Ultrasound Scatteromics: an Innovative Strategy for the Diagnosis of Pediatric Hepatic Steatosis

Fatty liver is a health issue that needs urgent attention today, and it has been confirmed as the main pathogenesis of non-viral liver cancer. Fatty liver is not just a problem for adults. More and more reports have pointed out that the prevalence of non-alcoholic fatty liver disease in children and adolescents has increased significantly because of irregular exercise habits and lifestyles such as high-calorie diets. For children, ultrasound imaging techniques are the most convenient and direct examination tools; among them, the controlled attenuation parameter (CAP) based on the vibration-controlled transient elastography (Fibroscan, Echosens, France) is the mainstream parameter for fatty liver diagnosis. However, the Fibroscan system is not suitable for being used in children's measurement situations such as too small rib spacing, poor coordination due to sedition, too thick subcutaneous fat, and time-consuming measurements; it also limits general abdominal scanning due to the lack of image scanning functions and only single-point measurement, resulting in a high degree of uncertainty in CAP measurements of pediatric fatty liver. Considering that the histopathological manifestation of pediatric fatty liver can be regarded as a process of microstructural changes. When ultrasound is transmitted into the liver of children, the generated ultrasound backscattered signals will be the information carrier of tissue characteristics, implying that the construction and implementation of "scatteromics" will be the subject of scientific research with disruptive innovation. Therefore, this research project is planned for three years to completely meet the clinical needs and pain points of pediatric fatty liver assessment by proposing imaging-based scatteromics diagnostic technique, which is constructed by combining backscattering parameters, machine learning, and deep learning models. In this first year, we will develop integrated backscatter, double Nakagami distribution parameters, homodyned-K parameters estimated using neural network, attenuation estimation using the Hilbert-Huang transform to calculate the instantaneous frequency; the diagnostic performances of all the proposed parameters will be explored. In the second year, through the integration of adaptive boosting and random forest and introducing the deep learning model U-net, we will realize the automatic image segmentation of liver parenchyma and scatteromics-based diagnostic technique. In the third, year, we will develop an ultrasound imaging system for scatteromics analysis, which will be compared with the CAP of Fibroscan in diagnosis to explore the advantages and potentials of scatteromics in the assessment of pediatric fatty liver disease. This research project is the first to propose scatteromics, expecting to publish three SCI journal papers per year, two conference presentations in IEEE International Ultrasonics Symposium, and two patents by the end of this research project. We will also look for technique transfer opportunities.

Project ID:PB10907-2928
External Project ID:MOST109-2223-E182-001-MY3