observations. On the other hand, the ICCs for ATI and UAP were 0.76 and 0.39, respectively. Our findings suggest that AC with AI-enhanced QUS could serve as a valuable tool for the non-invasive diagnosis of NAFLD.
BibTeX
@inproceedings{kwon2024application,
  author = {Hyuksool Kwon and Myeong-Gee Kim and Seok-Hwan Oh and Youngmin Kim and Guil Jung and Hyeon-Jik Lee and Sang-Yun Kim and Hyeon‐Min Bae},
  title = {{Application of Quantitative Ultrasonography and Artificial Intelligence for Assessing Severity of Fatty Liver: A Pilot Study}},
  booktitle = {Diagnostics},
  year = {2024},
  doi = {10.3390/diagnostics14121237},
}
Figure