Viscoelasticity Imaging to Assess Liver Cancer

Overview

Ultrasound (US) used for hepatocellular carcinoma (HCC) surveillance suffers from low sensitivity (60-78%) due to fatty liver, obesity, and diffusely nodular appearance in cirrhosis. Once a suspicious malignant lesion is detected at US, guidelines recommend contrast-enhanced US, magnetic resonance imaging (MRI) or computed tomography (CT) scans to confirm suspicion. The investigators' team has developed innovative quantitative US (QUS) techniques that have a high potential to improve tissue characterization in terms of sensitivity and specificity. The investigators hypothesize that advanced QUS providing tumor viscoelasticity assessment, sub-resolution tissue structure characterization and US attenuation in the framework of a machine learning classification model can improve HCC diagnosis compared with standard US. Early detection through systematic US surveillance translates into curative therapy in a higher proportion of patients and into improvements in survival rates. Thus, there is an urgent need to investigate innovative and cost-effective imaging techniques for improving detection and characterization of HCC. The proposed QUS methods are experimental and will be validated in this proof-of-concept clinical study. A major impact of this work, for patients and medical institutions, will be to improve early-stage detection and characterization of HCC, and offer alternatives in patients with negative or inconclusive conventional US. QUS are low-cost, non-invasive and non-irradiating imaging modalities available from a single exam (i.e., no additional imaging session is necessary).

RESEARCH QUESTION AND BACKGROUND: Primary liver cancer or hepatocellular carcinoma (HCC) is the fifth most common cancer in men and the seventh in women and is the second cause of cancer mortality worldwide. In Canada, HCC is the only cancer for which mortality is increasing. More than 80% of HCC cases occur in individuals with advanced liver fibrosis (cirrhosis) due to viral hepatitis infection (B and C), non-alcoholic fatty liver disease (NAFLD), and alcoholic liver disease. Once cirrhosis is established, there is a significantly increased risk of developing HCC. Furthermore, HCC is observed in obese diabetic individuals without cirrhosis, increasing the population of patients at risk with a disease that has high fatality rate. HCC surveillance is associated with significantly prolonged survival. However, only 52% of patients undergoing surveillance have early HCCs that are eligible for curative treatment, whereas remainder of patients have intermediate- or advanced-stage disease eligible for bridge or palliative treatment only. HCC surveillance is also associated with significant improvements in early-stage detection, curative-treatment rates, and survival, even after adjusting for lead-time bias. North American guidelines recommend ultrasound (US) surveillance every 6 months in at-risk patients who are non-cirrhotic hepatitis B carriers and cirrhotic. However, a key challenge for US is the low sensitivity (60-78%) for identifying a lesion due to liver steatosis and cirrhosis. Once a suspicious malignant lesion is detected at US, current American Association for the Study of Liver Diseases (AASLD) guidelines recommend contrast-enhanced US, magnetic resonance imaging (MRI) or computed tomography (CT) scans to confirm suspicion. GOAL: The long-term reaching goal is to develop US biomarkers of focal liver lesions and strategies to improve diagnostic sensitivity to HCC while maintaining a high specificity. This would constitute a major breakthrough because HCC diagnosis currently requires a combination of US for screening and confirmation using MRI, CT and less often biopsy. OBJECTIVES: 1) Develop a machine learning model based on QUS for classification of solid hepatocellular carcinomas identified at US and diagnosed with MRI (or biopsy if required); 2) Determine if QUS maps can improve visual detection of suspected lesions at US; 3) Compare performance of QUS- versus MRI-based viscoelastography for lesion characterization. Hypothesis: the investigators hypothesize that advanced QUS providing tumor viscoelasticity assessment, sub-resolution tissue structure characterization and US attenuation in the framework of a machine learning classification model can improve HCC diagnosis compared with standard US. METHODOLOGY - Study design: This will be a clinical study with two sequential cohorts: 1) a training cohort of 100 patients at risk for HCC to optimize QUS biomarkers for classification of solid liver lesions using MRI and/or biopsy as gold standard clinical references; and 2) a validation cohort of 100 patients to confirm diagnostic performance. Data analysis: Random forests machine learning to develop QUS classification models. Sensitivity and specificity to assess diagnostic accuracy, according to MRI and/or biopsy, with bootstrapping to obtain confidence intervals with training set. Confirmation of accuracy on test set. Inter-observer assessment of lesion detectability on clinical B-mode US versus QUS maps. Comparison of US- and MRI-based elasticity and viscosity according to diagnostic results.