Non-contrast abbreviated MRI for the detection of hepatocellular carcinoma in patients with Liver Imaging Reporting and Data System LR-3 and LR-4 observations in MRI

Abstract

Background and aims

With ultrasound sensitivity limited in hepatocellular carcinoma (HCC) surveillance and few prospective studies on non-contrast abbreviated MRI (NC-AMRI), this study aimed to assess its diagnostic performance in detecting HCC.

Methods

This prospective study involved cirrhotic patients with contrast-enhanced MRI (CE-MRI) Liver Imaging Reporting and Data System (LI-RADS) LR-3 and LR-4 observations detected during HCC surveillance. Patients underwent average 3 complete CE-MRI rounds at 3-6 months interval, with approximately 12-month follow-up. NC-AMRI included diffusion-weighted (DWI), T2-weighted imaging (T2WI), and T1-weighted imaging (T1WI). NC-AMRI protocol images were analysed for diagnostic performance, with subgroup analyses. CE-MRI and NC-AMRI images were independently reviewed by 2 experienced radiologists, with inter-reader agreement assessed with Kappa coefficient. The reference standard was the American Association for the Study of Liver Diseases-defined presence of arterial hypervascularity and washout during the portal-venous or delayed phases on CE-MRI.

Results

In 166 CE-MRI follow-ups of 63 patients (median age: 63 years; 60.3% male, 39.7% female), 12 patients developed HCC, with average size of 19.6 mm. The NC-AMRI (DWI + T2WI + T1WI) showed 91.7% sensitivity (95%CI, 61.5-99.8) and 91.6% specificity (95%CI, 86.0-95.4), area under receiver operating characteristic 0.92 (95%CI, 0.83-1.00). Across different Body Mass Index categories, lesion size, Child-Turcotte-Pugh classes, Albumin-Bilirubin (ALBI) grades, and Model for End-Stage Liver Disease classes, sensitivity remained consistent. However, specificity differed significantly between ALBI grade 1 and 2 (86.7% vs. 98.4%, P = .010), and between viral and non-viral cirrhosis (93.8% vs. 80.8%, P = .010).

Conclusions

NC-AMRI proved clinically feasible, and exhibits high diagnostic performance in HCC detection.

Advances in knowledge

This study highlights efficacy of NC-AMRI in detecting HCC among cirrhotic patients with LR-3 and LR-4 observations, representing significant progress in HCC surveillance.

Introduction

Liver cancer, ranking sixth in global cancer diagnoses and third in cancer-related deaths in 2020, poses a significant worldwide burden, with hepatocellular carcinoma (HCC) constituting over 80% of liver cancers and contributing to this growing challenge.¹ About 90% of HCC cases are linked to known risk factors, prompting the adoption of international guidelines to facilitate HCC screening among susceptible populations.^2–4 Patients with underlying cirrhosis, regardless of its cause, are the primary targets for HCC screening, as over 80% of HCC cases occur in patients with cirrhosis.⁵

HCC screening enhanced early tumour detection and improved overall survival rates.⁶ A meta-analysis of 59 studies indicated that HCC surveillance correlates with increased rates of early-stage detection, receipt of curative treatment, and improved overall survival [Hazard ratio (HR) 0.67, 95% CI, 0.61-0.72].⁷ Despite guidelines recommending biannual ultrasound (US) for HCC surveillance due to its non-invasiveness and accessibility, limitations persist in the critical role of liver imaging for the detection and diagnosis of HCC.^2–4,8 US efficacy heavily depends on operator skill and is challenging in severely obese patients. A meta-analysis of 41 studies revealed a low sensitivity of only 51.6% (95% CI, 43.3-60.5%) for early-stage HCC detection with US alone.⁹ Approximately 20% of patients, particularly those with obesity and non-viral cirrhosis, experienced visualization limitations with Ultrasound Liver Imaging Reporting and Data System (US LI-RADS) scores B or C, with the latter associated with reduced sensitivity for early HCC detection.^10,11 For cases of limited US visualization, the American Association for the Study of Liver Diseases (AASLD) recommends surveillance contrast-enhanced MRI or multiphase CT for comprehensive assessment.² The European Association for the Study of the Liver (EASL) suggests that in inconclusive imaging cases, particularly for lesions smaller than 2 cm, a biopsy is recommended for a definitive diagnosis.³ However, liver biopsy, despite its diagnostic utility, poses invasiveness, and obtaining adequate biopsy samples from lesions smaller than 1 cm remains challenging. Therefore, efficient HCC screening is pivotal for early tumor detection and potentially improved overall survival rates.

CT and MRI are essential in managing HCC, necessitating standardized imaging techniques and interpretations for accurate diagnosis and effective clinician communication, with contrast-enhanced MRI (CE-MRI) surpassing contrast-enhanced CT in HCC diagnosis due to its superior tumor detection efficacy, differential diagnosis capabilities, tissue contrast, and lack of radiation risk.^12,13 LI-RADS provides a comprehensive framework for reporting and interpretation, aiming for high specificity and positive predictive value (PPV) in diagnosis, tailored for high-risk populations where LR-3 and LR-4 observations indicate HCC risk.^12,14 Despite its excellent efficacy, CE-MRI faces challenges such as limited accessibility, complex technology, high costs due to contrast and the need for skilled personnel or additional staff, prolonged examination time, and contrast-related side effects.¹⁵ To address these issues, abbreviated MRI (AMRI) using essential sequences has gained acceptance, providing similar performance to conventional CE-MRI while saving time and reducing costs by eliminating redundant sequences.¹⁶

In our study, we focused on patients with CE-MRI LI-RADS LR-3 and LR-4 observations, a decision driven by factors such as the limited availability and higher cost associated with abbreviated MRI (AMRI)-based surveillance for hepatocellular carcinoma (HCC). This targeted approach allows for personalized risk stratification-based strategies, particularly in populations with the highest incidence of HCC.¹⁷ Notably, LR-3 lesions have been associated with HCC incidences ranging from 1.2% to 12.5% at 12 months, while LR-4 lesions exhibit incidences ranging from 30.8% to 44.0% at 12 months, as highlighted in a systematic review of 13 studies.¹⁸ Given that HCC surveillance is typically warranted in populations with an annual incidence of at least 1.5%,¹⁹ our study aims to address health economic concerns by recommending the selective use of non-contrast abbreviated MRI (NC-AMRI) for semiannual surveillance in patients with the highest HCC risk.

Several studies have reported various AMRI protocols, including NC-AMRI protocols.^20–32 A recent meta-analysis highlighted the efficacy of NC-AMRI in HCC detection, showing a sensitivity of 83% (95%CI: 79-87%) and specificity of 91% (95%CI: 88-93%), comparable to contrast-enhanced abbreviated MRI.³³ Most previous studies, predominantly retrospective, examined different AMRI protocols. Only two prospective studies reported NC-AMRI performance in HCC detection.^24,28 Our recent meta-analysis indicated that the protocol, combining all three sequences, including Diffusion-weighted (DWI) and T2-weighted imaging (T2WI) with T1-weighted imaging (T1WI), achieved a specificity of 87% and the highest sensitivity of 87%.³³ Prospective data on the HCC detection capability of NC-AMRI are still limited. Therefore, our study aimed to prospectively investigate the diagnostic performance of NC-AMRI for detecting HCC in patients with LI-RADS LR-3 and LR-4 observations on CE-MRI.

Methods

Study design

This study is a single-centred prospective study conducted at the King Chulalongkorn Memorial Hospital in Bangkok, Thailand.

Ethical considerations

The study was approved by the Institutional Review Board of the Faculty of Medicine, Chulalongkorn University (IRB Number—0412/66). Written informed consent was obtained from all participants by investigators.

Patients recruitment and selection

Patients attending clinic visits for HCC surveillance were recruited from the Center of Excellence for Innovation and Endoscopy in Gastrointestinal Oncology, Faculty of Medicine, Chulalongkorn University and out-patient department (OPD) of the Division of Gastroenterology and Hepatology, King Chulalongkorn Memorial Hospital (KCMH) in Bangkok, Thailand. Among them, patients undergoing HCC surveillance were recruited based on their prior evaluation with full contrast-enhanced MRI (CE-MRI) within 6 months before enrolment. This evaluation was prompted by suspicious lesion(s)/observation(s) detected by US, necessitating confirmation and characterization through CE-MRI, ensuring they were not indicative of HCC. They were assessed for the following inclusion criteria: (1) age at enrolment ≥18 years; (2) absence of previous HCC history; and (3) established diagnosis of cirrhosis pathologically and/or radiologically. Participants were excluded if they had one of the following exclusion criteria: (1) patients with pregnancy; (2) patients who are categorized as LR-5 by MRI LI-RADS; (3) patients with estimated glomerular filtration rate (eGFR) <30 mL/min/1.73 m²; and (d) patients who were absolutely contraindicated for MRI such as a cardiac pacemaker or implanted cardiac defibrillator.

Data collection

Patients’ demographic and clinical characteristics, including age, sex, body mass index (BMI), and aetiologies of chronic liver disease were collected. Laboratory data of interest, encompassing platelet count, creatinine, aspartate aminotransferase (AST), alanine aminotransferase (ALT), total bilirubin, albumin, international normalized ratio (INR), and alpha-fetoprotein (AFP), were meticulously documented. Additionally, calculations for the Child-Turcotte-Pugh (CTP), Model for End-Stage Liver Disease (MELD), and Albumin-Bilirubin (ALBI) scores were conducted. Patients have undergone both MRI and blood tests within a maximum interval of 2-4 weeks to ensure consistency in the assessment of liver imaging and tumour markers.

CE-MRI and NC-AMRI protocols

The MRI was performed on a 1.5-T scanner with an 8-channel phased array torso coil (MAGNETOM Sola with BioMatrix; Siemens Healthineers, Forchheim, Germany), including T1WI, DWI, T2WI, and contrast-enhanced T1WI using gadoxetic acid including pre-contrast, arterial, portal, transitional, and hepatobiliary (20 min) phases. All sequences were carried out axially. The NC-AMRI protocol comprised the following sequences: T1-weighted gradient in-phase (IP) and opposed-phase (OP) imaging conducted via breath-hold dual gradient-echo, with a slice thickness of 5 mm and a 1 mm gap, a field of view (FOV) of 380 mm, a repetition time (TR) of 143 ms, an in-phased echo time (TE) of 4.87 ms, an opposed-phased TE of 2.25 ms, a flip angle of 70 degrees, a matrix size of 189 × 320, and a total scan time of about 1.24 min, analysing only IP and OP images; T2-weighted imaging with fat saturation (T2WI FS) performed using a respiratory-triggered fast spin echo method with TR 2500 ms, TE 78 ms, flip angle 137 degrees, FOV 360 mm, matrix size 272 × 272 mm, slice thickness 3.5 mm with 0.5 mm gap; and DWI conducted using a respiratory-triggered spin echo—echo planar imaging scheme with TR 3500 ms, TE 61 ms, matrix size 134 × 108 mm, flip angle 90 degrees, FOV 380 mm, slice thickness 5 mm with 1 mm gap, using b-values of 50, 500, and 1000 s/mm² and with analysis of computed apparent diffusion coefficient (ADC) maps. The included images of NC-AMRI were extracted for analysis separately.

MRI images analysis

All images underwent anonymization and were shuffled into a random order for evaluation. Two experienced abdominal diagnostic radiologists, MS with 5 years of post-training experience and NT with 10 years of post-training experience from the Division of Diagnostic Radiology, Department of Radiology at the Faculty of Medicine, King Chulalongkorn Memorial Hospital, independently reviewed and interpreted the CE-MRI and NC-AMRI images using the MRI Liver Imaging Reporting and Data System (LI-RADS^®).¹⁴ In cases of asynchronous diagnoses, the 2 radiologists collaborated and reached a consensus report through discussion. Additionally, inter-observer variability was assessed using the kappa value. They interpreted the MRIs independently in a blinded manner, without access to clinical, laboratory data, or ultrasound reports. To minimize memory bias, the evaluations of CE-MRI and NC-AMRI images were conducted at separate intervals, with a washout period of 4-6 weeks between evaluations.

All NC-AMRI observations were classified into three categories: positive (category 3), subthreshold (category 2), and negative (category 1). The positive criterion for HCC diagnosis was assigned to category 3, while categories 1 and 2 were considered negative. NC-AMRI category 3 (positive) included observation(s) ≥10 mm with “not definitive benign features” characterized by diffusion restriction and mild to moderate T2 hyperintensity, with T1WI revealing fat in the lesion(s), threshold growth of previously noted observations, or the presence of new thrombus in the vein. Category 2 (subthreshold) was assigned to observations/lesions < 10 mm showing “not definitely benign” features. Category 1 (negative) was assigned when no lesion/observation was observed or when lesion(s)/observation(s) exhibited ‘definitely benign’ features, defined as marked T2 hyperintensity (Table S1). For CE-MRI, LI-RADS was adapted to interpret liver observations (Figure S1). Patients with LI-RADS LR-5 lesions were considered positive for HCC diagnosis. CE-MRIs were deemed negative when no lesion was found, or when all observations were scored as LI-RADS LR-1 or LR-2. MRIs with lesions assessed as LI-RADS LR-3 or LR-4 were also considered negative.

Reference standard

The reference standard for this study involved classifying each CE-MRI as either positive or negative for HCC based on the identification of typical radiological hallmarks. These assessments followed the operational definition of HCC outlined in the AASLD guidelines.² Diagnosis of HCC on full MRI relies on observing arterial hypervascularity in the arterial phase and washout in the portal venous or transitional phases, in addition to compatible lesion size and enhancement patterns.

Surveillance period and screening interval

Patients classified as LR-3 by full CE-MRI assessment with high-risk classification underwent surveillance every 6 months, while those with super-high-risk classification had surveillance every 4 months. Patients classified as LR-4 by full CE-MRI assessment underwent surveillance every 3 months. High-risk and super-high-risk classification followed the proposed Consensus-Based Clinical Practice Guidelines for the identification of high-risk populations for HCC by the Japan Society of Hepatology (JSH). The definitions are as follows: The super-high-risk population includes patients with hepatitis B-related liver cirrhosis and hepatitis C-related liver cirrhosis. The high-risk population comprises individuals with chronic hepatitis B, chronic hepatitis C, and liver cirrhosis caused by factors other than hepatitis B virus (HBV) or hepatitis C virus (HCV). (Table S2).³⁴ An overview of the study is presented in Figure 1 with a flowchart. The study was a single-centered prospective study, with patients prospectively enrolled. The follow-up period for each patient lasted approximately 12 months or until a diagnosis of HCC was established, whichever occurred first. If patients received a diagnosis of HCC via full gadoxetic acid-enhanced MRI, their participation in the study was terminated, and they were promptly referred for treatment. If no evidence of HCC was found on a full gadoxetic acid-enhanced MRI, patients resumed the surveillance schedule based on the MRI LI-RADS classifications.

Figure 1.

Flowchart providing the overview of the study design. *Repeating DWI, T2WI, and pre-T1WI is unnecessary. Immediate additional sequences of post-contrast enhanced T1W can be done as a recalled second MRI.

Statistical analysis

Baseline characteristics were summarized by presenting continuous variables as mean and standard deviation (SD) or median (range) as appropriate. Categorical variables were described using frequency and percentage. Kappa coefficient (κ) was used to evaluate inter-reader agreement. The level of agreement was interpreted as poor (κ  <  0), slight (0 ≤ κ  ≤  0.2), fair (0.2 < κ  ≤  0.4), moderate (0.4 < κ  ≤  0.6), substantial (0.6 < κ  ≤  0.8), or almost perfect (κ > 0.8). The diagnostic performance of NC-AMRI protocols was compared to the reference standard. Sensitivity, specificity, PPV, negative predictive value (NPV), and accuracy of NC-AMRI (DWI with T2WI FS and T1WI) and other non-contrast MRI sequences (DWI with T2WI FS; T2WI FS alone; and DWI alone) were calculated.

The subgroup analyses were done based on tumor size (≤ 10 mm, 11-20 mm, and > 20 mm), patients’ BMI (≤ 23 kg/m², 23-30 kg/m², and > 30 kg/m²), CTP class (CTP A and B), ALBI Grade (1 and 2), MELD score class (≤ 9 and 10-19), and etiologies of chronic liver disease (Chronic Hepatitis B and C-related cirrhosis categorized as “viral cirrhosis”; and cirrhosis related to Alcoholic Liver Disease and metabolic dysfunction-associated steatotic liver disease categorized as “non-viral cirrhosis”). Statistical analyses were performed using STATA, version 17.0 (StataCorp LLC, College Station, TX, United States) and for all tests, P-values <.05 were considered statistically significant.

Results

Baseline characteristics of participants

Table 1 outlines baseline characteristics of the 63 enrolled patients. The cohort underwent a total of 166 follow-up MRI examinations, with an average of three MRI examinations per patient. Among them, 38 were males (60.3%) and 25 females (39.7%), with a median age of 63.0 years (IQR: 52.6-69.8). The findings also depicted the distribution of patients according to BMI, revealing that 28.6% (n = 18) had a BMI ≤ 23 kg/m², 60.3% (n = 38) fell within the range of 23-30 kg/m², and 11.1% (n = 7) had a BMI > 30 kg/m². The cohort primarily had hepatitis B (n = 31, 49.2%), followed by hepatitis C (n = 20, 31.7%), alcoholic liver disease (ALD) (n = 7, 11.1%), and metabolic dysfunction-associated steatotic liver disease (MASLD) (n = 5, 7.9%). All patients had cirrhosis, with MRI LI-RADS showing LR-3 in 77 lesions (91.7%) and LR-4 in 7 lesions (8.3%), with 84 observed lesions, sized at a median of 10 mm (IQR: 8-14 mm). In the study, 12 patients (19.0%) were classified as high-risk, while 51 patients (81.0%) were categorized as super-high-risk. The median AFP level was 2.84 ng/mL (IQR: 1.80-4.56). CTP class distribution included 56 cases in class A (88.9%) and 7 in class B (11.1%), while MELD scores were ≤ 9 in 31 patients (49.2%) and ranged from 10 to 19 in 32 patients (50.8%). ALBI grade 1 was observed in 35 patients (55.6%) and ALBI grade 2 in 28 patients (44.4%). Clinical and laboratory data from 166 MRI examinations during follow-up visits were also provided in Table S3.

Table 1.

Baseline characteristics of patients at the time of enrolment.

	n = 63
Number of follow-up MRI examinations	166
Average number of MRI examinations per patient	3
Age (years) [Median (IQR)]	63.0 (52.6-69.8)
Sex (n, %)
Male	38, 60.3
Female	25, 39.7
Body mass index (BMI) (n, %), kg/m2
≤23	18, 28.6
23-30	38, 60.3
>30	7, 11.1
Underlying liver disease (n, %)
Hepatitis B	31, 49.2
Hepatitis C	20, 31.7
Alcoholic liver disease (ALD)	7, 11.1
Metabolic dysfunction-associated steatotic liver disease (MASLD)	5, 7.9
Diagnosis of cirrhosis (n, %)	63, 100.0
MRI LI-RADS observations
LR-3	77, 91.7
LR-4	7, 8.3
Risk classifications (n, %)
High-risk patients	12, 19.0
Super-high-risk patients	51, 81.0
Number of lesions	84
Size of lesions (mm) [Median (IQR)]	10 (8-14)
Laboratory findings
Total bilirubin (mg/dL) [Median (IQR)]	0.97 (0.63-1.41)
Serum albumin (g/dL) [Median (IQR)]	4.0 (3.6-4.3)
International normalized ratio (INR) [Median (IQR)]	1.14 (1.08-1.26)
Aspartate transaminase (AST) (IU/mL) [Median (IQR)]	30 (25-40)
Alanine transaminase (ALT) (IU/mL) [Median (IQR)]	24 (18-36)
Creatinine (mg/dL) [Median (IQR)]	0.86 (0.71-1.14)
Platelets count (×109/L) [Median (IQR)]	133 (87-196)
Alpha-fetoprotein (ng/mL) [Median (IQR)]	2.84 (1.80-4.56)
Child-Turcotte-Pugh (CTP) class (n, %)
A	56, 88.9
B	7, 11.1
Model for end-stage liver disease (MELD) score (n, %)
≤9	31, 49.2
10-19	32, 50.8
ALBI (Albumin-Bilirubin) grade (n, %)
ALBI grade 1	35, 55.6
ALBI grade 2	28, 44.4

Abbreviations: IQR = interquartile range; LI-RADS = Liver Imaging Reporting and Data System.

Interobserver variability of non-contrast and abbreviated MRI protocols

Table 2 illustrates the interobserver variability of NC-AMRI protocols between two radiologists, expressed as κ values. The reference standard CE-MRI, achieved a κ value of 0.96, indicating almost perfect interobserver agreement. The agreement between two radiologists was assessed for various non-contrast and abbreviated MRI sequences, including DWI alone, T2WI FS alone, DWI + T2WI FS, and NC-AMRI consisting of DWI, T2WI FS, and T1WI. The results demonstrate high agreement levels, with κ values ranging from 0.87 to 0.96 across different sequences. Specifically, the highest level of agreement was observed for DWI alone, with a κ value of 0.96, followed closely by NC-AMRI and DWI + T2WI FS, both yielding κ values of 0.89. T2WI FS alone also demonstrated almost perfect agreement but had the lowest level of agreement, with a κ value of 0.87.

Table 2.

Interobserver variability of NC-AMRI protocols (κ value).

Sequences	Kappa (κ value)
CE-MRI	0.96
NC-AMRI (DWI + T2WI FS + T1WI)	0.89
DWI alone	0.96
T2WI FS alone	0.87
DWI + T2WI FS	0.89

Abbreviations: CE-MRI = contrast-enhanced MRI; DWI = diffusion-weighted Imaging; NC-AMRI = non-contrast abbreviated MRI; T2WI FS = T2-weighted Imaging with fat saturation; T1WI = T1-weighted Imaging.

Diagnostic performance of non-contrast and abbreviated MRI for HCC detection

Table 3 summarizes the diagnostic performance of NC-AMRI sequences on a per-patient basis. The sensitivity, specificity, PPV, NPV, area under the receiver operating characteristic curve (AUROC), and accuracy are presented along with their respective 95% CI. The NC-AMRI protocol comprising DWI, T2WI FS, and T1WI demonstrated a sensitivity of 91.7% (95%CI: 61.5-99.8) and specificity of 91.6% (95%CI: 86.0-95.4), with an AUROC of 0.92 (95%CI: 0.83-1.00). DWI alone and DWI + T2WI FS sequences exhibited similar sensitivities of 91.7% (95%CI: 61.5-99.8) but slightly lower specificities at 89.0% (95%CI: 82.9-93.4) and 89.6% (95%CI: 83.7-93.9), respectively. T2WI FS alone demonstrated a perfect sensitivity of 100.0% (95%CI: 73.5-100.0) but lower specificity at 61.7% (95%CI: 53.5-69.4).

Table 3.

Diagnostic performance of non-contrast abbreviated MRI sequences (per-patient and per-lesion basis).

	Sensitivity	Specificity	PPV	NPV	AUROC	Accuracy
	[% (95%CI)]	[% (95%CI)]	[% (95%CI)]	[% (95%CI)]	(95%CI)	[% (95%CI)]
Per-patient basis
NC-AMRI (DWI + T2WI FS + T1WI)	91.7 (61.5-99.8)	91.6 (86.0-95.4)	45.8 (25.6-67.2)	99.3 (96.1-100.0)	0.92 (0.83-1.00)	91.5 (86.2-95.3)
DWI alone	91.7 (61.5-99.8)	89.0 (82.9-93.4)	39.3 (21.5-59.4)	99.3 (96.0-100.0)	0.90 (0.82-0.99)	89.2 (83.4-93.4)
T2WI FS alone	100.0 (73.5-100.0)	61.7 (53.5-69.4)	16.9 (9.0-27.7)	100.0 (96.2-100.0)	0.81 (0.77-0.85)	64.5 (56.7-71.7)
DWI + T2WI FS	91.7 (61.5-99.8)	89.6 (83.7-93.9)	40.7 (22.4-61.2)	99.3 (96.1-100.0)	0.91 (0.82-0.99)	89.8 (84.1-93.9)
Per-lesion basis
NC-AMRI (DWI + T2WI FS + T1WI)	91.7 (61.5-99.8)	94.2 (90.4-96.9)	45.8 (25.6-67.2)	99.5 (97.4-100.0)	0.93 (0.85-1.00)	94.1 (90.3-96.7)
DWI alone	91.7 (61.5-99.8)	92.5 (88.2-95.6)	39.3 (21.5-59.4)	99.5 (97.4-100.0)	0.92 (0.84-1.00)	92.4 (88.3-95.5)
T2WI FS alone	100.0 (73.5-100.0)	65.9 (59.4-72.1)	13.5 (7.2-22.4)	100.0 (97.6-100.0)	0.83 (0.80-0.86)	67.6 (61.3-73.5)
DWI + T2WI FS	91.7 (61.5-99.8)	92.9 (88.8-95.9)	40.7 (22.4-61.2)	99.5 (97.4-100.0)	0.92 (0.84-1.00)	92.9 (88.8-95.8)

Abbreviations: AUROC = area under the receiver operating characteristic; DWI = diffusion-weighted imaging; NC-AMRI = non-contrast abbreviated MRI; NPV = negative predictive value; PPV = positive predictive value; T1WI = T1-weighted imaging; T2WI FS = T2-weighted imaging with fat saturation.

Table 3 also presents the diagnostic performance of NC-AMRI sequences on a per-lesion basis. The NC-AMRI protocol consisting of DWI, T2WI FS, and T1WI demonstrated a sensitivity of 91.7% (95%CI: 61.5-99.8) and specificity of 94.2% (95%CI: 90.4-96.9), with an AUROC of 0.93 (95%CI: 0.85-1.00). DWI alone exhibited similar sensitivity but slightly lower specificity at 92.5% (95%CI: 88.2-95.6). T2WI FS alone showed perfect sensitivity at 100.0% (95%CI: 73.5-100.0) but lower specificity at 65.9% (95% CI: 59.4-72.1). DWI + T2WI FS had a sensitivity of 91.7% (95%CI: 61.5-99.8) and specificity of 92.9% (95%CI: 88.8-95.9).

False negative and false positive NC-AMRI examinations

Flowchart detailing patient pathways through the study was described in Figure 2. During the follow-up visits, false positive and false negative cases were identified in the MRI examinations conducted using NC-AMRI, which comprised DWI + T2WI FS + T1WI, with reference to CE-MRI. One false negative and 12 false positive NC-MRI examinations were detected during the follow-up visits. The only false negative case presented mild hyperintensity in T2WI FS and lacked signal intensities in DWI and T1WI, resulting in a negative NC-AMRI report (Figure 3). Among the 12 false positive examinations in 8 patients, all NC-AMRI examinations displayed ‘not definitely benign’ features (Figure S2). However, they did not meet the reference standard criteria of CE-MRI, characterized by arterial hypervascularity in the arterial phase and washout in the portal venous or delayed phases. Specifically, one patient initially categorized as LR-3 developed HCC after 3 rounds of follow-up MRI examinations within 9.7 months (Figure 4), while another patient classified as LR-4 exhibited “not definitely benign” features in the NC-AMRI examination and developed HCC after 4.8 months (Figure 5). Additionally, two patients with LR-3 and three patients with LR-4 remained negative according to CE-MRI after 3 rounds of follow-up visits within the 12-month study period. Furthermore, one patient initially categorized as LR-3 transitioned to LR-4 within 8.4 months, displaying ‘not definitely benign’ features without HCC characteristics according to CE-MRI.

Figure 2.

Flowchart depicting patient pathways through the study.

Figure 3.

A 47-year-old man with a 1.0-cm nodule in the hepatic segment V showed a false-negative case on NC-AMRI. On T2-weighted Imaging (T2WI), faint hyperintensity was observed in segment V of the liver (A, arrow). DWI (Diffusion-weighted Imaging) (b = 1000 s/mm²) (B) displayed no hyperintense signal and no lesion detected on ADC mapping (C), while there was slightly hypointense signal on T1-weighted imaging with fat suppression (D). After gadoxetic acid injection, the mass exhibited arterial enhancing lesion (arrow, E) and washout enhancement on the portal venous phase (arrow, F). (Informed consent was obtained from the patient included in this figure for the use of MRI images).

Figure 4.

A 50-year-old man presented with HCC in segment VI, exhibiting ‘not definitely benign’ features on NC-AMRI. Diffusion restriction was evident on DWI (Diffusion-weighted Imaging) (b = 500 s/mm²) (A), accompanied by high signal intensity on T1-weighted imaging (B). A mild hyperintense signal was observed on T2 Single-shot Turbo spin (C), confirming the presence of HCC on NC-AMRI. However, following gadoxetic acid injection, the mass displayed a slight arterial enhancing lesion (arrow, D) but lacked washout on the portal venous phase (arrow, E). Subsequent gadoxetic acid-enhanced MRI, after 3 rounds within 9.7 months, revealed typical features of HCC with prominent arterial hyperenhancement (arrow, F) and venous washout (arrow, G). (Informed consent was obtained from the patient included in this figure for the use of MRI images).

Figure 5.

A 70-year-old man presented with HCC in segment VI. Diffusion restriction was evident on DWI (Diffusion-weighted Imaging) (b = 1000 s/mm²) (circle) (A), along with mild to moderate hyperintensity on T2-weighted imaging (B), and high signal intensity on T1-weighted imaging in segment VI of the liver (circle) (C). However, after gadoxetic acid injection, the mass displayed a prominent arterial enhancing lesion (circle, D) but lacked washout on the portal venous phase (E), resulting in CE-MRI LI-RADS LR-4 observation and a ‘false positive’ for NC-AMRI. After 4.8 months, subsequent gadoxetic acid-enhanced MRI revealed typical features of HCC, including more prominent arterial hyperenhancement (circle, F), and typical venous washout (circle, G). (Informed consent was obtained from the patient included in this figure for the use of MRI images).

Subgroup analysis of sensitivity and specificity of NC-AMRI protocol

Table 4 illustrates sensitivity and specificity in a subgroup analysis based on several variables: BMI, lesion size, CTP Class, ALBI Grade, MELD Class, and underlying aetiology. Across different BMI categories, sensitivity showed no significant differences (P = .677), and specificity remained consistent (P = .928). Lesion size also did not significantly affect sensitivity (P = .677) or specificity (P = .169). Similarly, no significant differences in sensitivity were observed between CTP classes A and B (P = .640), nor between ALBI grades 1 and 2 (P = .378). However, specificity differed significantly between ALBI grade 1 and 2 (86.7%, 95%CI: 77.9-92.9 vs. 98.4%, 95%CI: 91.6-100.0, P = .010). Additionally, sensitivity was comparable across MELD classes (P = .217), with no significant difference in specificity (P = .813). Notably, sensitivity was consistent across underlying aetiologies (P = .640), but a significant difference in specificity (P = .030) between viral and non-viral cirrhosis (93.8%, 95%CI: 88.1-97.3 vs. 80.8%, 95%CI: 60.6-93.4, P = .010).

Table 4.

Subgroup analysis of sensitivity and specificity.

Variables	Subgroup	Sensitivity (%) (95%CI)	P-value	Specificity (%) (95%CI)	P-value
Body mass index (BMI)a	≤23.0 kg/m2	100.0 (15.8-100.0)	.677	91.3 (79.2-97.6)	.928
	23.0-30.00 kg/m2	85.7 (42.1-99.6)		92.1 (84.5-96.8)
	>30.00 kg/m2	100.0 (29.2-100.0)		89.5 (0.88-98.7)
Size of lesionsa	≤10 mm	100.0 (15.8-100.0)	.677	96.2 (91.9-98.6)	.169
	11-20 mm	85.7 (42.1-99.6)		90.0 (79.5-96.2)
	>20 mm	100.0 (29.2-100.0)		88.9 (51.8-99.7)
Child-Turcotte-Pugh (CTP) classb	A	90.0 (55.5-99.7)	.640	91.4 (85.4-95.5)	.795
	B	100.0 (15.8-100.0)		93.3 (68.1-99.8)
Albumin-Bilirubin (ALBI) gradeb	1	85.7 (42.1-99.6)	.378	86.7 (77.9-92.9)	.010
	2	100 (47.8-100.0)		98.4 (91.6-100.0)
Model for end-stage liver disease (MELD) classb	≤9	80.0 (28.4-99.5)	.217	91.1 (83.2-96.1)	.813
	10-19	100.0 (59.0-100.0)		92.2 (82.7-97.4)
Underlying aetiologyb	Viral cirrhosis	90.0 (55.5-99.7)	.640	93.8 (88.1-97.3)	.030
	Non-viral cirrhosis	100.0 (15.8-100.0)		80.8 (60.6-93.4)

^aChi-square test for comparison.

^bMcNemar test for comparison.

Characteristics of patients diagnosed with HCC

The characteristics of 12 patients with HCC are summarized in Table S4. They had a mean age of 64.8 years, with the majority being male (66.7%) and an average BMI of 26.05 kg/m². The most common underlying liver diseases were hepatitis B (50.0%), hepatitis C (33.3%), and ALD (16.7%). HCC characteristics included a mean (± standard deviation) lesion size of 19.6 mm (± 9.1 mm) and a median AFP level of 3.23 ng/mL (IQR: 2.36-17.4). Among patients diagnosed with HCC, 83.3% (n = 10) belonged to the super-high-risk category, while 16.7% (n = 2) were classified as high-risk. Most patients were classified as CTP class A (83.3%), with the rest as CTP class B (16.7%). Regarding MELD scores, 41.7% had scores of ≤ 9, while 58.3% had scores ranging from 10 to 19. ALBI grade 1 was predominant in 58.33% of patients, whereas 41.7% were classified as ALBI grade 2. BCLC Staging showed that the majority of patients were in stage 0 (75.0%), with the remaining 25.0% classified as stage A.

Discussion

In our study, NC-AMRI, including DWI, T2WI FS, and T1WI sequences, demonstrated a per-lesion sensitivity of 91.7% and specificity of 94.2%. This performance surpassed that of the routine surveillance tool of US with AFP, which exhibited 74.1% sensitivity and 87.9% specificity as per a recent meta-analysis.⁹ Our findings also paralleled those of a previous study where a NC-AMRI protocol, incorporating DWI and T2WI FS along with AFP, achieved a sensitivity of 92.8%.³¹ Sensitivities and specificities observed both on a per-lesion and per-patient basis in our study were also similar to a recent systematic review and meta-analysis conducted by our group, demonstrating pooled sensitivities of 83% and pooled specificities of 91% for NC-AMRI protocols.³³ Therefore, our study highlighted the robust diagnostic performance of NC-AMRI in detecting HCC, contributing valuable insights to existing clinical knowledge.

Concerning each NC-AMRI protocol, NC-AMRI, consisting of DWI, T2WI, and T1WI, demonstrated the highest AUROC of 0.92 and 0.93 on a per-lesion and per-patient basis, respectively, compared to other protocols. This outcome indicated the complementary nature of these three sequences. DWI exhibited excellence in HCC detection, as evidenced by a study evaluating a DWI-only strategy for surveillance, revealing sensitivity and specificity of 83% and 98%, respectively.²⁴ T2WI facilitates both lesion detection and characterization, with mild to moderate T2 hyperintensity serving as an ancillary feature for HCC detection. This feature, defined as “not definitely benign,” could make valuable for liver lesion characterization when combined with DWI.¹⁶ Moderately T2WI sequences appeared more suitable for surveillance due to better lesion conspicuity and signal-to-noise ratio, although further investigations are needed to establish the optimal T2 pulse sequence for non-contrast MRI.³⁵ T1WI contributes to assessing intralesional fat, another ancillary feature supporting HCC detection, and can be acquired in a single breath-hold using most current MRI systems.¹⁶ The advantage of considering these three sequences was evident in the MAGNUS-HCC clinical trial, which assessed the diagnostic performance of annual NC-AMRI (T2WI, DWI, and T1-weighted in/opposed images) for HCC surveillance in cirrhotic patients compared to biannual US. NC-AMRI exhibited higher per-patient sensitivity (71.0%) and positive predictive value (61.1%) than US (sensitivity, 45.2%; PPV, 33.3%).³⁶

The subgroup analysis revealed consistently high diagnostic performance of NC-AMRI across different BMI groups. This finding suggested that NC-AMRI may be particularly beneficial for patients with high BMI who are unsuitable for HCC surveillance by ultrasound. Previous studies have indicated that a BMI of ≥30 kg/m² is significantly associated with ultrasound surveillance failure,³⁷ while a BMI ≥ 25 kg/m² is associated with inadequate echogenic window, leading to HCC detection failure by ultrasound.³⁸ This underscored the potential of NC-AMRI to address the limitations posed by BMI in ultrasound-based surveillance for HCC. In the subgroup analysis across lesion sizes, the sensitivity was lowest at 85.7% for lesions sized 11-20 mm, but there were no significant differences observed among the three size categories (≤10 mm, 11-20 mm, and >20 mm). Similarly, high specificity ranging from 88.9% to 96.2% was observed across different sizes, with no significant differences noted. In our study, 12 patients developed HCC with a mean lesion size of 19.6 mm and a median AFP of 3.23 ng/mL. The majority of patients were in BCLC stage 0 (n = 9), with the remaining 3 patients classified as stage A. Therefore, NC-AMRI proved beneficial for cirrhotic patients by enabling the detection of HCC at an early stage for curative treatment, which may ultimately improve patient outcomes. Compared with ultrasound, which exhibited only 51.6% sensitivity for early-stage HCC according to a meta-analysis,⁹ NC-AMRI could be highly beneficial for early HCC detection.

In our study, while the sensitivity of NC-AMI did not differ between viral and non-viral cirrhosis aetiologies, the specificity was significantly lower in non-viral cirrhosis at 80.8% compared to 93.8% in viral cirrhosis. However, the diagnostic performance still surpassed that of ultrasound in this patient population, where US visualization was compromised.¹⁰ There are concerns regarding poor visualization, which was more likely to occur in patients with obesity and non-viral aetiologies of cirrhosis, increasingly common populations undergoing HCC surveillance.^39,40 Patients with suboptimal visualization are at high risk for surveillance failure and may require alternative HCC surveillance strategies. This association partly explains why surveillance failure is less common in patients with non-cirrhotic hepatitis B infection but increases among those with cirrhosis.⁴¹ Therefore, NC-AMRI may be a valuable surveillance tool for patients with severe limitations in ultrasound visualization, particularly among those with obesity and non-viral aetiologies of cirrhosis. Another notable finding from our study was that the diagnostic performance of NC-AMRI remained consistent regardless of the severity of chronic liver disease, assessed by CTP class, MELD score, and ALBI grade (Table 4). Sensitivities and specificities across these subgroups showed no significant differences, indicating the robustness of NC-AMRI in detecting HCC across various stages of liver disease severity.

Our study encountered several limitations. First, we used CE-MRI as the reference standard instead of histopathology due to the invasive nature of biopsies and the need for repeated follow-ups. This choice may yield inconclusive results because of technical issues like motion artefacts, inadequate contrast enhancement, or poor image quality, potentially compromising the reliability of CE-MRI for assessing diagnostic performance. Nevertheless, we addressed this limitation by implementing structured protocols with appropriate sequence selection and parameters (eg, slice thickness, field of view). Additionally, we utilized a hepatocyte-specific contrast agent, gadoxetic acid, and engaged the expertise of well-experienced radiologists specialized in abdominal imaging. Importantly, the interpretation achieved perfect agreement, with κ-values exceeding 0.8 for all sequences, further enhancing the reliability of our findings. Secondly, the limited 12-month follow-up period might have contributed somewhat to false-positive MRI findings during cross-sectional data collection. In our study, the reference standard relied solely on arterial phase hyperenhancement and venous washout in transitional phases, as per the AASLD guideline. However, in actual clinical practice, HCC diagnosis typically involves multidisciplinary discussions in tertiary centres, considering ancillary features of HCC based on MRI LI-RADS. In our study, patients with false-positive NC-AMRI examinations classified as LR-4 on CE-MRI received treatment after multidisciplinary team discussions. Therefore, we could not correlate these false-positive NC-AMRI examinations with histopathology findings to confirm whether they are actual false positives. Regarding patients with false-positive NC-AMRI examinations classified as LR-3 on CE-MRI, these patients did not develop HCC within the 12-month follow-up period of our study. We checked subsequent rounds after our 12-month follow-up MRIs, and these patients remained LR-3 on additional 6-monthly CE-MRIs. Therefore, if NC-AMRI is incorporated into clinical practice, patients with a history of false-positive results should undergo CE-MRI only for subsequent follow-ups, with individualized intervals such as 6-monthly CE-MRI for LR-3 patients. This approach would prevent unnecessary costs and burdens on health service systems, as positive findings on NC-AMRI would remain unchanged, and only CE-MRI would be beneficial for detecting HCC.

Although our study enrolled a relatively small number of patients, we meticulously calculated the sample size before initiating this prospective trial, and the number of patients surpassed our initial calculations. Conducted at a single centre, the number of MRI examinations performed was comparable to previous prospective studies.^24,28,42 Our study specifically assessed the performance of NC-AMRI for HCC in a true surveillance setting, targeting patients with CE-MRI LI-RADS LR-3 and LR-4 observations. Our prospective approach allowed us to focus on specific patient cohorts, offering valuable insights into NC-AMRI efficacy for HCC surveillance in this high-risk population. By concentrating on this subset of patients, our study contributed to the development of personalized risk stratification-based strategies with significant clinical implications. While our study may have a smaller sample size, its prospective nature and focus on high-risk patients provide unique and valuable contributions to the existing literature on HCC surveillance. However, further validation of diagnostic performance in multicentre large trials is necessary before introducing NC-AMRI into real clinical practice for HCC surveillance.

Our study focused on monitoring LR-4 observations using MRI, which may vary in accuracy for HCC diagnosis. A systematic review on MRI analysis of HCC reported that 74% (95%CI: 67%-80%) of LR-4 observations were confirmed as HCC.⁴³ In a multicentre cohort study of 93 patients with LR-4 observations, 33 patients (80.5%) developed HCC over a median follow-up of 8 months.⁴⁴ Therefore, patients with cirrhosis and LR-4 observations face a heightened risk of HCC progression. The flowchart in Figure S2 depicts the surveillance rounds and outcomes for each patient with false-positive NC-AMRI examinations, highlighting the comparison between NC-AMRI and CE-MRI findings. We were unable to correlate the false-positive NC-AMRI examinations classified as LR-4 on CE-MRI (Patients 2, 3, 4, and 8) with histopathology findings because these patients received loco-regional treatment, mostly by microwave ablation, after multidisciplinary team discussions. When NC-AMRI is positive once, it should raise awareness of potential HCC development, necessitating close surveillance with CT or MRI imaging. Our cohort included 7 LR-4 observations in 7 patients, of which 5 LR-4 observations (71%) progressed to HCC within a 1-year follow-up period. These findings corroborated previous studies, underscoring the imperative of close surveillance using CT or MRI for LR-4 observations.

While the accuracy of surveillance tests is crucial, factors such as affordability, compliance, and cost-effectiveness are also essential. The NC-AMRI protocol typically requires 15 min or less, focusing solely on essential sequences for HCC detection, which reduces the burden on time and human resources.⁴⁵ Although NC-AMRI is more affordable and less time-consuming than CE-MRI, it still costs more and takes longer than US examinations. Due to the limited accessibility and higher costs of MRI compared to US, a tailored approach to HCC surveillance using AMRI is necessary. This approach should focus on individuals most likely to benefit rather than applying it universally.⁴⁵ Further investigation and cost-effectiveness analysis are needed to identify the target populations for AMRI-based HCC surveillance. A recent cost-utility analysis demonstrated that, compared to biannual US with AFP, NC-AMRI proved to be a cost-effective HCC surveillance strategy in cirrhotic patients. The incremental cost-effectiveness ratios fall within the willingness-to-pay thresholds in both Thailand and the United States settings.⁴⁶ Effective HCC surveillance programs rely on both the diagnostic accuracy of surveillance imaging techniques and patient compliance with surveillance protocols. However, a meta-analysis revealed a pooled adherence rate of only 52% to HCC surveillance programs, indicating the importance of patient preferences and values in determining adherence.⁴⁷ To investigate patient preferences for HCC surveillance methods, Woolen et al conducted a recent study among individuals at risk of developing HCC.⁴⁸ Surveillance benefits (51.3%) were valued more than the risk of physical harm (7.6%), financial harm (15.2%), convenience (9.3%), and test logistics (16.7%). Based on simulations including all possible tests, patients preferred abbreviated MRI (29.0%), MRI (23.3%), or novel blood-based biomarkers (20.9%) to ultrasound alone (3.4%) or with AFP (8.8%). Therefore, our study, demonstrating the high diagnostic performance for HCC detection, may align with patient preferences, supporting the rationale for improved health outcomes despite the associated elevated costs. Additionally, it may serve as a suitable tool for improving patient outcomes and adherence to HCC surveillance protocols.

NC-AMRI offers many advantages as an HCC surveillance tool, including repeatability without the burden of cost and time. If NC-AMRI fails or image quality is poor, it can be immediately repeated. Implementation at various centres is feasible since NC-AMRI requires less expertise compared to CE-MRI. Additionally, all sites and vendors can perform T1WI by IP-OP imaging, T2WI, and DWI with the specified b values, which strengthens utility of NC-AMRI. There are significant differences in image acquisition and quality among the three main vendors: Siemens, Philips, and GE. Therefore, for NC-AMRI to be truly adopted for HCC screening, it is essential to conduct validation studies across vendors and involve large patient cohorts. Future work also should include a direct comparison between same-day US and NC-AMRI with independent analysis. This approach would highlight the improvements offered by NC-AMRI by identifying lesions missed by US but detected by NC-AMRI.

Conclusion

Improving the performance of HCC surveillance is a significant challenge in managing patients with chronic liver disease and liver cancer. NC-AMRI stands out as a promising tool in this arena, offering enhanced early HCC detection and potential survival benefits. NC-AMRI demonstrated high diagnostic performance in HCC detection among patients with LR-3 and LR-4 observations in CE-MRI. Nevertheless, prospective studies involving a larger patient cohort are necessary to compare the diagnostic performance of NC-AMRI with that of US. Such studies may help identify subgroups in which NC-AMRI could replace US as a surveillance tool.

Author contributions

Roongruedee Chaiteerakij, Sombat Treeprasertsuk, and Soe Thiha Maung developed the protocol. RC and ST reviewed and provided inputs to the protocol, and supervised the work. RC and STM enrolled participants in the study. Natthaporn Tanpowpong and Minchanat Satja provided inputs by interpreting MRI images. STM collected data, arranged statistical analysis and draft manuscript development. RC provided essential edits, and gave final approval with her mentorship. All authors have reviewed and confirmed the final version of the manuscript before submission.

Supplementary material

Supplementary material is available at BJR online.

Funding

No funding was taken from any pharmaceutical company. This study was funded by Ratchadapiseksompotch Research Fund by the Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand (Grant Number - GA67/075). R.C. acknowledges the grant support from the Gastroenterological Association of Thailand. S.T.M. acknowledges support from the Graduate Scholarship Programme for ASEAN or Non-ASEAN Countries by Chulalongkorn University for Master of Science (M.Sc.) in Clinical Sciences (International Program) and Clinical Fellowship program at the Division of Gastroenterology, Department of Internal Medicine, Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand.

Conflicts of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

All data, materials, and software applications support the published claims and comply with field standards.

Informed consent to participate and publish

All patient records were thoroughly anonymized and de-identified before analysis. Written informed consent was obtained from all participants by investigators. The study was approved by the Institutional Review Board of the Faculty of Medicine, Chulalongkorn University (IRB Number—0412/66).

Statement of Human Rights

All procedures performed in this study were in accordance with the ethical standards of the Institutional Review Board (IRB) of the Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand, and with the 1964 Helsinki declaration and its later amendments.