Performance of Ultrasound Techniques and the Potential of Artificial Intelligence in the Evaluation of Hepatocellular Carcinoma and Non-Alcoholic Fatty Liver Disease

Abstract

Global statistics show an increasing percentage of patients that develop non-alcoholic fatty liver disease (NAFLD) and NAFLD-related hepatocellular carcinoma (HCC), even in the absence of cirrhosis. In the present review, we analyzed the diagnostic performance of ultrasonography (US) in the non-invasive evaluation of NAFLD and NAFLD-related HCC, as well as possibilities of optimizing US diagnosis with the help of artificial intelligence (AI) assistance. To date, US is the first-line examination recommended in the screening of patients with clinical suspicion of NAFLD, as it is readily available and leads to a better disease-specific surveillance. However, the conventional US presents limitations that significantly hamper its applicability in quantifying NAFLD and accurately characterizing a given focal liver lesion (FLL). Ultrasound contrast agents (UCAs) are an essential add-on to the conventional B-mode US and to the Doppler US that further empower this method, allowing the evaluation of the enhancement properties and the vascular architecture of FLLs, in comparison to the background parenchyma. The current paper also explores the new universe of AI and the various implications of deep learning algorithms in the evaluation of NAFLD and NAFLD-related HCC through US methods, concluding that it could potentially be a game changer for patient care.

Ultrasound‐Based Machine Learning Approach for Detection of Nonalcoholic Fatty Liver Disease

“…The algorithm will then be able to apply the learned algorithm to analyze the new ultrasound images and to make predictions (external testing set). 18,34,[41][42][43] In our previous studies, AutoML has shown comparable performance to radiologists indicating that these models could help clinical decision making. 23,25 However, to our knowledge there are no prior studies that have prospectively investigated the feasibility of AutoML Vision for the diagnosis of NAFLD on ultrasound.…”

“…Hyperechogenicity is also seen in the presence of fibrosis and early cirrhosis which can reduce the reliability of ultrasound where there are coexisting liver diseases 16 . To overcome the limitations of ultrasound in evaluating low levels of hepatic steatosis, more advanced ultrasound techniques (ie, elastography) and machine learning have been developed 17,18 . Aside from the low accuracy and intra/interobserver variability of conventional ultrasound, ultrasound‐based liver fat quantification is not feasible in people with high body mass index (BMI) 19 …”

Current status of ultrasonography in national cancer surveillance program for hepatocellular carcinoma in South Korea: a large-scale multicenter study

“…5 Moreover, USG sensitivity tends to decline even further in patients with fatty liver or liver cirrhosis. [24][25][26] Considering the increasing incidence of obesity and fatty liver, it is important to consider whether additional tests such as liquid biopsy should be performed in these patients in addition to surveillance USG. [27][28][29] Our study had several limitations.…”

Clinical-radiomic analysis for non-invasive prediction of liver steatosis on non-contrast CT: A pilot study

“…Nowadays, imaging modalities behave well with favourable capacity in diagnosing hepatic steatosis, including US, MRI, CT, etc, ( Marchesini et al, 2016 ). Compared to CT and MRI, US remains the first-line imaging tool for NAFLD diagnosis due to the advantage of simplicity, reproductivity and inexpensiveness ( Monica et al, 2021 ). However, it has insufficient sensitivity and even fails to detect steatosis when <20% or facing high body mass index (BMI) individuals ( Ryan et al, 2002 ; Saadeh et al, 2002 ; Fishbein et al, 2005 ).…”