A vein-viewing application enabled detecting abdominal wall varices related to the presence of non-treated gastroesophageal varices: a cross-sectional study

Background Gastroesophageal varices (GOV) are a life-threatening complication in chronic liver disease. A method for non-invasively predicting GOV is crucial for management. This study aimed to determine whether a vein-viewing application can detect abdominal wall varices (AWV) and elucidate the relationship between AWV and GOV. Methods One-hundred patients with chronic liver diseases were prospectively enrolled. All the patients underwent esophagogastroduodenoscopy within three months of the enrollment. Unmanipulated images (UI) and vein-weighted images (VWI) were taken for assessing AWV by a vein-viewing application on iPhone. Two doctors independently evaluated both image types. We defined the grading of both UI and AWV as grade 0 (non-detectable), grade 1 (slightly detectable), and grade 2 (distinct). Results The causes of liver diseases among the 71 men and 29 women (median age, 70.5 yr) included Hepatitis B (n = 19), Hepatitis C (n = 21), alcoholism (n = 33), primary biliary cholangitis (n = 3), autoimmune hepatitis (n = 4) and others (n = 20). GOV was indicated in ﻿60﻿ patients, and half of them had not been treated previously (non-treated). VWI could significantly visualize AWV than UI (72% vs. 24%, p = 0.0005). The presence of cirrhosis (chronic hepatitis vs. cirrhosis = 64.6% vs. 91.4%, p = 0.004) and GOV (52.3% vs. 74.3%, p = 0.032) were significantly higher in the VWI-AWV grade 2 group. Multivariate analysis demonstrated that VWI-AWV grade 2 was an independent factor related to the presence of non-treated GOV [OR = 3.05 (1.24–7.53), p = 0.016]. Conclusions The vein-viewing application non-invasively detected AWV related to the presence of cirrhosis and GOV, and VWI-AWV grade 2 was an independent factor related to the presence of non-treated GOV.


Background
Gastroesophageal varices (GOV) are present in about half of the patients with cirrhosis [1]. Variceal bleeding is a life-threatening complication which accounts for 10-30% of all upper gastrointestinal bleeding [2]. Esophagogastroduodenoscopy (EGD) is the gold standard for the detection of GOV. Disadvantages of endoscopy include the risk of sedation, higher cost, bleeding, and risk of aspiration [3]. However, no recommendations on screening of GOV has been made in Japan [4]. Many less invasive methods for screening of GOV have been investigated [5]. Serum biomarkers including platelet count, FIB-4 index, aspartate aminotransferase to platelet ratio index (APRI), liver stiffness (LS), spleen stiffness (SS), LS-spleen diameter to platelet ratio, and Liver stiffness × spleen size/platelet count (LSPS) are reportedly useful for predicting esophageal varices [6][7][8][9][10][11][12][13][14][15]. The updated Baveno VI guidelines recommend that screening EGD can be avoided in patients with compensated advanced chronic liver disease who have liver stiffness < 20 kPa and a platelet count > 150,000/mm 3 [16].
We focused on abdominal wall varices (AWV) for predicting GOV. Several prominent collateral veins radiating from the umbilicus are termed the caput-medusae. The caput-medusae sign is an indicator of portal hypertension. It describes engorged paraumbilical veins radiating from the umbilicus within the adipose tissue of the anterior abdominal wall, creating portosystemic anastomoses [17]. However, in clinical settings, it could not be commonly identified. It is now considered a rare finding. The use of infrared photography for the visualization of AWV is reported in the literature [18]. There are no specific modalities for visualizing AWV. Therefore, we used a vein-viewing application on the iPhone instead of an infrared camera. This can visualize the high-contrast image of the vein by boosting oxyhemoglobin/deoxyhemoglobin absorption contrast and reducing the contribution of superficially scattered and specularly reflected light to the overall image.
In this study, we aimed to evaluate the efficacy of the vein-viewing application for detecting the AWV in patients with chronic liver disease and elucidating the relationship between AWV and GOV.

Methods
This was a single-center, prospective, cross-sectional study. Between November 2018 and September 2020, one-hundred adult patients in our hospital with any chronic liver disease (including cirrhosis) were prospectively enrolled. All the patients underwent EGD within three months of inclusion. Patients with skin diseases of the abdominal wall were not enrolled because of skin discoloration preventing successful imaging. We obtained both unmanipulated images (UI) and veinweighted images (VWI) with VeinSeek Pro (VeinSeek LLC, Los Angeles, CA) (https:// www. veins eek. com/) for each patient. VeinSeek Pro for iPhone can be downloaded via App Store for iPhone (https:// apps. apple. com/ us/ app/ veins eek-pro/ id117 45363 86). VeinSeek version 2 for android is also available; however, it does not work as well as VeinSeek Pro. We defined the grading of AWV as grade 0 (non-detectable), grade 1 (slightly detectable), and grade 2 (distinct) for both unmanipulated and VWI, respectively (Fig. 1). Both images were evaluated by two doctors (Dr. S and N) independently. We obtained the patient's information on biological gender, age, body mass index (BMI), and mental status (regarding hepatic encephalopathy) at the time of imaging. The following data: hemoglobin, total bilirubin, albumin, prothrombin time (PT), fibrosis index based on the four factors (FIB-4) index using age, aspartate transaminase (AST), alanine transaminase (ALT), and platelet values [19], and AST to platelet ratio index (APRI) [20] were also collected. The severity of cirrhosis was determined according to the Child-Pugh scoring system based on PT, albumin, bilirubin values, and the presence of encephalopathy or ascites. Patients were classified into Child A (5-6 points), B (7-9 points), and C (10-15 points) groups. Classification of GOV was according to the "general rules for recording the endoscopic findings of esophagogastric varices in Japan" [21]. Moreover, gastric varices were classified according to Sarin's classification [22]. Other abdominal imaging techniques (ultrasound, computerized tomography, or magnetic resonance imaging) were also applied for evaluating ascites. FibroScan measures of liver stiffness were also performed on patients without ascites.

Statistical analysis
The Student's t-test and chi-square test were applied for comparing the two groups as defined by the cutoff criteria. One-way ANOVA was applied for multiple comparisons. Interrater reliability was assessed by the Cohen's kappa coefficient. A Kappa > 0.7 indicates agreement between two operators. Logistic regression analysis was applied for multivariate analysis. The Spearman rank-order correlation coefficient (shown as rS) was used for evaluating the correlation between two variables. All statistical tests were performed using StatFlex (Windows ver. 6.0; Artech, Osaka, Japan). Values are expressed as median (range) or mean with a standard error of the mean (SEM). Categorical variables are shown as numbers. Statistical significance was set at p < 0.05.

Ethics approval and consent to participate
The study protocol was approved by the Institutional Review Board of Tottori University (No.18A152) under the guidelines of the 1975 Declaration of Helsinki. Written informed consent was obtained from all the participants.

Multivariate analysis of predicting factors for GOV
Multivariate analysis was applied for the factors related to GOV in Table 3. APRI and FIB-4 index were not selected because of including platelet count. Liver stiffness was also not selected because of the lack of data in 32 patients with ascites. Age ≥ 71 years [OR = 0.35 (0.14-0.85), p = 0.021] was an independent factor, and VWI-AWV grade2 [OR = 2.40 (0.91-6.33), p = 0.076] approached the borderline of significance. In this study, liver cirrhosis was lower in patients ≥ 71 years old (64% vs. 84%, p = 0.023). Therefore, age was negatively related to GOV (Table 5).

Multivariate analysis of factors for non-treated GOV
Multivariate analysis was also applied for non-treated GOV. It was also applied both with and without liver stiffness. Only VWI-AWV grade2 was an independent factor related to non-treated GOV [OR = 3.05 (1.24-7.53), p = 0.016] ( Table 6).

Relationship between parameters or shunts and VWI grading
Several parameters, such as hemoglobin, total bilirubin, and BMI, can affect VWI grading. However, there were no correlations observed between hemoglobin (Dr.S; rS = −0.092 p = 0.363, Dr.N; rS = −0.029 p = 0.777) or

Discussion
In this pilot study, we demonstrated that the vein-viewing application on iPhone could non-invasively detect AWV related to cirrhosis and GOV. This is the first report of the non-invasive method of simply taking AWV images that enables us to indicate the patients who should be applied for medical service and EGD. Among the forty patients without GOV the mean platelet count was over 150,000/mm 3 , and mean liver stiffness was 17 kPa. This suggested that the cutoff levels for avoiding EGD in the Baveno VI guidelines were practical in indicating patients with a low risk of GOV.
In our study, age was an independent factor negatively related to GOV. This was an unexpected result. The difference in etiology may have caused this. The number of patients with alcoholism was larger in the participant group under 71 years of age compared to the older group (42% vs. 24%, p = 0.056). Moreover, most of the patients with alcoholism had cirrhosis (87.9%).
The image-based method for the prediction of GOV has been validated. Further development will enhance the usefulness of this approach in future medical diagnostics. Smartphones and mobile devices have rapidly become part of everyday life around the world. In the current situation with COVID-19 the role of on-line medical services is increasingly important. The vein-viewing application on the iPhone was not originally developed for medical purposes; however, we have established that it is useful in detecting AWV in cirrhotic patients in a medical context.
In this study, VWI-AWV grade 2 was related to the presence of cirrhosis, high Child-Pugh score, the presence of ascites, the presence of GOV, the presence of PHG, and low platelet count. A weak positive correlation between total bilirubin and VWI grading can also be associated with liver dysfunction. Furthermore, multivariate analysis demonstrated that VWI-AWV grade 2 was an independent factor related to non-treated GOV. GOV treatment would alter the hemodynamics, including AWV. Our approach is therefore more meaningful for diagnosing naïve than treated patients. Intriguingly, eight patients (22.9%) who were identified as grade 2 had no AWV when assessed by UI. Five patients with GOV (three were untreated) were included in the eight patients. The interrater reliability was lower in VWI-AWV grade 0-1, indicating that identifying a slight AWV was difficult. However, the identification of grade 2 AWV was significantly higher by VWI in both doctors, and the reliability of VWI-AWV grade 2 was satisfactory.
Among twenty-two VWI-AWV grade 2 patients who had no history of GOV treatment, six patients did not have any GOV. In this group, four patients (67%) had cirrhosis. VWI-AWV grade 2 may therefore have the potential to identify not only GOV but also cirrhosis. However, the other two patients had no cirrhosis and GOV; this would be an entirely false positive. Novel technology is warranted for the improvement of the vein-viewing application to minimize this outcome.
The role of artificial intelligence is also rapidly growing in the medical field, such as pathology, EGD, mammography, brain diseases, and COVID-19 diagnosis [24][25][26][27][28]. Deep learning of AWV structures would provide a highly reproducible diagnosis of AWV. It also means that each person can check themselves with such applications on mobile devices in the future. Our effort should be focused on quantifying the imaging capabilities of mobile devices  on the human body and provide meaning and context to them. This study has several limitations. The cohort studied represented a small group of patients on which EGD could be performed. Selection bias was therefore inevitable. However, based on the promising results of this pilot study, a large-scale cohort study will be conducted for validation. Presently, there are no available objective data on detectability differences for skin color. One user from Zimbabwe commented on the App Store review that the app was helpful to patients with dark skin. Although it may work for different skin colors, verification is warranted.
In summary, the vein-viewing application could noninvasively detect AWV related to the presence of cirrhosis and GOV. VWI-AWV grade 2 was an independent factor related to the presence of non-treated GOV. This result suggests a future direction of medicine using consumer mobile devices as medical devices. The camera lens will be like the eyes on "Baymax," a prototype healthcare-providing robot on Disney animation.  Correlation between parameters and VWI grading. a no correlation between Hb and VWI grade, b slight positive correlation between T-bil and VWI grade, and c no correlation between BMI and VWI grade by Dr.S. d no correlation between Hb and VWI grade, e slight positive correlation between T-bil and VWI grade, and f no correlation between BMI and VWI grade by Dr.N. Hb hemoglobin, T-bil total bilirubin, BMI body mass index, VWI Vein-weighted image