Steatosis Quantification on Ultrasound Images by a Deep Learning Algorithm on Patients Undergoing Weight Changes

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Liver steatosis diagnosis is often subjective.
• Objective quantification using deep learning on ultrasound images is needed.
• This study evaluates a deep learning algorithm in patients with weight fluctuations.

Purpose of the Study:

• To assess the accuracy of a deep learning algorithm for quantifying liver steatosis.
• To correlate algorithm-derived steatosis scores with patient weight changes.
• To determine optimal ultrasound scanning views for the algorithm.

Main Methods:

• Retrospective analysis of ultrasound studies from 74 patients with ≥5% weight change.
• Application of a deep learning algorithm to classify steatosis across four scanning views.
• Correlation of steatosis scores with body weight changes and demographic data.

Main Results:

• The deep learning algorithm demonstrated high consistency across different views (R² > 0.86).
• Steatosis scores significantly correlated with body weight changes (R² = 0.62, p < 0.001).
• Men exhibited a higher liver steatosis/BMI ratio than women (p = 0.026).

Conclusions:

• The right intercostal view using 3-5 images is recommended for optimal steatosis quantification.
• The deep learning algorithm provides objective liver steatosis assessment.
• Quantified steatosis is linked to weight fluctuations and gender.