Artificial Intelligence vs. LIRADS in Diagnosing HCC on CT

A Prototype Artificial Intelligence Algorithm Versus Liver Imaging Reporting and Data System (LI-RADS) Criteria in Diagnosing Hepatocellular Carcinoma on Computed Tomography: a Randomized Trial

Liver cancer is the sixth most commonly diagnosed cancer and the fourth leading cause of cancer death worldwide. It is the 3rd most common cause of cancer death in Hong Kong. The five-year survival rates of liver cancer differ greatly with disease staging, ranging from 91.5% in early-stage to 11% in late-stage. The early and accurate diagnosis of liver cancer is paramount in improving cancer survival. Liver cancer is diagnosed radiologically via cross sectional imaging, e.g. computed tomography (CT), without the routine use of liver biopsy. However, with current internationally-recommended radiological reporting methods, up to 49% of liver lesions may be inconclusive, resulting in repeated scans and a delay in diagnosis and treatment. An artificial intelligence (AI) algorithm that that can accurately diagnosed liver cancer has been developed. Based on an interim analysis, the algorithm achieved a high diagnostic accuracy. The AI algorithm is now ready for implementation.

This study aims to prospective validate this AI algorithm in comparison with the current standard of radiological reporting in a randomized manner in the at-risk population undergoing triphasic contrast CT. This research project is totally independent and separated from the actual clinical reporting of the CT scan by the duty radiologist. The primary study outcome is the diagnostic accuracy of liver cancer, which will be unbiasedly based on a composite clinical reference standard.

Study Overview

• Status: Recruiting
• Conditions: HCC; Liver Cancer
• Intervention / Treatment: Diagnostic test: Prototype artificial intelligence algorithm; Diagnostic test: LI-RADS

Detailed Description

Liver cancer is the sixth most commonly diagnosed cancer and the fourth leading cause of cancer death worldwide. The main disease burden is found in East Asia, in which the age-standardized incidence is 26.8 and 8.7 per 100,000 in men and women respectively. In 2017, among the top 10 most common cancers in Hong Kong, liver cancer had the highest case fatality rate of 84.6%. The five-year survival rates of hepatocellular carcinoma (HCC) differ greatly with disease staging, ranging from 91.5% in <2 cm with surgical resection to 11% in >5 cm with adjacent organ involvement. The early and accurate diagnosis of HCC is paramount in improving cancer survival.

Unlike other common cancers, HCC is diagnosed by highly characteristic dynamic patterns on contrast-enhanced cross sectional imaging, without the need of pathological confirmation. The Liver Imaging Reporting and Data System (LI-RADS) was established to standardize the lexicon, interpretation and communication of radiological findings related to HCC. However, up to 49% of nodules identified in computed tomography (CT) in the at-risk population are categorized by LI-RADS as indeterminate, further delaying the establishment of diagnosis.

There are currently studies pioneering the application of artificial intelligence (AI) in the field of medical imaging. A interdisciplinary research team of clinicians, radiologists and statistical scientists, based on the clinical and radiological database of over 4,000 liver images, and have developed an AI algorithm to accurately diagnose liver cancer on CT. Based on retrospective data, an interim analysis found the AI algorithm able to achieve a diagnostic accuracy of >97% and a negative predictive value of >99%.

Can this novel prototype AI algorithm achieve a better performance in diagnosing HCC in the at-risk population when compared to LI-RADS? This question is especially relevant when the key to improved survival is early diagnosis, of which AI can potentially improve. Currently, errors in radiologist reporting are estimated to be 3-5% on a day-to-basis, equating to 40 million errors per annum worldwide. This prototype algorithm can be a solution to reduce human misinterpretation of radiological findings.

• Study Type: Interventional
• Enrollment (Anticipated): 250
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Wai-Kay Seto, MD; Phone Number: 85222553579; Email: wkseto@hku.hk
• Study Contact Backup: Name: Keith Chiu, FRCR; Phone Number: 85222553111; Email: kwhchiu@hku.hk

Study Locations

• Hong Kong
  • Hong Kong, Hong Kong
    • Recruiting
    • Department of Medicine, The University of Hong Kong, Queen Mary Hospital
    • Contact: Wai-Kay Seto, MD; Phone Number: +85222553579; Email: wkseto@hku.hk
    • Sub-Investigator: Keith Chiu, FRCR
    • Sub-Investigator: Lung Yi Mak, FRCP
    • Sub-Investigator: Man-Fung Yuen, MD

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years and older (Adult, Older Adult)
• Accepts Healthy Volunteers: No
• Genders Eligible for Study: All

Description

Inclusion Criteria:

• 1. Age >=18 years. 2. Defined as the at-risk population requiring regular liver ultrasonography surveillance. These include:

  1. Cirrhotic patients of any disease etiology,

  2. Chronic hepatitis B patients of age ≥40 years for men, age ≥50 years for women or with a family history of HCC.

     3. At least one new-onset focal liver nodule detected on liver ultrasonography.

     Exclusion Criteria:

     1. Liver nodules of <1 cm. Currently such nodules are not reported using LI-RADS criteria but are recommended for a repeat scan in 3-6 months. In patients with multiple liver nodules, the largest nodule will be assessed.
     2. Patients with contraindications for contrast CT imaging, including a history of contrast anaphylaxis and impaired renal function (glomerular filtration rate <30 ml/min).
     3. Patients with prior transarterial chemoembolization or other interventional procedures with intrahepatic injection of lipiodol. Lipiodol is extremely hyperdense on computed tomography and will preclude objective interpretation. Such patients were also excluded in the development of our prototype AI algorithm.

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Diagnostic
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: Single

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Active Comparator: Prototype AI algorithm; In-house prototype deep learning artificial intelligence algorithm	Diagnostic test: Prototype artificial intelligence algorithm; Developed by the University of Hong Kong
Placebo Comparator: LI_RADS interpretation; LI-RADS criteria will be assessed independently by two specified abdominal radiologists with at least 10 years of experience in cross-sectional abdominal imaging	Diagnostic test: LI-RADS; The Liver Imaging Reporting and Data System (LI-RADS) was established to standardize the lexicon, interpretation and communication of radiological findings related to HCC

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Diagnostic accuracy for HCC	Number of participants diagnosed with HCC using a composite clinical reference standard. A lesion will be considered positive for HCC based on histology (biopsy, surgical resection or explant) or achieving LR-5 criteria in subsequent imaging. A lesion will be considered negative for HCC if it demonstrated stability at imaging for at least 12 months, unequivocal spontaneous reduction, or disappearance in the absence of tumor treatment.	12 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Other diagnostic performance parameters for HCC	Number of participants diagnosed with HCC using a composite clinical reference standard. A lesion will be considered positive for HCC based on histology (biopsy, surgical resection or explant) or achieving LR-5 criteria in subsequent imaging. A lesion will be considered negative for HCC if it demonstrated stability at imaging for at least 12 months, unequivocal spontaneous reduction, or disappearance in the absence of tumor treatment.	12 months
Interpretation time	Mean time for AI interpretation for recruited participants	12 months
Occurrence of technical failures	Number of technical failures overall	12 months

Collaborators and Investigators

• Sponsor: The University of Hong Kong
• Collaborators: Education University of Hong Kong

Study record dates

Study Major Dates

• Study Start (Actual): March 19, 2021
• Primary Completion (Anticipated): December 31, 2025
• Study Completion (Anticipated): June 30, 2026

Study Registration Dates

• First Submitted: April 6, 2021
• First Submitted That Met QC Criteria: April 8, 2021
• First Posted (Actual): April 13, 2021

Study Record Updates

• Last Update Posted (Actual): May 18, 2022
• Last Update Submitted That Met QC Criteria: May 17, 2022
• Last Verified: May 1, 2022

More Information

Terms related to this study

• Keywords: imaging; HCC; CT; AI; deep learning; liver cancer
• Additional Relevant MeSH Terms: Digestive System Diseases; Neoplasms; Neoplasms by Site; Digestive System Neoplasms; Liver Diseases; Liver Neoplasms
• Other Study ID Numbers: UW 20-445

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: No
• IPD Plan Description: Available to bona fide researchers who approach to principal investigator

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No