Evaluation and Diagnosis of Nonalcoholic Fatty Liver Disease/Steatohepatitis by K-Fold Validation Based on Pathology, Including Real-Time Shear Wave Elastography and Noninvasive Biomarkers

This study aim to investigate the diagnostic accuracy of shear wave elastography (SWE) for diagnosing nonalcoholic steatohepatitis (NASH) and staging brosis in a cohort patients conrmed nonalcoholic fatty liver disease (NAFLD) by liver biopsy. Methods A total of 86 NAFLD patients and 17 normal-control were enrolled. The performance of SWE to diagnose NASH and stage brosis was evaluated on the basis of histopathological inammation grades and brosis stages according to Kleiner/Brunt et al.’s criteria classication, and compared to previous reported four noninvasive serum brotic scores, coupled with the k-fold-cross-validation and Delong test. Meanwhile, inuence of steatosis on liver stiffness measurements (LSMs) of SWE was also studied. LSMs (AUROC and brotic NASH: ≥ F2 stage (AUROC (cid:0) ≥ F3 (AUROC (cid:0) and =F4 stage (AUROC (cid:0) 0.94) with the cutoff values were 7.55, 7.65, 8.25 and 11.80 kPa, respectively. Compared with serum brotic scores, SWE had the highest AUROC for predicting ≥ F2, ≥ F3, =F4 by Delong test (all P (cid:0) 0.05). No statistic differences of LSMs were found among different steatosis levels (P=0.29).


Introduction
Nonalcoholic fatty liver disease (NAFLD), de ned by the presence of steatosis in > 5% of hepatocytes, is a spectrum of diseases strongly associated with insulin resistance [1]. In recent years, NAFLD prevalence is growing rapidly [2,3] and has become most common cause of chronic liver diseases in the world [4,5]. Based on histopathological features of hepatocellular damage, NAFLD encompasses a wide spectrum ranging from benign nonalcoholic fatty liver (NAFL), progressive non-alcoholic steatohepatitis (NASH), to cirrhosis nally [1]. Owing to NAFL usually remains stable for years without progression in most patients [6][7][8], but NASH always indicates an increased risk of brosis progression, cirrhosis and possibly hepatic comorbidities (such as hepatic failure, hepatocellular carcinoma) [9,10]. It is important to accurately differentiate NASL and NASH, which is attributed to formulate the optimum options for monitoring and treatment in NAFLD patients [11].
Liver biopsy as the gold standard has long been considered for assessing the stage of NAFLD. However, it is invasive, di cult to accept, and inconvenient for the repeat measurement of treatment reaction [1]. Thereby, several non-invasive methods have been studied in the last decade, including cytokeratin-18 fragments (CK-18) [12], NAFLD brosis score (NFS) [13], brosis 4 calculator (FIB-4) [14], and liver stiffness measurements (LSMs) based on transient elastography (TE) imaging [15,16], demonstrating good capabilities for diagnosis of NAFL, NASH, and advanced brosis. However, none of them can be used to substitute liver biopsy based on the convergent ndings from recently critical appraisals [17,18]. Furthermore, there has been no consensus on thresholds or strategies for use in non-invasive methods when trying to avoid biopsy according to the guidelines from European Association for the Study of the Liver (EASL) [1,18]. Consequently, more preferable strategies are urgently needed to improve the capability of the non-invasive measurements for diagnosis of NAFLD subtypes.
Shear wave elastography (SWE) based on a radiation force induced by focused ultrasonic beams in tissues has many advantages, including realtime, convenience to implement, high range of values (2-150 kPa), and high performance for diagnosis, etc. [18]. Recently, use of SWE has been reported to be effective for accurate diagnosis of brosis in various liver diseases [19][20][21][22][23][24]. Nevertheless, little research has been investigated in NASH patients. In addition, steatosis and brosis would co-exist in the same NAFLD patient. Thereby, the effect of steatosis on LSMs of SWE also needs to be assessed.
Hence, the purpose of this study is to (1) validate the diagnostic accuracy of SWE for NASH and quality criteria using histopathology as the reference; (2)

Patients
All the patients' private information is not included in the paper. We included patients aged 18 years or older in West China hospital from September 2015 to September 2019, whom pathologic examination con rmed the accumulation of lipid within more than 5% hepatocytes. All patients underwent liver biopsy and liver SWE measurement consecutively in one day. The exclusion criteria were showed in supplemental materials (1.Supplemental exclusion criteria of NAFLD and normal control group).

Clinical and Laboratory Assessments
Relevant clinical data: sex, age, weight, height, and alcohol intake (g/day), and laboratory results: total bile acid (TB), direct bile acid (DB), blood glucose, triglyceride (TG), alanine transaminase (ALT), aspartate transaminase (AST), alkaline phosphatase (ALP), lactic dehydrogenase (LDH) were collected at the time of liver biopsy. Body mass index (BMI) was calculated as weight (kg) divided by height (m) squared.

US examination and SWE
All subjects underwent ultrasound examination by SuperSonic Imagine AixPlorer (SuperSonic Imagine, Aix-en-Provence, France) before fast for at least 8 hours. SWE was conducted using a convex probe (SC6-1) by a pro cient sonographer performing ultrasound examinations, SWE measurement and needle biopsy. The examinees were placed in the supine position with the right-arm maximal extension. Two-dimension ultrasonic examination was rst performed before select SWE mode. The transducer was positioned in the right lobe of the liver through the intercostal space (segment V, VIII, or VII) with the transducer at 90° in relation to the liver capsule in an area free of artifacts and large vessels. The region of interest was placed a minimum of 1-2 cm and a maximum of 6 cm below the liver capsule. The analysis box was set to at least 10 mm. The patients transiently held their breath in a neutral position. Five successful measurements were repeated. The mean of the ve times SWE measurements expressed in kilopascals (kPa) was used as the representative measurement.

Histopathological assessment
After US and SWE examination, liver needle biopsy was performed in the same area of SWE examination.  Figure 2 showed the example of SWE images and HE stain sections.

Predictors of NASH.
The result of binary regression was showed in Table 2, only SWE have statistic difference with OR = 3.91 among multiple variables (P = 0.01). The LSMs (± SD) of NASH group (10.2 ± 4.5 kPa) were higher than both normal control (6.1 ± 0.9 kPa) and NAFL group (6.2 ± 0.7 kPa) (P < 0.001) (Fig. 3). However, there was no statistical difference between the normal control (6.1 ± 0.9 kPa) and NAFL group (6.2 ± 0.7 kPa) (P = 0.26). For the diagnosis of NASH by SWE and calculated optimal cutoff value, the data obtained were randomly divided into train sets and testing sets (ratio of 4:1) by caret package, and then analyzed by pROC package of R. As demonstrated in Fig. 4 and    (Table 4). Nevertheless, the optimal applications for stage brosis using NFS, FIB-4, BARD and APRI scores were different: APRI is appropriate to ≥ F2, NFS, FIB-4 and BARD score are more suitable for applying to ≥ F3 [13,14,25,26]. In fact, best application of four serum brosis scores also showed lower AUROCs for their respective optimum when compared with SWE (showed Table 5). Consequently, the LSMs on SWE was signi cantly superior (all P < 0.05) for diagnosis of liver brosis stage (F2-4), compared to the NFS, FIB-4, BARD and APRI scores respectively (Fig. 6).

The in uence of steatosis and in ammation on LSMs
In this study, steatosis, lobular in ammation and brosis would coexist in NASH patient. To assess the effect of steatosis on LSMs, we strati ed NAFL patients with different steatosis levels and NASH patients at each brosis stages with different in ammation grades. The LSMs (± SD) of NAFL patients with 5-33%, 33-66% and ≥ 66% of steatosis hepatocytes were 6.2 ± 0.6, 6.9 ± 0.5 and 6.0 ± 0.9 kPa, respectively, showing no statistic differences among different steatosis levels (F = 1.3, P = 0.29).

Discussion
The prognosis of NAFLD depends heavily on the histopathological severity [1]. Although liver biopsy has traditionally been considered the only reference method for evaluation of tissue damage, problem is that liver biopsy only gives a snapshot and not an insight into the dynamic changes during the process [31]. Moreover, it is invasive, di cult to accept, and susceptible to signi cant sampling variability and has a risk of severe complications [1]. Therefore, an ideal noninvasive surrogate marker or tests is urgently needed.
Several blood biomarkers and scores systems have been proposed for differential diagnosis of simple steatosis from NASH.
Among the clinical and laboratory parameters, the best representative biomarker is cytokeratin-18 with 66% of Sen and 82% of Spe [12]. Besides, NAFIC score (AUROC = 0.851) were reported to be useful for diagnosing the early stage of NASH via the combination with blood markers such as Palekar's score [32,33]. However, the use of laboratory indexes above is unusual and costly. Moreover, none of them are liver speci c and their results may be in uenced by changes in clearance and excretion of each individual parameter. Therefore, more longitudinally veri able data are needed. It is known that the best-studied imaging modality was magnetic resonance elastography (MRE), demonstrating the high accuracy (AUROC = 0.93) for discriminating patients with NASH from those with simple steatosis, with 94% of Sen and 73% of Spe [34]. However, the group design utilized in this study lacked the clear histological diagnosis between NASH and not-NASH disease. To date, noninvasive tests have not been validated for diagnosing NASH because of the small number of studies.
SWE is an emerging noninvasive method based on shear waves implemented on the diagnostic ultrasound system to provide quantitative analysis of tissue stiffness. In the present study, we evaluated the performance of SWE on the diagnosis of NAFLD and compared it with the four serum brosis scores using histopathology as reference. The LSMs on SWE was proved to be a reliable method that can not only bring up potentially steatohepatitis hint, but also roughly determine the stage of liver brosis and would not been in uenced by the presence of steatosis at the same time.
In the present study, we evaluated the performance of SWE in indentifying NASH in NAFLD patients and con rmed the good performance of SWE to diagnose NASH with an AUROC of 0.85 (0.73-0.92) at a cutoff of 7.55 kPa. To the best of our knowledge, this is the rst study to evaluated SWE in a clinic concerned NASH, which was deemed to steatosis with lobular and portal area in ammation rather than minor in ammation.
As for staging brosis, dozens of noninvasive models composed of blood biochemical biomarkers were reported to be useful, including NFS [13], FIB-4 [14], APRI [25], enhanced liver brosis (ELF) [35] and CA index [36]. However, APRI and FIB-4 were initially established for patients with hepatitis C virus (HCV) infection or HIV/HCV coinfection. ELF and CA index need some special tests. In this study, the AUROC of SWE LSMs on detection of brosis is superior to those of four scoring systems (NFS, FIB-4, APRI and BARD). This is reasonable because some blood markers of the scoring systems would be affected by races and diets, whereas the SWE LSMs is directly re ect stiffness of the liver and guided by a higher frame-rate B-mode image that can yield a more accurate measurement. Transient elastography (TE) was initially shown to be reliable for assessing brosis in patients with chronic hepatitis C [15]. TE also was reported to be useful to assess brosis in patients with NAFLD [21,37]. However, there were some limitations of TE. The rst is its one-dimensional imaging that may fail to obtain reliable LSMs.
As for NAFLD disease, some animal studies have reported that SWE is an e cient technique to differentiate NASH from less severe NAFL [39,40]. However, limited human data were available on NAFLD. Samir  , consistent with our results. However, it cannot be ignored that the in uence of severe steatosis on LSMs, because fatty attenuation of severe steatosis may lead to SWE measurement failure [24].
Nevertheless, several literatures had inconsistent conclusions. In some transient elastography studies, steatosis has been reported to have an effect on TE LSMs [45][46][47].
One of the strengths of the study is that we consulted clinically concerned NAFLD classi cations [11]: steatosis with minor in ammation is regarded as non-progressive and assigned to the NAFL group, rather than early NASH. Declarations Data sharing statement: The datasets generated and analysed during the current study are not publicly available due to ensure data privacy of the study participants but are available from the corresponding author on reasonable request.
Ethics approval and consent to participate: Informed consent was not required by the review board because patients did not need to be contacted for this retrospective data analysis. And patients' privacy is not included in this retrospective study. All experimental protocols were approved by Sichuan University. All methods were carried out in accordance with standards for reporting of diagnostic accuracy.
Con ict of interest: The authors have no nancial con icts of interest.  Results of shear wave elastography images (top row) and HE stain sections (bottom row) in patients with NAFLD having simple steatosis (A and D), non-cirrhosis NASH (B and E), and cirrhosis NASH (C and F). The mean liver stiffness was 5.5 kPa, 7.8 kPa, and 13.8 kPa, respectively.
No statistical difference was found for LSMs as compared with normal control and NAFL group (P 0.26).

Figure 4
Page 16/17 Graphs show 5-fold cross validation of ROC curves for diagnosing NASH by SWE in train sets, the mean AUROC=0.85.