Diagnostic Performance of Simulated Abbreviated MRI for Early-Stage Hepatocellular Carcinoma Screening: A Comparison to Conventional Dynamic Contrast-Enhanced MRI.

Purpose: To compare the per-patient diagnostic performance of simulated abbreviated MRI (AMRI) to that of conventional MRI (CMRI) with full-sequence dynamic gadoxetic acid (GA) enhancement for early-stage hepatocellular carcinoma (HCC) screening in high-risk patients. Materials and
Methods: A total of 201 consecutive patients at high-risk for HCC, who underwent 3T liver MRI, were included in this retrospective study. The AMRI protocol comprised T2-weighted imaging, hepatobiliary phase imaging after GA injection, and diffusion-weighted imaging. For each patient, two AMRI and CMRI image sets were independently reviewed by two radiologists. Inter-reader agreement was assessed using Cohen's kappa value. A composite reference standard was used to determine the diagnostic performance of each image set for each reader.
Results: A total of 93 HCCs were detected in 79 patients. The inter-reader agreement was almost perfect for both image sets (κ = 0.839, 0.948). In AMRI, the per-patient sensitivity and negative predictive values (NPV) were 94.9% and 96.4%, respectively. In CMRI, the per-patient sensitivity and NPV were 96.2% and 97.5%, respectively.
Conclusion: AMRI, using only three sequences, had a comparable diagnostic performance to CMRI in screening early-stage HCC. AMRI could be an alternative HCC screening tool for high-risk HCC patients. Copyrights

INTRODUCTION

Hepatocellular carcinoma (HCC) was predicted to be the sixth most commonly diagnosed cancer and the fourth leading cause of cancer death worldwide in 2018 (1). A previous study has shown that screening for HCC using ultrasound improved patient survival by permitting the early detection of HCC (2). The American Association for the Study of Liver Diseases and many international organizations recommend semi-annual HCC surveillance using ultrasound for high-risk patients to detect HCC at an early stage (345). Despite its low cost and high availability, the per-patient sensitivity of ultrasound for the detection of small and early HCCs is not reliable, as low as 60%–63% according to previous meta-analyses (6). Tzartzeva et al. (7) reported that US detected any stage HCC with 84% sensitivity, but early stage HCC with only 47% sensitivity.

Dynamic contrast-enhanced MRI is proposed for HCC surveillance because of its high diagnostic accuracy for the detection of early stage HCC (< 3 cm) (89). However, the high cost of individual dynamic contrast-enhanced MRI as well as relatively long examination times is an obstacle for the detection of HCC (10). Abbreviated MRI (AMRI), using only a few sequences of MRI, could be an alternative method for HCC screening if the AMRI could provide adequate diagnostic performance for HCC screening. There have been several studies with variable protocols for HCC detection and have shown its usefulness for HCC screening in the high-risk group of HCC (11121314151617). The AMRI protocols include T2-weighted imaging (T2WI) as a standard and can be divided into two protocols, including hepatobiliary phase (HBP) imaging after gadoxetic acid (GA) enhancement.

Treatment options and clinical outcomes in patients with HCC depend primarily on the stage of the tumor at the time of diagnosis (181920). In that sense, the detection of early stage HCC is more meaningful in screening high-risk populations for curative treatment and prolonged survival. Most AMRI studies have analyzed stage HCCs, including advanced stage and early stage, in the detection of HCCs. Only two studies have evaluated the role of AMRI in the detection of early stage HCC (1415). Two studies reported the diagnostic performance of AMRI combined with multiphasic CT (14) or an unenhanced AMRI protocol for detecting HCC (15). Park et al. (14) reported that AMRI comprising HBP, T2WI, and diffusion-weighted imaging (DWI) plus multiphasic CT showed comparable accuracy (91.2% vs. 87.6%), positive predictive value (PPV) (94.1% vs. 93.5%), and sensitivity (96.4% vs. 92.9%) to full-sequence gadoxetic acid-enhanced MRI using Liver Imaging Reporting and Data System (LI-RADS)-4/5 criteria in the HCC screening cohort. Another study (15) showed that the sensitivity of an unenhanced AMRI for detecting HCC of readers 1 and 2 was 88.0% and 86.2%, respectively.

The purpose of this study was to retrospectively evaluate the per-patient diagnostic performance of a simulated AMRI protocol consisting of T2WI, GA-enhanced HBP imaging, and DWI compared to conventional MRI (CMRI) for early stage HCC screening in high-risk patients.

MATERIALS AND METHODS

This single-center retrospective study was approved by our Institutional Review Board (IRB No. CNUHH-2020-121), and informed consent was waived. Patients who underwent GA-enhanced liver MRI at the hospital over a 2-year period from January 2017 to December 2018 for HCC screening/surveillance or diagnosis were included in this study (n = 330). To avoid duplication, only the first examination was included in each patient. The inclusion criteria included high-risk factors for HCC, such as liver cirrhosis. The exclusion criteria included a history of known HCC with previous treatment (n= 34), no follow-up examination (n = 59), and other malignancies with metastasis (n = 3). To compare the diagnostic performance of HCC screening and surveillance in the early stage, patients with advanced HCCs with > four HCCs (n = 14), HCCs larger than 3 cm (n = 18), or diffuse infiltrative HCCs (n = 4) were excluded. Finally, 201 patients were included in the study (Fig. 1).

Fig. 1

Flow chart of the study population.

HCC = hepatocellular carcinoma

Patient demographics and clinical characteristics are summarized in Table 1. A total of 134 lesions of 109 patients were detected and assessed. A total of 93 HCCs were detected in 79 patients (79/201, 39.3%). Of the 93 HCCs identified, 59 (63.4%) were 1–1.9 cm, 34 lesions (36.5%) were 2 cm or larger, and the average lesion size was 1.81 cm. Lesions other than HCCs (n = 41) were considered cirrhotic nodules such as regenerative nodules or dysplastic nodules (n = 29), hemangiomas (n = 10), nodular fibrosis (n = 1), and focal eosinophilic infiltration (n = 1).

Table 1

Patients and Lesion Demographics

Characteristics		Patients with HCC (n = 79, 39.3%)	Patients without HCC (n = 122, 60.7%)	p-Value
Age (mean ± SD)		61.85 ± 9.95	57.57 ± 10.64	0.558
Sex, %				0.354
	Male	18 (22.8)	35 (28.7)
	Female	61 (77.2)	87 (71.3)
Etiology of liver disease, %				0.102
	Hepatitis B	53 (67.1)	64 (52.5)
	Hepatitis C	3 (3.8)	9 (7.4)
	Alcoholic liver disease	16 (20.3)	40 (32.8)
	Others (NAFLD, AIH, etc.)	7 (8.8)	9 (7.4)
Child Pugh score, %				0.816
	A	62 (78.5)	97 (79.5)
	B	13 (16.4)	21 (17.2)
	C	4 (5.1)	4 (3.3)
Lesion size (mm, number of lesions = 134)				0.010
	10–19	59 (63.4)	29 (70.7)
	≥ 20	34 (36.6)	12 (29.3)
	Mean ± SD	1.90 ± 0.59	1.56 ± 0.49

AIH = autoimmune hepatitis, HCC = hepatocellular carcinoma, NAFLD = non-alcoholic fatty liver disease, SD = standard deviation

The reference standard was considered positive for HCC if the results of a pathologic examination within 12 months of the index MRI examination were positive (n = 20). In the remaining patients (n = 73), the diagnosis of HCC was based on the radiological hallmarks of American Association for the Study of Liver Diseases or the Korean Liver Cancer Study Group- National Cancer Center Korea, namely the arterial enhancement and washout on the portal venous or transitional phases (342122). After independent review, two reviewers re-evaluated the MR findings of each HCC in consensus in terms of the presence of findings favoring HCC to clarify the incidence of these findings.

Five cirrhotic nodules were confirmed by pathologic diagnosis, and the other 24 cirrhotic nodules were finally categorized as benign, following multidisciplinary board consensus. The reference standard for cirrhotic nodules was based on comprehensive contrast-enhanced MRI reading by radiologists 1 and 2 in consensus and no interval change in clincal and imaging follow-up examinations (at least a 12 months follow-up interval; mean interval, 18 months). Other lesions, such as hemangiomas, nodular fibrosis, and focal eosinophilic infiltration, were diagnosed based on characteristic imaging findings (232425).

MRI TECHNIQUES

All images were obtained using a 3T MR scanner (MAGNETOM Tim Trio; Siemens Medical Solution, Erlangen, Germany) with a phased-array body coil. The CMRI protocols (Table 2) consisted of a breath-hold axial T1- weighted in- and out-of-phase two-dimensional volumetric interpolated breath-hold examination (VIBE) imaging. Breath-hold T2WI was acquired using fat-suppressed half-Fourier acquisition single-shot turbo-spin echo techniques, and heavily T2WI also underwent breath-hold half-Fourier-acquired single-shot turbo spin echo (HASTE) sequencing without fat suppression. For contrast-enhanced dynamic T1-weighted imaging using 3D VIBE, 0.1 mL/kg (0.25 mmoL/mL) of GA (Primovist, Bayer Healthcare, Leverkusen, Germany) was injected with an MRI-compatible injector (Nemoto Kyorindo Co., Ltd., Tokyo, Japan) at a flow rate of 1 mL/sec, followed by a 20 mL 0.9% saline flush. The axial images were obtained at 30, 60, 120, 180, 600, and 900 s after injection of the contrast material, and 900 s (15 min) was considered for HBP imaging. Additionally, contrast-enhanced coronal images were obtained at 5 min. DWI with the simultaneous use of respiratory triggering was obtained in the axial plane by single-shot echo-planar imaging using three values (b = 50, 400, and 800 s/mm2). An afferent diffusion coefficient (ADC) map was reconstructed on a pixel- by-pixel basis using the standard software on the console (Syngo, Siemens Healthineers, Erlangen, Germany).

Table 2

Imaging Parameters of the MRI Protocols

	In & Opposed Phase T1WI	T2WI & Heavily T2WI	Dynamic Contrast-Enhancement	DWI
Pulse sequence	VIBE	HASTE	VIBE	EPI
TR/TE (ms)	4.4/1.4 & 2.3	2000/81166	3.5/1.3	4500/76
Section thickness (mm)	3	5	3	5
Interslice gap (mm)	0	0.5	0	0.5
Flip angle (°)	9	90-150	13
Field of view (mm)	300–400	300–400	300–400	300–400
Matrix number	320 × 256	320 × 256	352 × 194	160 × 120
B values (sec/mm2)	-	-	-	50, 400, 800

DWI = diffusion-weighted imaging, EPI = echoplanar imaging, HASTE = half-Fourier-acquired single-shot turbo spin echo, TE = echo time, TR = repetition time, T1WI = T1-weighted imaging, T2WI = T2-weighted imaging, VIBE = volumetric interpolated breath-hold examination

The acquisition scan time for our full conventional liver MRI protocol, including dynamic phases after GA injection, was approximately 40 min, compared to approximately 10 min (including the setup) for the simulated AMRI protocol, which included T2WI (3–4 min), HBP imaging (1 min), and DWI (4 min).

IMAGE ANALYSIS

Two radiologists (both board-certified radiologists with 15 years and 5 years of experience, respectively, in abdominal imaging) who were blinded to the clinical information and pathologic analyses, independently assessed two sets of images (AMRI and CMRI) per patient. The two sets of AMRI and CMRI images were interpreted at different times, at least four weeks apart, to reduce recall bias.

In the simulated AMRI, for each patient, all lesions ≥ 10 mm were detected and their signal intensities on T2WI and HBP imaging were assessed. Radiologists measured the lesion size on the HBP image because overestimation of the lesion size could be made on T2WI and DWI (26). Lesions smaller than 10 mm were neglected. First, all lesions were scored according to the scoring system (Table 3) of a previous study (12). In the scoring system, very hypersignal intensity on T2WI indicates bright signal intensity compared with signal intensity of the spleen, whereas very hyposignal intensity on T2WI means absolutely dark signal intensity compared with the signal intensity of hepatic parenchyma. Then, the nodule categorization was adjusted according to the signal characteristics on DWI. The lesions with restricted diffusion showing high signal intensity on DWI and low signal intensity on the ADC map were upgraded by one category. The final score for each lesion was considered to be the highest final score for any nodule on that examination. Final imaging scores of 4 or 5 were interpreted as positive examinations.

Table 3

Scoring of Abbreviated MRI

		Hepatobiliary Phase Imaging
		Hypointense	Isointense	Hyperintense
T2-weighted imaging
	Very hyperintense	2	2	2
	Mildly hyperintense	5	4	4
	Isointense	4	2	3
	Mildly hypointense	4	3	3
	Very hypointense	2	2	2

Separately, on the CMRI, all lesions ≥ 10 mm were detected and assessed according to LI-RADS v2018, using LI-RADS category 1 to 5 (2728). Lesions 10 mm or larger in the LI-RADS category of 4 and 5 were considered positive examinations (Fig. 2). Likewise, LI-RADS does not allow sub-centimeter observations to be considered definite HCC (2728). Sub-centimeter HCCs are unlikely to metastasize or become untreatable before the next surveillance examination six months to one year later.

Fig. 2

A 52-year-old male with hepatocellular carcinoma.

A-F. The mass (arrows) shows arterial enhancement (A), washout on the portal phase (B), hypointensity on hepatobiliary phase imaging (C), subtle hyperintensity on T2-weighted imaging (D), and restricted diffusion on diffusion-weighted imaging (E, F). The final imaging score for abbreviated MRI and conventional MRI is 5 and interpreted as positive examinations.

STATISTICAL ANALYSIS

To analyze the inter-reader agreement between the two image sets, Cohen's k statistics were applied, where 0–0.20 indicated slight agreement, 0.21–0.40 indicated fair agreement, 0.41–0.60 indicated moderate agreement, 0.61–0.80 indicated substantial agreement, and 0.81– 1.00 indicated almost perfect agreement (29). For each image set and reader, the per-patient sensitivity, specificity, PPV, negative predictive value (NPV), and accuracy were obtained with binominal 95% confidential intervals (CIs), to compare the per-patient diagnostic performance of the two AMRI and CMRI protocols.

All statistical analyses were performed using IBM SPSS Statistics for Windows, version 23.0 (IBM Corp., Armonk, NY, USA) and MedCalc software version 18.11.3 (MedCalc Software, Mariakerke, Belgium). A p-value of less than 0.05 was considered significant.

RESULTS

INTER-READER AGREEMENT

Cohen's Kappa values for the two observers were 0.839 (0.764–0.915) for AMRI, and 0.948 (0.904–0.993) for CMRI, indicating that the inter-reader agreement between the two image sets was almost perfect for both image sets.

PER-PATIENT DIAGNOSTIC PERFORMANCE

In the AMRI, the mean per-patient sensitivity and NPV were 94.9% (95% CI: 90.3–97.8) and 96.4% (95% CI: 93.1–98.1), respectively. In the CMRI, the mean per-patient sensitivity and NPV were 96.2% (95% CI: 91.9–98.6) and 97.5% (95% CI: 94.6–98.8), respectively. A comparison of the mean sensitivity and NPV showed no statistically significant difference between the AMRI and CMRI (p > 0.05). Other per-patient diagnostic performance parameters for each set 1 and 2 are summarized in Table 4.

Table 4

Per-Patient Diagnostic Performance of AMRI and CMRI

		AMRI (%)	CMRI (%)	p-Value
Reader 1
	Sensitivity	94.9 (87.5–98.6)	98.7 (93.1–99.9)	0.060
	Specificity	87.7 (80.5–93.0)	94.3 (88.5–97.7)	0.033
	PPV	83.3 (75.6–89.0)	91.8 (84.4–95.8)	0.015
	NPV	96.4 (91.1–98.6)	99.1 (94.3–99.9)	0.137
	Accuracy	90.5 (85.6–94.2)	96.0 (92.3–98.3)	0.046
Reader 2
	Sensitivity	94.9 (87.5–98.6)	93.7 (85.8–97.9)	0.761
	Specificity	86.1 (78.6–91.6)	95.1 (89.6–98.2)	0.004
	PPV	81.5 (73.9–87.3)	92.5 (84.9–96.4)	0.002
	NPV	96.3 (90.9–98.5)	95.9 (90.8–98.2)	0.960
	Accuracy	89.5 (84.4–93.4)	94.5 (90.4–97.2)	0.096

Data are percentage (95% confidential interval).

AMRI = abbreviated MRI, CMRI = conventional MRI, NPV = negative predictive value, PPV = positive predictive value

FALSE-NEGATIVE AND FALSE-POSITIVE LESIONS

There were six false-negative examinations on AMRI, two for both readers, two for reader A and two for reader B. Both readers scored the two AMRI examinations as score 1 or 2, whereas the CMRI showed single LI-RADS category 4 observations for each (probably HCC).

Reader A missed two mild hyperintense lesions on T2WI and isointense lesions on the HBP images of two patients, which were confirmed on follow-up dynamic enhanced CT and transarterial chemoembolization. Reader B missed one lesion with mild hypointensity on HBP imaging, but it was proven to be HCC in follow-up studies. Other lesions showing very hyperintensity on T2WI and hyperintensity on HBP were scored as score 2 by reader B on the AMRI due to hyperintensity on T2WI, but the CMRI showed a typical dynamic enhancement pattern of HCC with LI-RADS category 5, which was pathologically confirmed as HCC on surgery. There is no known malignancy other than HCC.

There were 22 false-positive examinations on AMRI, ten by both readers, five by reader A, and seven by reader B. Most of the false-positive lesions were presumed to be cirrhotic nodules (Fig. 3), such as regenerative nodules (RNs) and dysplastic nodules (DNs) (19/22, 86.4%). Other false-positive lesions were a hemangioma (1/22, 4.5%), nodular fibrosis (1/22, 4.5%), and focal eosinophilic infiltration (1/22, 4.5%). The cirrhotic nodules were hypointense lesions on HBP imaging (18/22, 81.8%), mild hyperintense lesions on T2WI (9/22, 40.9%), or restricted diffusion on DWI (4/22, 18.1%).

Fig. 3

False positive lesion on AMRI in a 57-year-old male.

A-F. The nodule (arrows) showing hypointensity on hepatobiliary phase (A), subtle hypointensity on on T2-weighted imaging (B), and no diffusion restriction on diffusion-weighted imaging (C) on AMRI, assigned a score of 4. However, on dynamic contrast-enhanced MRI, the nodule (arrows) showed hyperintensity on pre-contrast T1-weighted imaging (D), no enhancement on arterial phase image (E), and washout on portal phase image (F). The tumor marker was within the normal range and the nodule was stable at the 18-month follow-up; it was considered a benign cirrhotic nodule.

AMRI = abbreviated MRI

DISCUSSION

Our study simulated the performance of an AMRI by focusing on a subset of sequences, including T2WI, HBP, and DWI. The advantage of AMRI is that it is a shortened version of the CMRI examination of the liver for the surveillance of HCC in cirrhotic patients. AMRI, including T2WI, HBP imaging, and DWI, can be performed at a lower cost in a shorter time than CMRI, thus making it more suitable for routine HCC screening (11121316). The T2WI sequence of MRI serves to differentiate benign cysts and some hemangiomas from malignant lesions, such as HCC (30). Most lesions of non-hepatocyte origin, including HCC as well as benign lesions, show hypointensity on HBP imaging, which can explain the high sensitivity of MRI (3132). DWI is included to potentially detect HCC nodules that are not recognized or reported on other acquisitions (3334).

Our study showed that the inter-reader agreement was almost perfect, which means that this method might be generalizable. The mean per-patient sensitivity and NPV of AMRI were 94.9% and 96.4%, respectively. In contrast, the mean per-patient sensitivity and NPV of CMRI were 96.2% and 97.5%, respectively. The AMRI in our study had sensitivity and NPV in the surveillance of early stage HCC, clinically comparable to CMRI with full sequences. One report (15) showed that the sensitivity of each of the two readers was 88.0% and 86.2% in the detection of early HCC on unenhanced AMRI, which was lower than that of our results. This is because our study performed AMRI, including HBP imaging. Besa et al. (11) reported that AMRI including HBP imaging had a per-patient sensitivity of 80.6% and an NPV of 94.2%. Marks et al. (12) reported that the mean per-patient sensitivity and NPV were 82.6% and 93.2%, respectively. The NPVs in previous studies including HBP imaging (1112) were similar to our results, but the sensitivities were lower than in our study. Although previous studies included large-sized advanced HCCs, they showed lower sensitivity compared to our study. The reason why the sensitivity of our study was good compared to the previous study may be that our study used only 3T MRI with good resolution and thin slice thickness of the HBP imaging, whereas previous studies used 1.5T or thick slice thickness of the HBP imaging.

In our study, there were several false-negative cases where mild hyperintensity on T2WI or mild hypointense lesions on HBP imaging were not detected and missed, probably due to a background of heterogeneous hepatic parenchyma. Patients with cirrhotic liver, a known common risk factor for HCC, show a multinodular appearance of mixed-signal intensity or diffuse heterogeneous appearance, which obscures small lesions. This is a relatively common limitation in ultrasound, which is currently used for screening for HCC as well as AMRI (73536).

The per-patient specificity of our study was relatively low due to the higher false-positive rate of AMRI. Most of the false-positive lesions were presumed to be cirrhotic nodules, following multidisciplinary board consensus. Typically, RNs show iso- to hypointensity on T2WI, iso- to hyperintensity on HBP imaging, and DNs demonstrate variable signal intensity on T2WI and HBP images (3738). In clinical practice, if there are positive lesions suspected of HCC in screening AMRI protocols, dynamic contrast-enhanced MRI needs to be considered.

In our study, it took 30–40 min to perform the CMRI examinations. On the other hand, AMRI could be completed within 10 min if the GA injection was performed outside the MRI scanner. AMRI has a very short scan time because AMRI uses only a few sequences compared to CMRI, so it can be expected to be inexpensive and more beneficial to patients. A few studies have compared the cost effectiveness of AMRI for HCC surveillance (1139). Besa et al. (11) found a 30%–35% cost saving when using a similar AMRI consisting of HBP imaging and DWI compared to a full contrast-enhanced CMRI. In a study to evaluate the cost-effectiveness of risk-stratified HCC screening, AMRI had a lower incremental cost-effectiveness ratio and was the most cost-effective strategy across a wide range of HCC incidences compared with ultrasound, CT, and CMRI (39). Further numerical verification is needed to accurately identify cost reductions.

Our study has several limitations. First because of the retrospective study design, there was a selection bias in patient sampling. The prevalence of HCC in this study is higher than that in a prospective study (9). We are concerned about the overestimation of the performance of AMRI. Further prospective research is needed to confirm our results. However, all examinations were performed within a single center and with a single 3T MRI, which may not necessarily translate to scanners of different model or field strength. Second, the benign lesions in this study included hemangiomas, cirrhotic nodules, and focal inflammatory or fibrotic lesions. Although these benign liver lesions are common and their differentiation from hypervascular HCCs is an important diagnostic issue, no other types of benign liver lesions were included. Finally, histopathologic confirmation of the lesions was not available, and consensus reading of the MRI and follow-up clinical studies were used as the reference standard based on guidelines for diagnosing HCC (3420).

In conclusion, AMRI using only three sequences had a comparable sensitivity and NPV, compared to CMRI for the screening of early stage HCC in cirrhotic patients. Therefore, it could be an alternative HCC screening tool for high-risk HCC patients. A multi-center prospective trial is needed to confirm these findings.