not inf l uenced by confounding factors and is operator-independent.However, it may have applicability limitations due to claustrophobia,metal implants or pacemakers, or inability to suspend respiration forimage acquisition. Moreover, the availability and the costs should betaken into account when it is used for screening purpose.Controlled Attenuation Parameter (CAP) is available on the Fibro-Scan device. CAP measures the attenuation, in decibel/meter, of theUS beam that traverses the liver tissue. The performance of the CAPhas been assessed in cohort of patients and in a meta-analysis with indi-vidual patient’s data, taking the histologic grading of steatosis as thereference.Other new ultrasound-based techniques are nowadays available andtheir performance is under evaluation. These techniques are based onthe quantif i cation of the ultrasound attenuation in the tissue or the esti-mation of the speed of sound within soft tissues, that varies slightlywith fat content: an increase in fat content leads to a decrease in thespeed of sound.Grading fatty liver and detecting liver pathologicalfeatures - Is there consensus between sonographersand radiologists?David Sheng-Liang Yang, 1 Michal Schneider, 2 Paul Lombardo 21 Department of Ultrasound, Canberra Imaging Group, Canberra,ACT, Australia,2 Department of Medical Imaging and RadiationSciences, Monash University, Clayton, VIC, AustraliaIntroduction: Hepatic steatosis is the leading cause of chronic liverdisease. Reliable detection and staging of liver disease is important tofacilitate diagnosis and treatment. The aim of this study was to evaluatethe interobserver agreement between trainee sonographers, qualif i edsonographers and radiologists in grading non-alcoholic fatty liver dis-ease (NAFLD) and detecting common liver pathological features inBmode images.Methods: 150 B-mode liver ultrasound images from 50 adult patientsreferred for an abdominal ultrasound were obtained retrospectivelyfrom a PACS system. The images were independently graded for theseverity of hepatic steatosis (normal, mild, moderate or severe) and thedetection of incidental 0f i ndings, focal fatty sparing, liver surface irreg-ularity and rounded liver edge (present or absent) by 17 qualif i ed, sixtrainee sonographers and six radiologists. Fleiss’ kappa statistics wereused to calculate interobserver agreement among participants in grad-ing common liver pathological features. Intraclass correlation coeff i -cient (ICC) were used to calculate the level of absolute agreementamong participants in grading NAFLD.Results: The interobserver agreement rates among trainee sonogra-phers for the detection of incidental f i ndings, focal fatty sparing, liversurface irregularity and rounded liver edge were: k=0.243, 0.486,0.155 and 0.079 respectively. Among qualif i ed sonographers, theagreement rates were: k =0.323, 0.428, 0.167 and 0.152 respectively.Among radiologists, the agreement rates were: k=0.156, 0.266, 0.015and 0.154 respectively. The intraclass correlation coeff i cient averagescores among trainee sonographers, qualif i ed sonographers, all thesonographers combined and radiologists were: 0.927, 0.978, 0.985 and0.954 respectively.Conclusion: Visual assessment of common liver pathology in B-modeimaging has low interobserver agreement among sonographers andradiologists, but excellent inter-rater reliability in grading NAFLD.The low agreement levels are likely caused by a lack of standardisedassessment criteria. The development of a standardised criteria forstaging NAFLD and common liver pathological features arerecommended.Keywords: Hepatic steatosis, interobserver agreement, radiologists,sonographers, grading.Assessing hepatic steatosis on ultrasound imagingusing deep learningJennifer TangRadiology Registrar, The Royal Melbourne Hospital, Melbourne, VIC,AustraliaPurpose: To develop a deep learning algorithm which can assesshepatic steatosisMaterials and Methods: All abdominal ultrasounds from a tertiarycentre from January 2013 - November 2017 were reviewed. Studieswith known focal hepatobiliary pathology were excluded. Of the 1509candidate studies, 505 were considered suitable for inclusion. Fromeach study, four ultrasound images of the right lobe of the liver werechosen, two transverse and two longitudinal. Two radiologists and oneradiology fellow reviewed each panel of four images and assigned anumber of 1 (normal) - 7(severe steatosis). Subsequently, a consensuswas agreed upon by all three radiologists, which was used as the goldstandard. 2020 images were divided by patients into training, valida-tion and testing sets in a ratio of 8:1:1. Images were scaled to256£256. A modif i ed Densenet model pre-trained upon ImageNetwas trained and validated. Quadratic weighted Cohen’s Kappa waschosen as the evaluation metric.Results: The most common category was two (normal-mild steatosis),with 159 (32%) studies. 17 (3%) were in the least common categorywhich was category seven (severe steatosis). The radiologists achieveda quadratic weighted Cohen’s Kappa of 0.82 - 0.95 and the Densenetmodel achieved a quadratic weighted Cohen’s Kappa of 0.87. Themost commonly misclassif i ed category amongst the radiologists wasthree (mild) whilst the most commonly misclassif i ed categories for theDensenet model were tied at three and four (mild, mild-moderate).Conclusions: The deep learning algorithm performs similarly to sub-specialty trained radiologists in the assessment of hepatic steatosis onultrasound.Elastography of the spleenGiovanna FerraioliDepartment of “Scienze Clinico-Chirurgiche, Diagnostiche ePediatriche”, Medical School University of Pavia, Pavia, ItalySpleen stiffness (SS) might ref l ect portal pressure better than liverstiffness (LS). In fact, the severity of portal hypertension (PH) dependsalso on the increase in porto-systemic f l ow that cannot be measured byLS. PH leads to spleen congestion that increases the stiffness of theorgan; moreover, experimental data have shown that it induces splenicf i brosis.The repeatability of SS measurements has been assessed in healthysubjects, and it has been shown that the intra-observer and inter-observer agreement is higher for the liver than for the spleen, and thatthere is an improved repeatability of SS measurements at all sites withtraining. Also in patients with clinically signif i cant portal hypertension(CSPH) LS was more feasible than SS. In these patients, both LS andSS correlate with HVPG. SS can predict clinical complications in com-pensated cirrhosis It has been reported that, in these patients, SS has ahigh discriminative ability for predicting oesophageal variceal bleedingthan LS, with AUROCs of 0.86 for SS and 0.67 for LS. An algorithm torule out CSPH combining shear-wave elastography of liver and spleen,has been proposed.The Baveno VI consensus in PH suggests that patients with a liverstiffness <20 kPa and with a platelet count >150,000 have a very lowrisk of having varices requiring treatment, and can avoid screeningendoscopy. It has been shown that combining SS measurement withthe Baveno VI criteria a higher number of screening endoscopy may beavoided. LS and SS measurements may stratify the risk of varicealAbstracts S33
