Deep learning and digital pathology powers prediction of HCC development in steatotic liver disease.

BACKGROUND AND AIMS: Identifying patients with steatotic liver disease who are at a high risk of developing HCC remains challenging. We present a deep learning (DL) model to predict HCC development using hematoxylin and eosin-stained whole-slide images of biopsy-proven steatotic liver disease. APPROACH AND RESULTS: We included 639 patients who did not develop HCC for ≥7 years after biopsy (non-HCC class) and 46 patients who developed HCC <7 years after biopsy (HCC class). Paired cases of the HCC and non-HCC classes matched by biopsy date and institution were used for training, and the remaining nonpaired cases were used for validation. The DL model was trained using deep convolutional neural networks with 28,000 image tiles cropped from whole-slide images of the paired cases, with an accuracy of 81.0% and an AUC of 0.80 for predicting HCC development. Validation using the nonpaired cases also demonstrated a good accuracy of 82.3% and an AUC of 0.84. These results were comparable to the predictive ability of logistic regression model using fibrosis stage. Notably, the DL model also detected the cases of HCC development in patients with mild fibrosis. The saliency maps generated by the DL model highlighted various pathological features associated with HCC development, including nuclear atypia, hepatocytes with a high nuclear-cytoplasmic ratio, immune cell infiltration, fibrosis, and a lack of large fat droplets. CONCLUSIONS: The ability of the DL model to capture subtle pathological features beyond fibrosis suggests its potential for identifying early signs of hepatocarcinogenesis in patients with steatotic liver disease.

Effect of bone density on the drill-hole diameter made by a cannulated drill bit in cancellous bone.

BACKGROUND: When a pilot hole is made prior to a screw's insertion into bone, the same drill bit is used irrespective of the bone quality. However, osteoporotic bone is fragile and this may affect the hole diameter, which is of particular concern in cancellous bone. In this study, the relationship between bone density and drill-hole diameter was investigated assuming a pre-drilling process in screw-only osteosynthesis in the metaphysis and epiphysis. METHODS: Two types of drill bit (triple-flute [T] and quadruple-flute [Q]) with different shapes and diameters were prepared: type T bits with 3.5 mm and 4.4 mm diameters, and type Q bits with 3.5 mm and 4.2 mm diameters. Drilling was performed manually in simulated bones with four densities: 5, 10, 15, and 20 pounds per cubic foot. We measured the hole diameters with a coordinate measuring machine and analyzed the relationship between the drill-hole diameters and the densities of the simulated bones. We then compared the screw pull-out strength between the two 3.5-diameter drill bits. RESULTS: In all cases, the diameters of the drill holes were larger than those of the drill bits. The relationship between the drill-hole diameters and the bone densities was a negative linear correlation. Enlarging the hole diameter decreased the screw pull-out strength. CONCLUSIONS: For cannulated drill bits of 3.5, 4.2 and 4.4 mm diameter, the diameter of the drill hole in cancellous bone obtained by the manual drilling technique tends to be larger in low-density (e.g., osteoporotic) compared to high-density (e.g., healthy) bone.

The impact of COVID-19 on the diagnosis and treatment of HCC: analysis of a nationwide registry for advanced liver diseases (REAL).

The number of cancer cases diagnosed during the coronavirus disease 2019 (COVID-19) pandemic has decreased. This study investigated the impact of the pandemic on the clinical practice of hepatocellular carcinoma (HCC) using a novel nationwide REgistry for Advanced Liver diseases (REAL) in Japan. We retrieved data of patients initially diagnosed with HCC between January 2018 and December 2021. We adopted tumor size as the primary outcome measure and compared it between the pre-COVID-19 (2018 and 2019) and COVID-19 eras (2020 and 2021). We analyzed 13,777 patients initially diagnosed with HCC (8074 in the pre-COVID-19 era and 5703 in the COVID-19 era). The size of the maximal intrahepatic tumor did not change between the two periods (mean [SD] = 4.3 [3.6] cm and 4.4 [3.6] cm), whereas the proportion of patients with a single tumor increased slightly from 72.0 to 74.3%. HCC was diagnosed at a similar Barcelona Clinic Liver Cancer stage. However, the proportion of patients treated with systemic therapy has increased from 5.4 to 8.9%. The proportion of patients with a non-viral etiology significantly increased from 55.3 to 60.4%. Although the tumor size was significantly different among the etiologies, the subgroup analysis showed that the tumor size did not change after stratification by etiology. In conclusion, the characteristics of initially diagnosed HCC remained unchanged during the COVID-19 pandemic in Japan, regardless of differences in etiology. A robust surveillance system should be established particularly for non-B, non-C etiology to detect HCC in earlier stages.

Development of a transformer model for predicting the prognosis of patients with hepatocellular carcinoma after radiofrequency ablation.

INTRODUCTION: Radiofrequency ablation (RFA) is a widely accepted, minimally invasive treatment modality for patients with hepatocellular carcinoma (HCC). Accurate prognosis prediction is important to identify patients at high risk for cancer progression/recurrence after RFA. Recently, state-of-the-art transformer models showing improved performance over existing deep learning-based models have been developed in several fields. This study was aimed at developing and validating a transformer model to predict the overall survival in HCC patients with treated by RFA. METHODS: We enrolled a total of 1778 treatment-naïve HCC patients treated by RFA as the first-line treatment. We developed a transformer-based machine learning model to predict the overall survival in the HCC patients treated by RFA and compared its predictive performance with that of a deep learning-based model. Model performance was evaluated by determining the Harrel's c-index and validated externally by the split-sample method. RESULTS: The Harrel's c-index of the transformer-based model was 0.69, indicating its better discrimination performance than that of the deep learning model (Harrel's c-index, 0.60) in the external validation cohort. The transformer model showed a high discriminative ability for stratifying the external validation cohort into two or three different risk groups (p < 0.001 for both risk groupings). The model also enabled output of a personalized cumulative recurrence prediction curve for each patient. CONCLUSIONS: We developed a novel transformer model for personalized prediction of the overall survival in HCC patients after RFA treatment. The current model may offer a personalized survival prediction schema for patients with HCC undergoing RFA treatment.

Fatal Severe Fever with Thrombocytopenia Syndrome Virus and Pasteurella multocida Coinfection.

We herein report a case of severe fever with thrombocytopenia syndrome (SFTS) with Pasteurella multilocida bacteremia in a 65-year-old man with alcoholic cirrhosis who was admitted to our hospital with anorexia and severe fatigue. Laboratory tests revealed pancytopenia and liver and kidney dysfunction. After admission, he developed impaired consciousness, mucosal hemorrhaging, and septic shock. SFTS virus was detected on polymerase chain reaction testing of blood and throat swabs, and Pasteurella multocida was detected on blood culture. Despite being treated with invasive mechanical ventilation, vasopressors, and antibiotics, the patient's condition progressively deteriorated, and he died four days after admission.

Proton Beam Therapy versus Radiofrequency Ablation for Patients with Treatment-Naïve Single Hepatocellular Carcinoma: A Propensity Score Analysis.

Introduction: Proton beam therapy (PBT) is known to be an effective locoregional treatment for hepatocellular carcinoma (HCC). However, few comparative studies in treatment-naïve cases have been reported. The aim of this study was to compare the survival outcomes of PBT with those of radiofrequency ablation (RFA) in patients with treatment-naïve solitary HCC. Methods: Ninety-five consecutive patients with treatment-naïve HCC, a single nodule measuring ≤5 cm in diameter, and a Child-Pugh score of ≤8 who were treated with PBT at the University of Tsukuba Hospital between 2001 and 2013 were enrolled in the study. In addition, 836 patients with treatment-naïve HCC treated by RFA at the University of Tokyo Hospital during the same period were analyzed as controls. Recurrence-free survival (RFS) and overall survival (OS) were compared in 83 patient pairs after propensity score matching. Results: The 1-year, 3-year, and 5-year RFS rates were 86.6%, 49.5%, and 35.5%, respectively, in the PBT group and 59.5%, 34.0%, and 20.9% in the RFA group (p = 0.058); the respective OS rates were 97.6%, 77.8%, and 57.1% in the PBT group and 95.1%, 81.7%, and 67.7% in the RFA group (p = 0.16). Regarding adverse effects, no grade 3 or higher adverse events were noted in the PBT; however, two grade 3 adverse events occurred within 30 days of RFA in the RFA group: one hemoperitoneum and one hemothorax. Discussion: After propensity score matching, PBT showed no significant difference in RFS and OS compared to RFA. PBT can be an alternative for patients with solitary treatment-naïve HCC.

Machine learning for individualized prediction of hepatocellular carcinoma development after the eradication of hepatitis C virus with antivirals.

BACKGROUND AND AIMS: Accurate risk stratification for hepatocellular carcinoma (HCC) after achieving a sustained viral response (SVR) is necessary for optimal surveillance. We aimed to develop and validate a machine learning (ML) model to predict the risk of HCC after achieving an SVR in individual patients. METHODS: In this multicenter cohort study, 1742 patients with chronic hepatitis C who achieved an SVR were enrolled. Five ML models were developed including DeepSurv, gradient boosting survival analysis, random survival forest (RSF), survival support vector machine, and a conventional Cox proportional hazard model. Model performance was evaluated using Harrel' c-index and was externally validated in an independent cohort (977 patients). RESULTS: During the mean observation period of 5.4 years, 122 patients developed HCC (83 in the derivation cohort and 39 in the external validation cohort). The RSF model showed the best discrimination ability using seven parameters at the achievement of an SVR with a c-index of 0.839 in the external validation cohort and a high discriminative ability when the patients were categorized into three risk groups (P <0.001). Furthermore, this RSF model enabled the generation of an individualized predictive curve for HCC occurrence for each patient with an app available online. CONCLUSIONS: We developed and externally validated an RSF model with good predictive performance for the risk of HCC after an SVR. The application of this novel model is available on the website. This model could provide the data to consider an effective surveillance method. Further studies are needed to make recommendations for surveillance policies tailored to the medical situation in each country. IMPACT AND IMPLICATIONS: A novel prediction model for HCC occurrence in patients after hepatitis C virus eradication was developed using machine learning algorithms. This model, using seven commonly measured parameters, has been shown to have a good predictive ability for HCC development and could provide a personalized surveillance system.

Clinical utility of postablation liver tumor biopsy and possibility of gene mutation analysis.

AIM: Radiofrequency ablation (RFA) is regarded as a first-line treatment for hepatocellular carcinoma (HCC) at an early stage. When treated with RFA, tumor biopsy may not be performed due to the risk of neoplastic seeding. We previously revealed that the risk of neoplastic seeding is significantly reduced by performing biopsies after RFA. In this study, we investigated the possibility of pathological evaluation and gene mutation analysis of post-RFA tumor specimens. METHODS: Radiofrequency ablation was undertaken on diethylnitrosamine-induced mouse liver tumor, and tumor samples with or without RFA were subjected to whole exome sequencing. Post-RFA human liver tumor specimens were used for detection of TERT promoter mutations and pathological assessment. RESULTS: The average somatic mutation rate, sites of mutation, and small indels and base transition patterns were comparable between the nontreated and post-RFA tumors. We identified 684 sites of nonsynonymous somatic substitutions in the nontreated tumor and 704 sites of nonsynonymous somatic substitutions in the post-RFA tumor, with approximately 85% in common. In the human post-RFA samples, the TERT promoter mutations were successfully detected in 40% of the cases. Pathological evaluation was possible with post-RFA specimens, and in one case, the diagnosis of adenocarcinoma was made. CONCLUSION: Our findings suggest that post-RFA liver tumor biopsy is a useful and safe method for obtaining tumor samples that can be used for gene mutation analysis and for pathological assessment.

Treatment and outcome of hepatorenal syndrome in Japan: a retrospective cohort study using a national inpatient database.

BACKGROUND: Hepatorenal syndrome (HRS) is a life-threatening complication of end-stage liver disease. This study aimed to clarify the status of HRS in Japan by analyzing the Japanese Diagnosis Procedure Combination database. METHODS: Patients hospitalized for cirrhosis and HRS from July 2010 to March 2019 were sampled. They were divided into two groups according to their prognosis upon discharge: the transplant-free survival group and the death or liver transplantation group. The two groups' baseline patient characteristics and treatments were compared. RESULTS: The mean age of the 1,412 participants was 67.3 years (standard deviation: 12.3 years), and 65.4% were male. The Child-Pugh grades was B and C in 18.8% and 81.2%, respectively. Hepatocellular carcinoma was present in 27.1% of the patients, and the proportion of spontaneous bacterial peritonitis was 2.3%. Albumin, noradrenaline, and dopamine were administered to 57.9%, 8.0%, and 14.9% of the patients, respectively; 7.0% of the patients underwent renal replacement therapy; and 5.0% were admitted to the intensive care unit. Intravenous antibiotics were administered to 30.8% of the patients. A total of 925 patients (65.5%) died or underwent liver transplantation. In addition to a higher proportion of patients with poor baseline liver function, the death or liver transplantation group included more males, patients with hepatocellular carcinoma, and those with spontaneous bacterial peritonitis. CONCLUSIONS: HRS in Japan has a high mortality rate. Albumin was administered to over 50% of participants. Although noradrenaline is recommended in Japanese clinical guidelines, dopamine was more frequently used as a vasoconstrictor in clinical practice.

Steroid administration for ischemic complications after radiofrequency ablation: A retrospective study.

AIM: The aim of this study was to evaluate the effects of steroids on ischemic complications after radiofrequency ablation. METHODS: A total of 58 patients with ischemic complications were divided into two groups according to corticosteroid use or non-use. RESULTS: A total of 13 patients who were administered steroids had a shorter duration of fever than those who were not administered steroids (median 6.0 vs. 2.0 days; p < 0.001). Linear regression analysis showed that steroid administration was associated with a reduction of 3.9 days in the duration of fever (p = 0.008). CONCLUSIONS: Steroid administration for ischemic complications after radiofrequency ablation may reduce the risk of fatal outcomes by blocking systemic inflammatory reactions.

Impact of a Novel Digital Therapeutics System on Nonalcoholic Steatohepatitis: The NASH App Clinical Trial.

INTRODUCTION: Management of nonalcoholic steatohepatitis (NASH) is a currently unmet clinical need. Digital therapeutics (DTx) is an emerging class of medicine that delivers evidence-based therapeutic interventions. This study was aimed at investigating the efficacy of DTx in patients with NASH. METHODS: We conducted a multicenter, single-arm, 48-week trial in 19 patients with biopsy-confirmed NASH. All patients received a DTx intervention with a newly developed smartphone application. The primary endpoint was change in the nonalcoholic fatty liver disease activity score (NAS) without worsening of liver fibrosis. The secondary endpoints included improvement of the NAS by ≥2 points without worsening of liver fibrosis, change in the body weight, and regression of fibrosis. RESULTS: After the 48-week DTx intervention, improvement of the NAS was observed in 68.4% (13/19) of patients. The mean change in the NAS from baseline to the end of the intervention was -2.05 ± 1.96 ( P < 0.001 when compared with the threshold of -0.7). A decrease in the NAS by ≥ 2 points was achieved in 11 (57.9%). The average weight loss at the end of the intervention was 8.3% ( P < 0.001). Reduction of the fibrosis stage was observed in 58.3% when the analysis was limited to patients with stage F2/3 fibrosis. There were no serious adverse events that could be considered as being related to the DTx intervention. DISCUSSION: DTx for NASH was found to be highly efficacious and well-tolerated. Further evaluation of the DTx intervention for NASH in a phase 3 trial is warranted.

Trend in industry payments to infectious disease physicians in the United States: a seven-year analysis of nonresearch payments from the Open Payments Database between 2014 and 2020.

OBJECTIVE: To evaluate the trend in nonresearch payments made by the industries to the infectious disease physicians in the United States since the launch of the Open Payments Database and during the COVID-19 pandemic. METHODS: Descriptive analysis was performed for the nonresearch payments made to all infectious disease physicians listed in the Open Payments Database between 2014 and 2020. Using the generalized estimating equation models with panel data of monthly and yearly payment per physician, the payment trend since the inception of the Open Payments Database and during the early stage of the COVID-19 pandemic were evaluated. RESULTS: A total of 7901 (81.5%) infectious disease physicians received $156 837 987 in nonresearch payments between 2014 and 2020. Median annual payments were $197 to $220. Monthly nonresearch per-physician payments and number of physicians with payments rapidly decreased by 58.6% (95% CI: 49.7%‒65.9%, p < 0.001) and by 54.4% (95% CI: 52.7%‒56.1%, p < 0.001) at the beginning of the COVID-19 pandemic, respectively. However, the per-physician payments and number of physicians with payments slightly increased every month right after onset of the pandemic. Both per-physician payments and the number of physicians with payments decreased by 2.6% (95% CI: 0.45‒4.7, p 0.018) and 2.0% (95% CI: 1.6%‒2.4%, p < 0.001) since the inception of the Open Payments Database, respectively. DISCUSSION: The nonresearch payments and number of infectious disease physicians accepting payments had decreased since the inception of the Open Payments Database. Furthermore, the non-research payments to infectious disease physicians suddenly decreased by more than half due to the COVID-19 pandemic.

Redox state as assessed using the measurement of human non-mercaptalbumin in embryo culture media is associated with successful embryo development in human in vitro fertilization.

The role of oxidative stress in the pathogenesis of various diseases has been attracting attention. We speculated as to whether the redox state of treatment solutions used for various diseases may play a role in treatment success. In the current study, we focused on the human embryo culture medium used for in vitro fertilization (IVF). A total of 173 oocytes from a total of 91 patients treated with IVF were enrolled. The redox state was assessed by measuring the levels of human non-mercaptalbumin (HNA). We analyzed factors related to blastocyst formation on day 5 or 6 after insemination. We also developed a random forest (RF) model for the prediction of blastocyst formation. The variable importance in the predictive model was assessed using the mean decrease in the Gini impurity. Blastocyst formation was observed in 41.04% (71/173) of the oocytes and was associated with a lower %HNA in the culture medium, a younger patient age, and the fertilization method (standard IVF or intracytoplasmic sperm injection). The RF model developed using these factors and 70% of the samples (training set, n = 121) was validated in the remaining testing set (n = 52) and produced an area under the curve of 0.761, where the %HNA in the culture medium was the most important variable for predicting blastocyst formation. In conclusion, lower levels of oxidative stress in embryo culture media were associated with the success of IVF treatment. The redox state of treatment solutions should be considered to support treatment success.

Current status of primary liver cancer and decompensated cirrhosis in Japan: launch of a nationwide registry for advanced liver diseases (REAL).

BACKGROUND: We developed a nationwide database that stores data of patients with primary liver cancer (PLC) and decompensated cirrhosis (DC) on an admission basis. METHODS: A database was constructed using the National Clinical Database, a nationwide registry platform for various diseases in Japan. Mutual data exchange was possible with the Nationwide Follow-up Survey of Primary Liver Cancer in Japan by the Liver Cancer Study Group of Japan. The stored data on the admission of patients with PLC, DC, or both, included treatment details as well as patient characteristics. RESULTS: A total of 37,705 admissions (29,489 PLC, 10,077 DC, and 1862 for both) in 21,376 patients from 224 hospitals were analyzed. The proportions of patients with hepatitis B, hepatitis C, and non-viral etiology were 11.9%, 36.2%, and 42.6%, respectively, in PLC, and 7.5%, 23.8%, and 55.0%, respectively, in DC. The mean ages (± standard deviation) on admission with PLC and DC were 73 ± 10 and 68 ± 13 years, respectively. The Barcelona Clinic Liver Cancer (BCLC) stage for PLC was 0, A, B, C, and D in 22.0%, 17.1%, 29.6%, 15.1%, and 5.1%, respectively. Treatment modalities for PLC were resection, ablation, transarterial chemoembolization, and systemic therapy in 18.4%, 22.8%, 33.7%, and 11.4%, respectively. A vasopressin receptor V2 antagonist was used in 38.2% in addition to conventionally used loop diuretics and aldosterone antagonists for DC. CONCLUSIONS: The distribution of treatment options for PLC on admission differed from that of the initial treatment. Newly introduced drugs are widely used in patients with DC.

Improved prognosis of hepatitis C-related hepatocellular carcinoma in the era of direct-acting antivirals.

The prognostic impact of direct-acting antivirals (DAAs) on patients with hepatitis C-related hepatocellular carcinoma (C-HCC) is still unclear. This study aimed to evaluate the prognosis of C-HCC in the DAA era. We enrolled 1237 consecutive patients with treatment-naive C-HCC who underwent radical radiofrequency ablation between 1999 and 2019. We also enrolled 350 patients with nonviral HCC as controls. We divided these patients into three groups according to the year of initial treatment: 1999-2005 (cohort 1), 2006-2013 (cohort 2), and 2014-2019 (cohort 3). The use of antiviral agents and their effect in patients with C-HCC was investigated. Overall survival was evaluated for each cohort using the Kaplan-Meier method and a multivariable Cox proportional hazards regression model. Sustained virologic response (SVR) was achieved in 52 (10%), 157 (26%), and 102 (74%) patients with C-HCC in cohorts 1-3, respectively. The 3- and 5-year survival rates of patients with C-HCC were 82% and 59% in cohort 1; 80% and 64% in cohort 2; and 86% and 78% in cohort 3, respectively (p = 0.003). Multivariable analysis adjusted for age, liver function, and tumor extension showed that the prognosis of C-HCC improved in cohort 3 compared to cohort 1 (adjusted hazard ratio [aHR], 0.49; 95% confidence interval [CI], 0.32-0.73; p < 0.001), whereas the prognosis of nonviral HCC did not improve significantly (aHR, 0.96; 95% CI, 0.59-1.57; p = 0.88). The prognosis of C-HCC drastically improved with the advent of DAAs.

Inhibiting SCAP/SREBP exacerbates liver injury and carcinogenesis in murine nonalcoholic steatohepatitis.

Enhanced de novo lipogenesis mediated by sterol regulatory element-binding proteins (SREBPs) is thought to be involved in nonalcoholic steatohepatitis (NASH) pathogenesis. In this study, we assessed the impact of SREBP inhibition on NASH and liver cancer development in murine models. Unexpectedly, SREBP inhibition via deletion of the SREBP cleavage-activating protein (SCAP) in the liver exacerbated liver injury, fibrosis, and carcinogenesis despite markedly reduced hepatic steatosis. These phenotypes were ameliorated by restoring SREBP function. Transcriptome and lipidome analyses revealed that SCAP/SREBP pathway inhibition altered the fatty acid (FA) composition of phosphatidylcholines due to both impaired FA synthesis and disorganized FA incorporation into phosphatidylcholine via lysophosphatidylcholine acyltransferase 3 (LPCAT3) downregulation, which led to endoplasmic reticulum (ER) stress and hepatocyte injury. Supplementation with phosphatidylcholines significantly improved liver injury and ER stress induced by SCAP deletion. The activity of the SCAP/SREBP/LPCAT3 axis was found to be inversely associated with liver fibrosis severity in human NASH. SREBP inhibition also cooperated with impaired autophagy to trigger liver injury. Thus, excessively strong and broad lipogenesis inhibition was counterproductive for NASH therapy; this will have important clinical implications in NASH treatment.

Lipid Profiles of Human Serum Fractions Enhanced with CD9 Antibody-Immobilized Magnetic Beads.

Blood samples are minimally invasive and can be collected repeatedly, but they are far from the site of disease and the target molecules are diluted by the large amount of blood. Therefore, we performed lipidomics using immunoprecipitation as a method to enrich specific fractions of serum. In this study, a CD9 antibody was immobilized on magnetic beads to enrich CD9-containing components in the serum for lipidomics. The percentages of phospholipids recovered from serum by methanol and isopropanol extractions were not significantly different, but triglycerides were barely recovered from serum by methanol extraction, requiring the use of isopropanol. However, once the serum was enriched with CD9 magnetic beads, triglycerides, and phospholipids were recovered at similar levels in both methanol and isopropanol extractions. Therefore, it is possible that the triglyceride fraction of the whole serum and the triglyceride fraction were enriched in CD9 magnetic beads differ in localization and properties. In addition, the variation per disease was small in general serum lipidomics; however, the difference per disease appeared larger when CD9 magnetic bead enrichment was employed.

Artificial intelligence (AI) models for the ultrasonographic diagnosis of liver tumors and comparison of diagnostic accuracies between AI and human experts.

BACKGROUND: Ultrasonography (US) is widely used for the diagnosis of liver tumors. However, the accuracy of the diagnosis largely depends on the visual perception of humans. Hence, we aimed to construct artificial intelligence (AI) models for the diagnosis of liver tumors in US. METHODS: We constructed three AI models based on still B-mode images: model-1 using 24,675 images, model-2 using 57,145 images, and model-3 using 70,950 images. A convolutional neural network was used to train the US images. The four-class liver tumor discrimination by AI, namely, cysts, hemangiomas, hepatocellular carcinoma, and metastatic tumors, was examined. The accuracy of the AI diagnosis was evaluated using tenfold cross-validation. The diagnostic performances of the AI models and human experts were also compared using an independent test cohort of video images. RESULTS: The diagnostic accuracies of model-1, model-2, and model-3 in the four tumor types are 86.8%, 91.0%, and 91.1%, whereas those for malignant tumor are 91.3%, 94.3%, and 94.3%, respectively. In the independent comparison of the AIs and physicians, the percentages of correct diagnoses (accuracies) by the AIs are 80.0%, 81.8%, and 89.1% in model-1, model-2, and model-3, respectively. Meanwhile, the median percentages of correct diagnoses are 67.3% (range 63.6%-69.1%) and 47.3% (45.5%-47.3%) by human experts and non-experts, respectively. CONCLUSION: The performance of the AI models surpassed that of human experts in the four-class discrimination and benign and malignant discrimination of liver tumors. Thus, the AI models can help prevent human errors in US diagnosis.