UNICORN: Ultrasound Nakagami Imaging via Score Matching and Adaptation for Assessing Hepatic Steatosis

A research team from KAIST has developed UNICORN (Ultrasound Nakagami Imaging via Score Matching and Adaptation), an AI-powered method that significantly improves ultrasound-based detection of hepatic steatosis (fatty liver disease). Unlike conventional B-mode ultrasound that struggles with detailed tissue characterization, UNICORN analyzes the statistical properties of backscattered ultrasound signals using score matching—a technique from machine learning that estimates probability distributions. This approach provides a closed-form mathematical estimator for Nakagami parameters, which describe how ultrasound waves scatter through different tissue types.

Traditional Nakagami imaging methods face two major limitations: they require optimal window size selection and suffer from estimator instability, both degrading image resolution. UNICORN overcomes these by offering pixel-by-pixel parameter estimation rather than analyzing only specific regions of interest with fixed window sizes. This enables comprehensive tissue mapping across the entire ultrasound image, resulting in high-resolution visualization of fat accumulation patterns that conventional methods cannot achieve.

The researchers validated UNICORN using real patient envelope data, demonstrating its clinical utility for detecting hepatic steatosis. The method showed robustness and generalizability across different cases, providing clear visual distinction in backscattered statistics associated with fatty liver conditions. By transforming ultrasound from a qualitative screening tool to a quantitative imaging modality, UNICORN could enable earlier and more accurate diagnosis without requiring more expensive imaging technologies like MRI or CT scans.