Applying Machine Learning in Liver Disease and Transplantation: A Comprehensive Review

Clinical and Radiological Fusion: A New Frontier in Predicting Post-Transplant Diabetes Mellitus

This study developed a predictive model for Post-Transplant Diabetes Mellitus (PTDM) by integrating clinical and radiological data to identify at-risk kidney transplant recipients. In a retrospective analysis across three Mayo Clinic sites, clinical metrics were combined with deep learning analysis of pre-transplant CT images, focusing on body composition parameters like adipose tissue and muscle mass instead of BMI or other biomarkers. Among 2,005 nondiabetic kidney recipients, 335 (16.7%) developed PTDM within the first year. PTDM patients were older, had higher BMIs, elevated triglycerides, and were more likely to be male and non-White. They exhibited lower skeletal muscle area, greater visceral adipose tissue (VAT), more intermuscular fat, and higher subcutaneous fat (all p < 0.001). Multivariable analysis identified age (OR: 1.05, 95% CI: 1.03-1.08, p < 0.0001), family diabetes history (OR: 1.55, CI: 1.14-2.09, p = 0.0061), White race (OR: 0.43, CI: 0.28-0.66, p < 0.0001), and VAT area (OR: 1.37, CI: 1.14-1.64, p = 0.0009) as predictors. The combined model achieved C-statistic of 0.724 (CI: 0.692-0.757), outperforming the clinical-only model (C-statistic 0.68). Patients with PTDM in the first year had higher mortality than those without PTDM. This model improves predictive precision, enabling accurate identification and intervention for at risk patients.

Machine Learning Prediction Model of Waitlist Outcomes in Patients with Primary Sclerosing Cholangitis

Background

Liver transplantation is essential for many people with primary sclerosing cholangitis (PSC). People with PSC are less likely to receive a deceased donor liver transplant compared with other causes of chronic liver disease. This disparity may stem from the inaccuracy of the model for end-stage liver disease (MELD) in predicting waitlist mortality or dropout for PSC. The broad applicability of MELD across many causes comes at the expense of accuracy in prediction for certain causes that involve unique comorbidities. We aimed to develop a model that could more accurately predict dynamic changes in waitlist outcomes among patients with PSC while including complex clinical variables.

Methods

We developed 3 machine learning architectures using data from 4666 patients with PSC in the Scientific Registry of Transplant Recipients (SRTR) and tested our models on our institutional data set of 144 patients at the University Health Network (UHN). We evaluated their time-dependent concordance index (C-index) for mortality prediction and compared it against MELD-sodium and MELD 3.0.

Results

Random survival forest (RSF), a decision tree-based survival model, outperformed MELD-sodium and MELD 3.0 in both the SRTR and the UHN test data set using the same bloodwork variables and readily available demographic data. It achieved a C-index of 0.868 (SD 0.020) and 0.771 (SD 0.085) on the SRTR and UHN test data, respectively. Training a separate RSF model using the UHN data with PSC-specific achieved a C-index of 0.91. In addition to high MELD score, increased white blood cells, time on the waiting list, platelet count, presence of Autoimmune hepatitis-PSC overlap, aspartate aminotransferase, female sex, age, history of stricture dilation, and extremes of body weight were the top-ranked features predictive of the outcomes.

Conclusions

Our RSF model offers more accurate waitlist outcome prediction in PSC. The significant performance improvement with the inclusion of PSC-specific variables highlights the importance of disease-specific variables for predicting trajectories of clinically distinct presentations.

Clinical decision support and electronic interventions to improve care quality in chronic liver diseases and cirrhosis

Significant quality gaps exist in the management of chronic liver diseases and cirrhosis. Clinical decision support systems-information-driven tools based in and launched from the electronic health record-are attractive and potentially scalable prospective interventions that could help standardize clinical care in hepatology. Yet, clinical decision support systems have had a mixed record in clinical medicine due to issues with interoperability and compatibility with clinical workflows. In this review, we discuss the conceptual origins of clinical decision support systems, existing applications in liver diseases, issues and challenges with implementation, and emerging strategies to improve their integration in hepatology care.

A Supervised Explainable Machine Learning Model for Perioperative Neurocognitive Disorder in Liver-Transplantation Patients and External Validation on the Medical Information Mart for Intensive Care IV Database: Retrospective Study

BACKGROUND

Patients undergoing liver transplantation (LT) are at risk of perioperative neurocognitive dysfunction (PND), which significantly affects the patients' prognosis.

OBJECTIVE

This study used machine learning (ML) algorithms with an aim to extract critical predictors and develop an ML model to predict PND among LT recipients.

METHODS

In this retrospective study, data from 958 patients who underwent LT between January 2015 and January 2020 were extracted from the Third Affiliated Hospital of Sun Yat-sen University. Six ML algorithms were used to predict post-LT PND, and model performance was evaluated using area under the receiver operating curve (AUC), accuracy, sensitivity, specificity, and F1-scores. The best-performing model was additionally validated using a temporal external dataset including 309 LT cases from February 2020 to August 2022, and an independent external dataset extracted from the Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) database including 325 patients.

RESULTS

In the development cohort, 201 out of 751 (33.5%) patients were diagnosed with PND. The logistic regression model achieved the highest AUC (0.799) in the internal validation set, with comparable AUC in the temporal external (0.826) and MIMIC-Ⅳ validation sets (0.72). The top 3 features contributing to post-LT PND diagnosis were the preoperative overt hepatic encephalopathy, platelet level, and postoperative sequential organ failure assessment score, as revealed by the Shapley additive explanations method.

CONCLUSIONS

A real-time logistic regression model-based online predictor of post-LT PND was developed, providing a highly interoperable tool for use across medical institutions to support early risk stratification and decision making for the LT recipients.

The impact of artificial intelligence and machine learning in organ retrieval and transplantation: A comprehensive review

This narrative review examines the transformative role of Artificial Intelligence (AI) and Machine Learning (ML) in organ retrieval and transplantation. AI and ML technologies enhance donor-recipient matching by integrating and analyzing complex datasets encompassing clinical, genetic, and demographic information, leading to more precise organ allocation and improved transplant success rates. In surgical planning, AI-driven image analysis automates organ segmentation, identifies critical anatomical features, and predicts surgical outcomes, aiding pre-operative planning and reducing intraoperative risks. Predictive analytics further enable personalized treatment plans by forecasting organ rejection, infection risks, and patient recovery trajectories, thereby supporting early intervention strategies and long-term patient management. AI also optimizes operational efficiency within transplant centers by predicting organ demand, scheduling surgeries efficiently, and managing inventory to minimize wastage, thus streamlining workflows and enhancing resource allocation. Despite these advancements, several challenges hinder the widespread adoption of AI and ML in organ transplantation. These include data privacy concerns, regulatory compliance issues, interoperability across healthcare systems, and the need for rigorous clinical validation of AI models. Addressing these challenges is essential to ensuring the reliable, safe, and ethical use of AI in clinical settings. Future directions for AI and ML in transplantation medicine include integrating genomic data for precision immunosuppression, advancing robotic surgery for minimally invasive procedures, and developing AI-driven remote monitoring systems for continuous post-transplantation care. Collaborative efforts among clinicians, researchers, and policymakers are crucial to harnessing the full potential of AI and ML, ultimately transforming transplantation medicine and improving patient outcomes while enhancing healthcare delivery efficiency.

Predicting Immune Flares in Untreated Chronic Hepatitis B Patients Using Novel Risk Factors and the FLARE-B Score

BACKGROUND AND AIMS

Risk factors of chronic hepatitis B (CHB) immune flares are poorly understood. The primary aim of this study was to discover predictors of the CHB flare in non-cirrhotic, untreated CHB patients and develop a simple risk-stratifying score to predict the CHB flare. The secondary aim was to compare different machine learning methods for prediction.

METHODS

A retrospective cohort of untreated, non-cirrhotic CHB patients with normal baseline ALT was followed up over time until an immune flare as defined by ALT twice the upper limit of normal. Statistical learning and machine learning algorithms were used to develop predictive models using baseline variables. Bootstrap validation was used to internally validate the models.

RESULTS

Of 405 patients (median age 44y; 41% male, 10% HBeAg positive), 67 (17%) experienced an immune flare by 5 years (annual incidence 4.0%). Predictors of flare included raised serum globulin, younger age, HBeAg positive status, higher viral load and raised liver stiffness. A simple predictive model "FLARE-B" had optimism-adjusted 1, 3 and 5-year AUCs of 0.813, 0.728 and 0.702, respectively. The random survival forest algorithm had the highest optimism-adjusted AUCs of 0.861, 0.766 and 0.725, respectively.

CONCLUSIONS

New, novel predictors of the CHB flare include a raised serum globulin and possibly raised liver stiffness and the absence of liver steatosis. FLARE-B can be used to risk-stratify individuals and potentially guide personalized management strategies such as monitoring schedules and proactive antiviral treatment in high-risk patients.

The Aachen ACLF ICU score predicts ICU mortality in critically ill patients with acute-on-chronic liver failure

Acute-on-chronic liver failure (ACLF) defines a heterogeneous syndrome involving acute decompensation in patients with pre-existing liver disease accompanied by (multi-)organ failure. This study aimed to develop a simple, reliable machine learning (ML) model to predict mortality in ACLF patients receiving intensive care unit (ICU) treatment. Data from 206 patients admitted to the ICU at RWTH Aachen University Hospital between 2015 and 2021 were retrospectively analyzed with ICU mortality as the primary outcome. An ICU mortality prediction model was developed by logistic regression and validated by 5-fold cross validation. Performance metrics were assessed to evaluate the model's accuracy and compare to existing mortality scores. ICU mortality was 60%. The chronic-liver-failure-consortium ACLF score (CLIF-C ACLFs) was the best predictor of ICU mortality. ML generated seven models using five to thirteen features. The best-performing model included CLIF-C ACLFs, number of organ failures, Horovitz quotient (FiO2/PaO2), FiO2 and lactate. The newly developed Aachen ACLF ICU (ACICU) score demonstrated exceptional predictive accuracy for ICU mortality (AUROC 0.96), underscoring its potential for mortality and futility assessment in critically ill ACLF patients complementing existing prognostic tools. The ACICU score www.acicu-score.com is an easy-to-use tool for predicting ICU mortality in patients with ACLF offering high predictive performance.

Insights into ALD and AUD diagnosis and prognosis: Exploring AI and multimodal data streams

The rapid evolution of artificial intelligence and the widespread embrace of digital technologies have ushered in a new era of clinical research and practice in hepatology. Although its potential is far from realization, these significant strides have generated new opportunities to address existing gaps in the delivery of care for patients with liver disease. In this review, we discuss how artificial intelligence and opportunities for multimodal data integration can improve the diagnosis, prognosis, and management of alcohol-associated liver disease. An emphasis is made on how these approaches will also benefit the detection and management of alcohol use disorder. Our discussion encompasses challenges and limitations, concluding with a glimpse into the promising future of these advancements.

Machine-learning methodologies to predict disease progression in chronic hepatitis B in Africa

BACKGROUND

Little is known about the determinants of disease progression among African patients with chronic HBV infection.

METHODS

We used machine-learning models with longitudinal data to establish predictive algorithms in a well-characterized cohort of Ethiopian HBV-infected patients without baseline liver fibrosis. Disease progression was defined as an increase in liver stiffness to >7.9 kPa or initiation of treatment based on meeting the eligibility criteria.

RESULTS

Twenty-four of 551 patients (4.4%) experienced disease progression after a median follow-up time of 69 months. A random forest model based on a combination of available laboratory tests (standard hematology and biochemistry) demonstrated the best predictive properties with the AUROC ranging from 0.82 to 0.88.

CONCLUSION

We conclude that combined metrics based on simple and available laboratory tests had good predictive properties and should be explored further in larger HBV cohorts.

A Machine Learning Algorithm Avoids Unnecessary Paracentesis for Exclusion of SBP in Cirrhosis in Resource-limited Settings

BACKGROUND & AIMS

Despite the poor prognosis associated with missed or delayed spontaneous bacterial peritonitis (SBP) diagnosis, <15% get timely paracentesis, which persists despite guidelines/education in the United States. Measures to exclude SBP non-invasively where timely paracentesis cannot be performed could streamline this burden.

METHODS

Using Veterans Health Administration Corporate Data Warehouse (VHA-CDW) we included patients with cirrhosis between 2009 and 2019 who underwent timely paracentesis and collected relevant clinical information (demographics, cirrhosis severity, medications, vitals, and comorbidities). XGBoost-models were trained on 75% of the primary cohort, with 25% reserved for testing. The final model was further validated in 2 cohorts: Validation cohort #1: In VHA-CDW, those without prior SBP who received 2nd early paracentesis, and Validation cohort #2: Prospective data from 276 non-electively admitted University hospital patients.

RESULTS

Negative predictive values (NPVs) at 5%,10%, and 15% probability cutoffs were examined. Primary cohort: n = 9643 (mean age, 63.1 ± 8.7 years; 97.2% men; SBP, 15.0%) received first early paracentesis. Testing-set NPVs for SBP were 96.5%, 93.0%, and 91.6% at the 5%, 10%, and 15% probability thresholds, respectively. In Validation cohort #1: n = 2844 (mean age, 63.14 ± 8.37 years; 97.1% male; SBP, 9.7%) with NPVs were 98.8%, 95.3%, and 94.5%. In Validation cohort #2: n = 276 (mean age, 56.08 ± 9.09; 59.6% male; SBP, 7.6%) with NPVs were 100%, 98.9%, and 98.0% The final machine learning model showed the greatest net benefit on decision-curve analyses.

CONCLUSIONS

A machine learning model generated using routinely collected variables excluded SBP with high NPV. Applying this model could ease the need to provide paracentesis in resource-limited settings by excluding those unlikely to have SBP.

Machine learning based on alcohol drinking-gut microbiota-liver axis in predicting the occurrence of early-stage hepatocellular carcinoma

BACKGROUND

Alcohol drinking and gut microbiota are related to hepatocellular carcinoma (HCC), but the specific relationship between them remains unclear.

AIMS

We aimed to establish the alcohol drinking-gut microbiota-liver axis and develop machine learning (ML) models in predicting the occurrence of early-stage HCC.

METHODS

Two hundred sixty-nine patients with early-stage HCC and 278 controls were recruited. Alcohol drinking-gut microbiota-liver axis was established through the mediation/moderation effect analyses. Eight ML algorithms including Classification and Regression Tree (CART), Gradient Boosting Machine (GBM), K-Nearest Neighbor (KNN), Logistic Regression (LR), Neural Network (NN), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost) were applied.

RESULTS

A total of 160 pairs of individuals were included for analyses. The mediation effects of Genus_Catenibacterium (P = 0.024), Genus_Tyzzerella_4 (P < 0.001), and Species_Tyzzerella_4 (P = 0.020) were discovered. The moderation effects of Family_Enterococcaceae (OR = 0.741, 95%CI:0.160-0.760, P = 0.017), Family_Leuconostocaceae (OR = 0.793, 95%CI:0.486-3.593, P = 0.010), Genus_Enterococcus (OR = 0.744, 95%CI:0.161-0.753, P = 0.017), Genus_Erysipelatoclostridium (OR = 0.693, 95%CI:0.062-0.672, P = 0.032), Genus_Lactobacillus (OR = 0.655, 95%CI:0.098-0.749, P = 0.011), Species_Enterococcus_faecium (OR = 0.692, 95%CI:0.061-0.673, P = 0.013), and Species_Lactobacillus (OR = 0.653, 95%CI:0.086-0.765, P = 0.014) were uncovered. The predictive power of eight ML models was satisfactory (AUCs:0.855-0.932). The XGBoost model had the best predictive ability (AUC = 0.932).

CONCLUSIONS

ML models based on the alcohol drinking-gut microbiota-liver axis are valuable in predicting the occurrence of early-stage HCC.

Machine learning for predicting device-associated infection and 30-day survival outcomes after invasive device procedure in intensive care unit patients

This study aimed to preliminarily develop machine learning (ML) models capable of predicting the risk of device-associated infection and 30-day outcomes following invasive device procedures in intensive care unit (ICU) patients. The study utilized data from 8574 ICU patients who underwent invasive procedures, sourced from the Medical Information Mart for Intensive Care (MIMIC)-IV version 2.2 database. Patients were allocated into training and validation datasets in a 7:3 ratio. Seven ML models were employed for predicting device-associated infections, while five models were used for predicting 30-day survival outcomes. Model performance was primarily evaluated using the receiver operating characteristic (ROC) curve for infection prediction and the survival model's concordance index (C-index). Top-performing models progressively reduced the number of variables based on their importance, thereby optimizing practical utility. The inclusion of all variables demonstrated that extreme gradient boosting (XGBoost) and extra survival trees (EST) models yielded superior discriminatory performance. Notably, when restricted to the top 10 variables, both models maintained performance levels comparable to when all variables were included. In the validation cohort, the XGBoost model, with the top 10 variables, achieved an area under the curve (AUC) of 0.810 (95% CI 0.808-0.812), an area under the precision-recall curve (AUPRC) of 0.226 (95% CI 0.222-0.230), and a Brier score (BS) of 0.053 (95% CI 0.053-0.054). The EST model, with the top 10 variables, reported a C-index of 0.756 (95% CI 0.754-0.757), a time-dependent AUC of 0.759 (95% CI 0.763-0.775), and an integrated Brier score (IBS) of 0.087 (95% CI 0.087-0.087). Both models are accessible via a web application. The internally evaluated XGBoost and EST models demonstrated exceptional predictive accuracy for device-associated infection risks and 30-day survival outcomes post-invasive procedures in ICU patients. Further validation is required to confirm the clinical utility of these two models in future studies.

Enhanced Artificial Intelligence Methods for Liver Steatosis Assessment Using Machine Learning and Color Image Processing: Liver Color Project

BACKGROUND

The use of livers with significant steatosis is associated with worse transplantation outcomes. Brain death donor liver acceptance is mostly based on subjective surgeon assessment of liver appearance, since steatotic livers acquire a yellowish tone. The aim of this study was to develop a rapid, robust, accurate, and cost-effective method to assess liver steatosis.

METHODS

From June 1, 2018, to November 30, 2023, photographs and tru-cut needle biopsies were taken from adult brain death donor livers at a single university hospital for the study. All the liver photographs were taken by smartphones then color calibrated, segmented, and divided into patches. Color and texture features were then extracted and used as input, and the machine learning method was applied. This is a collaborative project between Vall d'Hebron University Hospital and Barcelona MedTech, Pompeu Fabra University, and is referred to as LiverColor.

RESULTS

A total of 192 livers (362 photographs and 7240 patches) were included. When setting a macrosteatosis threshold of 30%, the best results were obtained using the random forest classifier, achieving an AUROC = 0.74, with 85% accuracy.

CONCLUSION

Machine learning coupled with liver texture and color analysis of photographs taken with smartphones provides excellent accuracy for determining liver steatosis.

Laboratory variables-based artificial neural network models for predicting fatty liver disease: A retrospective study

Background

The efficacy of artificial neural network (ANN) models employing laboratory variables for predicting fatty liver disease (FLD) remains inadequately established. The study aimed to develop ANN models to precisely predict FLD.

Methods

Of 12,058 participants undergoing the initial FLD screening, 7,990 eligible participants were included. A total of 6,309 participants were divided randomly into the training (4,415 participants, 70%) and validation (1,894 participants, 30%) sets for developing prediction models. The performance of ANNs was additionally tested in the testing set (1,681 participants). The area under the receiver operating characteristic curve (AUROC) was employed to assess the models' performance.

Results

The 18-variable, 11-variable, 3-variable, and 2-variable models each achieved robust FLD prediction performance, with AUROCs over 0.92, 0.91, and 0.89 in the training, validation, and testing, respectively. Although slightly inferior to the other three models in performance (AUROC ranges: 0.89-0.92 vs 0.91-0.95), the 2-variable model showed 80.3% accuracy and 89.7% positive predictive value in the testing. Incorporating age and gender increased the AUROCs of the resulting 20-variable, 13-variable, 5-variable, and 4-variable models each to over 0.93, 0.92, and 0.91 in the training, validation, and testing, respectively.

Conclusions

Implementation of the ANN models could effectively predict FLD, with enhanced predictive performance via the inclusion of age and gender.

Construction and validation of risk prediction models for pulmonary embolism in hospitalized patients based on different machine learning methods

Objective

This study aims to apply different machine learning (ML) methods to construct risk prediction models for pulmonary embolism (PE) in hospitalized patients, and to evaluate and compare the predictive efficacy and clinical benefit of each model.

Methods

We conducted a retrospective study involving 332 participants (172 PE positive cases and 160 PE negative cases) recruited from Guangdong Medical University. Participants were randomly divided into a training group (70%) and a validation group (30%). Baseline data were analyzed using univariate analysis, and potential independent risk factors associated with PE were further identified through univariate and multivariate logistic regression analysis. Six ML models, namely Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Naive Bayes (NB), Support Vector Machine (SVM), and AdaBoost were developed. The predictive efficacy of each model was compared using the receiver operating characteristic (ROC) curve analysis and the area under the curve (AUC). Clinical benefit was assessed using decision curve analysis (DCA).

Results

Logistic regression analysis identified lower extremity deep venous thrombosis, elevated D-dimer, shortened activated partial prothrombin time, and increased red blood cell distribution width as potential independent risk factors for PE. Among the six ML models, the RF model achieved the highest AUC of 0.778. Additionally, DCA consistently indicated that the RF model offered the greatest clinical benefit.

Conclusion

This study developed six ML models, with the RF model exhibiting the highest predictive efficacy and clinical benefit in the identification and prediction of PE occurrence in hospitalized patients.

Machine Learning Approach for Cardiovascular Death Prediction among Nonalcoholic Steatohepatitis (NASH) Liver Transplant Recipients

Cardiovascular disease is the leading cause of mortality among nonalcoholic steatohepatitis (NASH) patients who undergo liver transplants. In the present study, machine learning algorithms were used to identify important risk factors for cardiovascular death and to develop a prediction model. The Standard Transplant Analysis and Research data were gathered from the Organ Procurement and Transplantation Network. After cleaning and preprocessing, the dataset comprised 10,871 patients and 92 features. Recursive feature elimination (RFE) and select from model (SFM) were applied to select relevant features from the dataset and avoid overfitting. Multiple machine learning algorithms, including logistic regression, random forest, decision tree, and XGBoost, were used with RFE and SFM. Additionally, prediction models were developed using a support vector machine, Gaussian naïve Bayes, K-nearest neighbors, random forest, and XGBoost algorithms. Finally, SHapley Additive exPlanations (SHAP) were used to increase interpretability. The findings showed that the best feature selection method was RFE with a random forest estimator, and the most critical features were recipient and donor blood type, body mass index, recipient and donor state of residence, serum creatinine, and year of transplantation. Furthermore, among all the outcomes, the XGBoost model had the highest performance, with an accuracy value of 0.6909 and an area under the curve value of 0.86. The findings also revealed a predictive relationship between features and cardiovascular death after liver transplant among NASH patients. These insights may assist clinical decision-makers in devising strategies to prevent cardiovascular complications in post-liver transplant NASH patients.

Deep Learning-Based Prediction Modeling of Major Adverse Cardiovascular Events After Liver Transplantation

Objective

To validate deep learning models' ability to predict post-transplantation major adverse cardiovascular events (MACE) in patients undergoing liver transplantation (LT).

Patients and Methods

We used data from Optum's de-identified Clinformatics Data Mart Database to identify liver transplant recipients between January 2007 and March 2020. To predict post-transplantation MACE risk, we considered patients' demographics characteristics, diagnoses, medications, and procedural data recorded back to 3 years before the LT procedure date (index date). MACE is predicted using the bidirectional gated recurrent units (BiGRU) deep learning model in different prediction interval lengths up to 5 years after the index date. In total, 18,304 liver transplant recipients (mean age, 57.4 years [SD, 12.76]; 7158 [39.1%] women) were used to develop and test the deep learning model's performance against other baseline machine learning models. Models were optimized using 5-fold cross-validation on 80% of the cohort, and model performance was evaluated on the remaining 20% using the area under the receiver operating characteristic curve (AUC-ROC) and the area under the precision-recall curve (AUC-PR).

Results

Using different prediction intervals after the index date, the top-performing model was the deep learning model, BiGRU, and achieved an AUC-ROC of 0.841 (95% CI, 0.822-0.862) and AUC-PR of 0.578 (95% CI, 0.537-0.621) for a 30-day prediction interval after LT.

Conclusion

Using longitudinal claims data, deep learning models can efficiently predict MACE after LT, assisting clinicians in identifying high-risk candidates for further risk stratification or other management strategies to improve transplant outcomes based on important features identified by the model.

Computed tomography-based prediction of pancreatitis following biliary metal stent placement with the convolutional neural network

Background and study aims Pancreatitis is a potentially lethal adverse event of endoscopic transpapillary placement of a self-expandable metal stent (SEMS) for malignant biliary obstruction (MBO). Deep learning-based image recognition has not been investigated in predicting pancreatitis in this setting. Patients and methods We included 70 patients who underwent endoscopic placement of a SEMS for nonresectable distal MBO. We constructed a convolutional neural network (CNN) model for pancreatitis prediction using a series of pre-procedure computed tomography images covering the whole pancreas (≥ 120,960 augmented images in total). We examined the additional effects of the CNN-based probabilities on the following machine learning models based on clinical parameters: logistic regression, support vector machine with a linear or RBF kernel, random forest classifier, and gradient boosting classifier. Model performance was assessed based on the area under the curve (AUC) in the receiver operating characteristic analysis, positive predictive value (PPV), accuracy, and specificity. Results The CNN model was associated with moderate levels of performance metrics: AUC, 0.67; PPV, 0.45; accuracy, 0.66; and specificity, 0.63. When added to the machine learning models, the CNN-based probabilities increased the performance metrics. The logistic regression model with the CNN-based probabilities had an AUC of 0.74, PPV of 0.85, accuracy of 0.83, and specificity of 0.96, compared with 0.72, 0.78, 0.77, and 0.96, respectively, without the probabilities. Conclusions The CNN-based model may increase predictability for pancreatitis following endoscopic placement of a biliary SEMS. Our findings support the potential of deep learning technology to improve prognostic models in pancreatobiliary therapeutic endoscopy.

Enhancing liver fibrosis diagnosis and treatment assessment: a novel biomechanical markers-based machine learning approach

Accurate diagnosis and treatment assessment of liver fibrosis face significant challenges, including inherent limitations in current techniques like sampling errors and inter-observer variability. Addressing this, our study introduces a novel machine learning (ML) framework, which integrates light gradient boosting machine and multivariate imputation by chained equations to enhance liver status assessment using biomechanical markers. Building upon our previously established multiscale mechanical characteristics in fibrotic and treated livers, this framework employs Gaussian Bayesian optimization for post-imputation, significantly improving classification performance. Our findings indicate a marked increase in the precision of liver fibrosis diagnosis and provide a novel, quantitative approach for assessing fibrosis treatment. This innovative combination of multiscale biomechanical markers with advanced ML algorithms represents a transformative step in liver disease diagnostics and treatment evaluation, with potential implications for other areas in medical diagnostics.

Liver Diseases: Science, Fiction and the Foreseeable Future

This Editorial precedes the Special Issue entitled "Novel Challenges and Therapeutic Options for Liver Diseases". Following a historical outline of the roots of hepatology, we provide a brief insight into our colleagues' contributions in this issue on the current developments in this discipline related to the prevention of liver diseases, the metabolic dysfunction-associated steatotic liver disease (or non-alcoholic fatty liver disease, respectively), liver cirrhosis, chronic viral hepatitides, acute-on-chronic liver failure, liver transplantation, the liver-microbiome axis and microbiome transplantation, and telemedicine. We further add some topics not covered by the contributions herein that will likely impact future hepatology. Clinically, these comprise the predictive potential of organokine crosstalk and treatment options for liver fibrosis. With regard to promising developments in basic research, some current findings on the genetic basis of metabolism-associated chronic liver diseases, chronobiology, metabolic zonation of the liver, aspects of the aging liver against the background of demography, and liver regeneration will be presented. We expect machine learning to thrive as an overarching topic throughout hepatology. The largest study to date on the early detection of liver damage-which has been kicked off on 1 March 2024-is highlighted, too.

LightGBM outperforms other machine learning techniques in predicting graft failure after liver transplantation: Creation of a predictive model through large-scale analysis

BACKGROUND

The incidence of graft failure following liver transplantation (LTx) is consistent. While traditional risk scores for LTx have limited accuracy, the potential of machine learning (ML) in this area remains uncertain, despite its promise in other transplant domains. This study aims to determine ML's predictive limitations in LTx by replicating methods used in previous heart transplant research.

METHODS

This study utilized the UNOS STAR database, selecting 64,384 adult patients who underwent LTx between 2010 and 2020. Gradient boosting models (XGBoost and LightGBM) were used to predict 14, 30, and 90-day graft failure compared to conventional logistic regression model. Models were evaluated using both shuffled and rolling cross-validation (CV) methodologies. Model performance was assessed using the AUC across validation iterations.

RESULTS

In a study comparing predictive models for 14-day, 30-day and 90-day graft survival, LightGBM consistently outperformed other models, achieving the highest AUC of.740,.722, and.700 in shuffled CV methods. However, in rolling CV the accuracy of the model declined across every ML algorithm. The analysis revealed influential factors for graft survival prediction across all models, including total bilirubin, medical condition, recipient age, and donor AST, among others. Several features like donor age and recipient diabetes history were important in two out of three models.

CONCLUSIONS

LightGBM enhances short-term graft survival predictions post-LTx. However, due to changing medical practices and selection criteria, continuous model evaluation is essential. Future studies should focus on temporal variations, clinical implications, and ensure model transparency for broader medical utility.

Diagnosis of Gallbladder Disease Using Artificial Intelligence: A Comparative Study

Gallbladder (GB) disease is a common pathology that needs correct and early diagnosis for the optimum medical treatment. Early diagnosis is crucial as any delay or misdiagnosis can worsen the patient situation. Incorrect diagnosis could also lead to an escalation in patient symptoms and poorer clinical outcomes. The use of Artificial Intelligence (AI) techniques, ranging from Machine Learning (ML) to Deep Learning (DL) to predict disease progression, identify abnormalities, and estimate mortality rates associated with GB disorders has increased over the past decade. To this end, this paper provides a comprehensive overview of the AI approaches used in the diagnosis of GB illnesses. This review compiles and compares relevant papers from the last decade to show how AI might enhance diagnostic precision, speed, and efficiency. Therefore, this survey gives researchers the opportunity to find out both the diagnosis of GB diseases and AI techniques in one place. The maximum accuracy rate by ML was when using SVM with 96.67%, whilst the maximum accuracy rate by DL was by utilising a unique structure of VGG, GoogleNet, ResNet, AlexNet and Inception with 98.77%. This could provide a clear path for further investigations and algorithm’s development to boost diagnostic results to improve the patient’s condition and choose the appropriate treatment.

Machine Learning Radiomics Liver Function Model for Prognostic Prediction After Radical Resection of Advanced Gastric Cancer: A Retrospective Study

PURPOSE

We aimed to establish a machine learning radiomics liver function model to explore how liver function affects the prognosis of patients with gastric cancer (GC).

METHODS

Patients with advanced GC were retrospectively enrolled in this study. Eight machine learning radiomic models were constructed by extracting radiomic features from portal-vein-phase contrast-enhanced computed tomography (CE-CT) images. Clinicopathological features were determined using univariate and multifactorial Cox regression analyses. These features were used to construct a GC survival nomogram.

RESULTS

A total of 510 patients with GC were split into training and test cohorts in an 8:2 ratio. Kaplan-Meier analysis showed that patients with type I liver function had a better prognosis. Fifteen significant features were retained to establish the machine learning model. LightBGM showed the best predictive performance in the training (area under the receiver operating characteristic curve [AUC] 0.978) and test cohorts (AUC 0.714). Multivariate analysis revealed that gender, age, liver function, Nutritional Risk Screening 2002 (NRS-2002) score, tumor-lymph node-metastasis stage, tumor size, and tumor differentiation were independent risk factors for GC prognosis. The survival nomogram based on machine learning radiomics, instead of liver biochemical indicators, still had high accuracy (C-index of 0.771 vs. 0.773).

CONCLUSION

The machine learning radiomics liver function model has high diagnostic value in predicting the influence of liver function on prognosis in patients with GC.

The role of point-of-care ultrasound in the assessment of schistosomiasis-induced liver fibrosis: A systematic scoping review

BACKGROUND

Abdominal ultrasound imaging is an important method for hepatic schistosomiasis diagnosis and staging. Several ultrasound staging systems have been proposed, each attempting to standardise schistosomal periportal fibrosis (PPF) diagnosis. This review aims to establish the role of ultrasound in the diagnosis and staging of schistosomal PPF, and to map the evolution of ultrasound staging systems over time, focusing on internal validation and external reproducibility.

METHODS

A systematic search was undertaken on 21st December 2022 considering the following databases: PubMed/MEDLINE (1946-present), Embase (1974-present), Global Health (1973-present), Global Index Medicus (1901-present), and Web of Science Core Collection-Science Citation Index Expanded (1900-present) and the Cochrane Central Register of Controlled Trials (1996-present). Case reports, systematic reviews and meta-analyses, and studies exclusively using transient or shear-wave elastography were excluded. Variables extracted included study design, study population, schistosomal PPF characteristics, and diagnostic methods. The PRISMA-ScR (2018) guidelines were followed to inform the structure of the scoping analysis.

RESULTS

The initial search yielded 573 unique articles, of which 168 were removed after screening titles and abstracts, 43 were not retrieved due to full texts not being available online or through inter-library loans, and 170 were excluded during full text review. There were 192 remaining studies eligible for extraction. Of the extracted studies, 61.8% (76/123) of studies that reported study year were conducted after the year 2000. Over half of all extracted studies (59.4%; 114/192) were conducted in Brazil (26.0%; 50/192), China (18.8%; 36/192) or Egypt (14.6%; 28/192). For the species of schistosome considered, 77.6% (149/192) of studies considered S. mansoni and 21.4% (41/192) of studies considered S. japonicum. The ultrasound staging systems used took on three forms: measurement-based, feature-based and image pattern-based. The Niamey protocol, a measurement and image pattern-based system, was the most used among the staging systems (32.8%; 63/192), despite being the most recently proposed in 1996. The second most used was the Cairo protocol (20.8%; 40/192). Of the studies using the Niamey protocol, 77.8% (49/63) only used the image patterns element. Where ultrasound technology was specified, studies after 2000 were more likely to use convex transducers (43.4%; 33/76) than studies conducted before 2000 (32.7%; 16/49). Reporting on ultrasound-based hepatic diagnoses and their association with clinical severity was poor. Just over half of studies (56.2%; 108/192) reported the personnel acquiring the ultrasound images. A small number (9.4%; 18/192) of studies detailed their methods of image quality assurance, and 13.0% (25/192) referenced, discussed or quantified the inter- or intra-observer variation of the staging system that was used.

CONCLUSIONS

The exclusive use of the image patterns in many studies despite lack of specific acquisition guidance, the increasing number of studies over time that conduct ultrasound staging of schistosomal PPF, and the advances in ultrasound technology used since 2000 all indicate a need to consider an update to the Niamey protocol. The protocol update should simplify and prioritise what is to be assessed, advise on who is to conduct the ultrasound examination, and procedures for improved standardisation and external reproducibility.

Predicting the Climate Impact of Healthcare Facilities Using Gradient Boosting Machines

Health care accounts for 9-10% of greenhouse gas (GHG) emissions in the United States. Strategies for monitoring these emissions at the hospital level are needed to decarbonize the sector. However, data collection to estimate emissions is challenging, especially for smaller hospitals. We explored the potential of gradient boosting machines (GBM) to impute missing data on resource consumption in the 2020 survey of a consortium of 283 hospitals participating in Practice Greenhealth. GBM imputed missing values for selected variables in order to predict electricity use and beef consumption (R2=0.82) and anesthetic gas desflurane use (R2=0.51), using administrative data readily available for most hospitals. After imputing missing consumption data, estimated GHG emissions associated with these three examples totaled over 3 million metric tons of CO2 equivalent emissions (MTCO2e). Specifically, electricity consumption had the largest total carbon footprint (2.4 MTCO2e), followed by beef (0.6 million MTCO2e) and desflurane consumption (0.03 million MTCO2e) across the 283 hospitals. The approach should be applicable to other sources of hospital GHGs in order to estimate total emissions of individual hospitals and to refine survey questions to help develop better intervention strategies.

Computed tomography-based machine learning for donor lung screening before transplantation

BACKGROUND

Assessment and selection of donor lungs remain largely subjective and experience based. Criteria to accept or decline lungs are poorly standardized and are not compliant with the current donor pool. Using ex vivo computed tomography (CT) images, we investigated the use of a CT-based machine learning algorithm for screening donor lungs before transplantation.

METHODS

Clinical measures and ex situ CT scans were collected from 100 cases as part of a prospective clinical trial. Following procurement, donor lungs were inflated, placed on ice according to routine clinical practice, and imaged using a clinical CT scanner before transplantation while stored in the icebox. We trained and tested a supervised machine learning method called dictionary learning, which uses CT scans and learns specific image patterns and features pertaining to each class for a classification task. The results were evaluated with donor and recipient clinical measures.

RESULTS

Of the 100 lung pairs donated, 70 were considered acceptable for transplantation (based on standard clinical assessment) before CT screening and were consequently implanted. The remaining 30 pairs were screened but not transplanted. Our machine learning algorithm was able to detect pulmonary abnormalities on the CT scans. Among the patients who received donor lungs, our algorithm identified recipients who had extended stays in the intensive care unit and were at 19 times higher risk of developing chronic lung allograft dysfunction within 2 years posttransplant.

CONCLUSIONS

We have created a strategy to ex vivo screen donor lungs using a CT-based machine learning algorithm. As the use of suboptimal donor lungs rises, it is important to have in place objective techniques that will assist physicians in accurately screening donor lungs to identify recipients most at risk of posttransplant complications.

Development of a machine learning-based model to predict hepatic inflammation in chronic hepatitis B patients with concurrent hepatic steatosis: a cohort study

BACKGROUND

With increasingly prevalent coexistence of chronic hepatitis B (CHB) and hepatic steatosis (HS), simple, non-invasive diagnostic methods to accurately assess the severity of hepatic inflammation are needed. We aimed to build a machine learning (ML) based model to detect hepatic inflammation in patients with CHB and concurrent HS.

METHODS

We conducted a multicenter, retrospective cohort study in China. Treatment-naive CHB patients with biopsy-proven HS between April 2004 and September 2022 were included. The optimal features for model development were selected by SHapley Additive explanations, and an ML algorithm with the best accuracy to diagnose moderate to severe hepatic inflammation (Scheuer's system ≥ G3) was determined and assessed by decision curve analysis (DCA) and calibration curve. This study is registered with ClinicalTrials.gov (NCT05766449).

FINDINGS

From a pool of 1,787 treatment-naive patients with CHB and HS across eleven hospitals, 689 patients from nine of these hospitals were chosen for the development of the diagnostic model. The remaining two hospitals contributed to two independent external validation cohorts, comprising 509 patients in validation cohort 1 and 589 in validation cohort 2. Eleven features regarding inflammation, hepatic and metabolic functions were identified. The gradient boosting classifier (GBC) model showed the best performance in predicting moderate to severe hepatic inflammation, with an area under the receiver operating characteristic curve (AUROC) of 0.86 (95% CI 0.83-0.88) in the training cohort, and 0.89 (95% CI 0.86-0.92), 0.76 (95% CI 0.73-0.80) in the first and second external validation cohorts, respectively. A publicly accessible web tool was generated for the model.

INTERPRETATION

Using simple parameters, the GBC model predicted hepatic inflammation in CHB patients with concurrent HS. It holds promise for guiding clinical management and improving patient outcomes.

FUNDING

This research was supported by the National Natural Science Foundation of China (No. 82170609, 81970545), Natural Science Foundation of Shandong Province (Major Project) (No. ZR2020KH006), Natural Science Foundation of Jiangsu Province (No.BK20231118), Tianjin Key Medical Discipline (Specialty), Construction Project, TJYXZDXK-059B, Tianjin Health Science and Technology Project key discipline special, TJWJ2022XK034, and Research project of Chinese traditional medicine and Chinese traditional medicine combined with Western medicine of Tianjin municipal health and Family Planning Commission (2021022).

The Integration of Artificial Intelligence into Clinical Practice

The purpose of this literature review is to provide a fundamental synopsis of current research pertaining to artificial intelligence (AI) within the domain of clinical practice. Artificial intelligence has revolutionized the field of medicine and healthcare by providing innovative solutions to complex problems. One of the most important benefits of AI in clinical practice is its ability to investigate extensive volumes of data with efficiency and precision. This has led to the development of various applications that have improved patient outcomes and reduced the workload of healthcare professionals. AI can support doctors in making more accurate diagnoses and developing personalized treatment plans. Successful examples of AI applications are outlined for a series of medical specialties like cardiology, surgery, gastroenterology, pneumology, nephrology, urology, dermatology, orthopedics, neurology, gynecology, ophthalmology, pediatrics, hematology, and critically ill patients, as well as diagnostic methods. Special reference is made to legal and ethical considerations like accuracy, informed consent, privacy issues, data security, regulatory framework, product liability, explainability, and transparency. Finally, this review closes by critically appraising AI use in clinical practice and its future perspectives. However, it is also important to approach its development and implementation cautiously to ensure ethical considerations are met.

Image-based AI diagnostic performance for fatty liver: a systematic review and meta-analysis

BACKGROUND

The gold standard to diagnose fatty liver is pathology. Recently, image-based artificial intelligence (AI) has been found to have high diagnostic performance. We systematically reviewed studies of image-based AI in the diagnosis of fatty liver.

METHODS

We searched the Cochrane Library, Pubmed, Embase and assessed the quality of included studies by QUADAS-AI. The pooled sensitivity, specificity, negative likelihood ratio (NLR), positive likelihood ratio (PLR), and diagnostic odds ratio (DOR) were calculated using a random effects model. Summary receiver operating characteristic curves (SROC) were generated to identify the diagnostic accuracy of AI models.

RESULTS

15 studies were selected in our meta-analysis. Pooled sensitivity and specificity were 92% (95% CI: 90-93%) and 94% (95% CI: 93-96%), PLR and NLR were 12.67 (95% CI: 7.65-20.98) and 0.09 (95% CI: 0.06-0.13), DOR was 182.36 (95% CI: 94.85-350.61). After subgroup analysis by AI algorithm (conventional machine learning/deep learning), region, reference (US, MRI or pathology), imaging techniques (MRI or US) and transfer learning, the model also demonstrated acceptable diagnostic efficacy.

CONCLUSION

AI has satisfactory performance in the diagnosis of fatty liver by medical imaging. The integration of AI into imaging devices may produce effective diagnostic tools, but more high-quality studies are needed for further evaluation.

Development and validation of a practical machine learning model to predict sepsis after liver transplantation

BACKGROUND

Postoperative sepsis is one of the main causes of mortality after liver transplantation (LT). Our study aimed to develop and validate a predictive model for postoperative sepsis within 7 d in LT recipients using machine learning (ML) technology.

METHODS

Data of 786 patients received LT from January 2015 to January 2020 was retrospectively extracted from the big data platform of Third Affiliated Hospital of Sun Yat-sen University. Seven ML models were developed to predict postoperative sepsis. The area under the receiver-operating curve (AUC), sensitivity, specificity, accuracy, and f1-score were evaluated as the model performances. The model with the best performance was validated in an independent dataset involving 118 adult LT cases from February 2020 to April 2021. The postoperative sepsis-associated outcomes were also explored in the study.

RESULTS

After excluding 109 patients according to the exclusion criteria, 677 patients underwent LT were finally included in the analysis. Among them, 216 (31.9%) were diagnosed with sepsis after LT, which were related to more perioperative complications, increased postoperative hospital stay and mortality after LT (all p < .05). Our results revealed that a larger volume of red blood cell infusion, ascitic removal, blood loss and gastric drainage, less volume of crystalloid infusion and urine, longer anesthesia time, higher level of preoperative TBIL were the top 8 important variables contributing to the prediction of post-LT sepsis. The Random Forest Classifier (RF) model showed the best overall performance to predict sepsis after LT among the seven ML models developed in the study, with an AUC of 0.731, an accuracy of 71.6%, the sensitivity of 62.1%, and specificity of 76.1% in the internal validation set, and a comparable AUC of 0.755 in the external validation set.

CONCLUSIONS

Our study enrolled eight pre- and intra-operative variables to develop an RF-based predictive model of post-LT sepsis to assist clinical decision-making procedure.

Prognostic models in end stage liver disease

Cirrhosis is a major cause of death worldwide, and is associated with significant health care costs. Even if milestones have been recently reached in understanding and managing end-stage liver disease (ESLD), the disease course remains somewhat difficult to prognosticate. These difficulties have already been acknowledged already in the past, when scores instead of single parameters have been proposed as valuable tools for short-term prognosis. These standard scores, like Child Turcotte Pugh (CTP) and model for end-stage liver disease (MELD) score, relying on biochemical and clinical parameters, are still widely used in clinical practice to predict short- and medium-term prognosis. The MELD score, which remains an accurate, easy-to-use, objective predictive score, has received significant modifications over time, in order to improve its performance especially in the liver transplant (LT) setting, where it is widely used as prioritization tool. Although many attempts to improve prognostic accuracy have failed because of lack of replicability or poor benefit with the comparator (often the MELD score or its variants), few scores have been recently proposed and validated especially for subgroups of patients with ESLD, as those with acute-on-chronic liver failure. Artificial intelligence will probably help hepatologists in the near future to fill the current gaps in predicting disease course and long-term prognosis of such patients.

Predicting post-liver transplant outcomes in patients with acute-on-chronic liver failure using Expert-Augmented Machine Learning

Liver transplantation (LT) is a treatment for acute-on-chronic liver failure (ACLF), but high post-LT mortality has been reported. Existing post-LT models in ACLF have been limited. We developed an Expert-Augmented Machine Learning (EAML) model to predict post-LT outcomes. We identified ACLF patients who underwent LT in the University of California Health Data Warehouse. We applied the RuleFit machine learning (ML) algorithm to extract rules from decision trees and create intermediate models. We asked human experts to rate the rules generated by RuleFit and incorporated these ratings to generate final EAML models. We identified 1384 ACLF patients. For death at 1 year, areas under the receiver-operating characteristic curve were 0.707 (confidence interval [CI] 0.625-0.793) for EAML and 0.719 (CI 0.640-0.800) for RuleFit. For death at 90 days, areas under the receiver-operating characteristic curve were 0.678 (CI 0.581-0.776) for EAML and 0.707 (CI 0.615-0.800) for RuleFit. In pairwise comparisons, both EAML and RuleFit models outperformed cross-sectional models. Significant discrepancies between experts and ML occurred in rankings of biomarkers used in clinical practice. EAML may serve as a method for ML-guided hypothesis generation in further ACLF research.

Artificial intelligence-based clinical decision support for liver transplant evaluation and considerations about fairness: A qualitative study

BACKGROUND

The use of large-scale data and artificial intelligence (AI) to support complex transplantation decisions is in its infancy. Transplant candidate decision-making, which relies heavily on subjective assessment (ie, high variability), provides a ripe opportunity for AI-based clinical decision support (CDS). However, AI-CDS for transplant applications must consider important concerns regarding fairness (ie, health equity). The objective of this study was to use human-centered design methods to elicit providers' perceptions of AI-CDS for liver transplant listing decisions.

METHODS

In this multicenter qualitative study conducted from December 2020 to July 2021, we performed semistructured interviews with 53 multidisciplinary liver transplant providers from 2 transplant centers. We used inductive coding and constant comparison analysis of interview data.

RESULTS

Analysis yielded 6 themes important for the design of fair AI-CDS for liver transplant listing decisions: (1) transparency in the creators behind the AI-CDS and their motivations; (2) understanding how the AI-CDS uses data to support recommendations (ie, interpretability); (3) acknowledgment that AI-CDS could mitigate emotions and biases; (4) AI-CDS as a member of the transplant team, not a replacement; (5) identifying patient resource needs; and (6) including the patient's role in the AI-CDS.

CONCLUSIONS

Overall, providers interviewed were cautiously optimistic about the potential for AI-CDS to improve clinical and equitable outcomes for patients. These findings can guide multidisciplinary developers in the design and implementation of AI-CDS that deliberately considers health equity.

Development of a liver graft assessment expert machine-learning system: when the artificial intelligence helps liver transplant surgeons

Background

The complex process of liver graft assessment is one point for improvement in liver transplantation. The main objective of this study is to develop a tool that supports the surgeon who is responsible for liver donation in the decision-making process whether to accept a graft or not using the initial variables available to it.

Material and method

Liver graft samples candidate for liver transplantation after donor brain death were studied. All of them were evaluated "in situ" for transplantation, and those discarded after the "in situ" evaluation were considered as no transplantable liver grafts, while those grafts transplanted after "in situ" evaluation were considered as transplantable liver grafts. First, a single-center, retrospective and cohort study identifying the risk factors associated with the no transplantable group was performed. Then, a prediction model decision support system based on machine learning, and using a tree ensemble boosting classifier that is capable of helping to decide whether to accept or decline a donor liver graft, was developed.

Results

A total of 350 liver grafts that were evaluated for liver transplantation were studied. Steatosis was the most frequent reason for classifying grafts as no transplantable, and the main risk factors identified in the univariant study were age, dyslipidemia, personal medical history, personal surgical history, bilirubinemia, and the result of previous liver ultrasound (p < 0.05). When studying the developed model, we observe that the best performance reordering in terms of accuracy corresponds to 76.29% with an area under the curve of 0.79. Furthermore, the model provides a classification together with a confidence index of reliability, for most cases in our data, with the probability of success in the prediction being above 0.85.

Conclusion

The tool presented in this study obtains a high accuracy in predicting whether a liver graft will be transplanted or deemed non-transplantable based on the initial variables assigned to it. The inherent capacity for improvement in the system causes the rate of correct predictions to increase as new data are entered. Therefore, we believe it is a tool that can help optimize the graft pool for liver transplantation.

Prediction of New-Onset Diabetes Mellitus within 12 Months after Liver Transplantation-A Machine Learning Approach

BACKGROUND

Liver transplantation (LT) is a routine therapeutic approach for patients with acute liver failure, end-stage liver disease and/or early-stage liver cancer. While 5-year survival rates have increased to over 80%, long-term outcomes are critically influenced by extrahepatic sequelae of LT and immunosuppressive therapy, including diabetes mellitus (DM). In this study, we used machine learning (ML) to predict the probability of new-onset DM following LT.

METHODS

A cohort of 216 LT patients was identified from the Disease Analyzer (DA) database (IQVIA) between 2005 and 2020. Three ML models comprising random forest (RF), logistic regression (LR), and eXtreme Gradient Boosting (XGBoost) were tested as predictors of new-onset DM within 12 months after LT.

RESULTS

18 out of 216 LT patients (8.3%) were diagnosed with DM within 12 months after the index date. The performance of the RF model in predicting the development of DM was the highest (accuracy = 79.5%, AUC 77.5%). It correctly identified 75.0% of the DM patients and 80.0% of the non-DM patients in the testing dataset. In terms of predictive variables, patients' age, frequency and time of proton pump inhibitor prescription as well as prescriptions of analgesics, immunosuppressants, vitamin D, and two antibiotic drugs (broad spectrum penicillins, fluocinolone) were identified.

CONCLUSIONS

Pending external validation, our data suggest that ML models can be used to predict the occurrence of new-onset DM following LT. Such tools could help to identify LT patients at risk of unfavorable outcomes and to implement respective clinical strategies of prevention.

Machine learning radiomics to predict the early recurrence of intrahepatic cholangiocarcinoma after curative resection: A multicentre cohort study

PURPOSE

Postoperative early recurrence (ER) leads to a poor prognosis for intrahepatic cholangiocarcinoma (ICC). We aimed to develop machine learning (ML) radiomics models to predict ER in ICC after curative resection.

METHODS

Patients with ICC undergoing curative surgery from three institutions were retrospectively recruited and assigned to training and external validation cohorts. Preoperative arterial and venous phase contrast-enhanced computed tomography (CECT) images were acquired and segmented. Radiomics features were extracted and ranked through their importance. Univariate and multivariate logistic regression analysis was used to identify clinical characteristics. Various ML algorithms were used to construct radiomics-based models, and the predictive performance was evaluated by receiver operating characteristic curves, calibration curves, and decision curve analysis.

RESULTS

127 patients were included for analysis: 90 patients in the training set and 37 patients in the validation set. Ninety-two patients (72.4%) experienced recurrence, including 71 patients exhibiting ER. Male sex, microvascular invasion, TNM stage, and serum CA19-9 were identified as independent risk factors for ER, with the corresponding clinical model having a poor predictive performance (AUC of 0.685). Fifty-seven differential radiomics features were identified, and the 10 most important features were utilized for modelling. Seven ML radiomics models were developed with a mean AUC of 0.87 ± 0.02, higher than the clinical model. Furthermore, the clinical-radiomics models showed similar predictive performance to the radiomics models (AUC of 0.87 ± 0.03).

CONCLUSION

ML radiomics models based on CECT are valuable in predicting ER in ICC.

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.

Machine Learning Algorithm Improves the Prediction of Transplant Hepatic Artery Stenosis or Occlusion: A Single-Center Study

ABSTRACT

The aim of this study was to determine if machine learning can improve the specificity of detecting transplant hepatic artery pathology over conventional quantitative measures while maintaining a high sensitivity.This study presents a retrospective review of 129 patients with transplanted hepatic arteries. We illustrate how beyond common clinical metrics such as stenosis and resistive index, a more comprehensive set of waveform data (including flow half-lives and Fourier transformed waveforms) can be integrated into machine learning models to obtain more accurate screening of stenosis and occlusion. We present a novel framework of Extremely Randomized Trees and Shapley values, we allow for explainability at the individual level.The proposed framework identified cases of clinically significant stenosis and occlusion in hepatic arteries with a state-of-the-art specificity of 65%, while maintaining sensitivity at the current standard of 94%. Moreover, through 3 case studies of correct and mispredictions, we demonstrate examples of how specific features can be elucidated to aid in interpreting driving factors in a prediction.This work demonstrated that by utilizing a more complete set of waveform data and machine learning methodologies, it is possible to reduce the rate of false-positive results in using ultrasounds to screen for transplant hepatic artery pathology compared with conventional quantitative measures. An advantage of such techniques is explainability measures at the patient level, which allow for increased radiologists' confidence in the predictions.

Novel online calculator to predict reduced risk of early recurrence from adjuvant transarterial chemoembolisation for patients with hepatocellular carcinoma

BACKGROUND

The role of adjuvant transarterial chemoembolisation (TACE) to reduce postoperative recurrence varies widely among patients undergoing hepatectomy with curative intent for hepatocellular carcinoma (HCC). Personalised predictive tool to select which patients may benefit from adjuvant TACE is lacking. This study aimed to develop and validate an online calculator for estimating the reduced risk of early recurrence from adjuvant TACE for patients with HCC.

METHODS

From a multi-institutional database, 2590 eligible patients undergoing curative-intent hepatectomy for HCC were enrolled, and randomly assigned to the training and validation cohorts. Independent predictors of early recurrence within 1 year of surgery were identified in the training cohort, and subsequently used to construct a model and corresponding prediction calculator. The predictive performance of the model was validated using concordance indexes (C-indexes) and calibration curves, and compared with conventional HCC staging systems. The reduced risk of early recurrence when receiving adjuvant TACE was used to estimate the expected benefit from adjuvant TACE.

RESULTS

The prediction model was developed by integrating eight factors that were independently associated with risk of early recurrence: alpha-fetoprotein level, maximum tumour size, tumour number, macrovascular and microvascular invasion, satellite nodules, resection margin and adjuvant TACE. The model demonstrated good calibration and discrimination in the training and validation cohorts (C-indexes: 0.799 and 0.778, respectively), and performed better among the whole cohort than four conventional HCC staging systems (C-indexes: 0.797 vs 0.562-0.673, all p<0.001). An online calculator was built to estimate the reduced risk of early recurrence from adjuvant TACE for patients with resected HCC.

CONCLUSIONS

The proposed calculator can be adopted to assist decision-making for clinicians and patients to determine which patients with resected HCC can significantly benefit from adjuvant TACE.

Automated screening of potential organ donors using a temporal machine learning model

Organ donation is not meeting demand, and yet 30-60% of potential donors are potentially not identified. Current systems rely on manual identification and referral to an Organ Donation Organization (ODO). We hypothesized that developing an automated screening system based on machine learning could reduce the proportion of missed potentially eligible organ donors. Using routine clinical data and laboratory time-series, we retrospectively developed and tested a neural network model to automatically identify potential organ donors. We first trained a convolutive autoencoder that learned from the longitudinal changes of over 100 types of laboratory results. We then added a deep neural network classifier. This model was compared to a simpler logistic regression model. We observed an AUROC of 0.966 (CI 0.949-0.981) for the neural network and 0.940 (0.908-0.969) for the logistic regression model. At a prespecified cutoff, sensitivity and specificity were similar between both models at 84% and 93%. Accuracy of the neural network model was robust across donor subgroups and remained stable in a prospective simulation, while the logistic regression model performance declined when applied to rarer subgroups and in the prospective simulation. Our findings support using machine learning models to help with the identification of potential organ donors using routinely collected clinical and laboratory data.

Complement inhibition alleviates donor brain death-induced liver injury and posttransplant cascade injury by regulating phosphoinositide 3-kinase signaling

Brain death (BD) donors are the primary source of donor organs for liver transplantation. However, the effects of BD on donor livers and outcomes after liver transplantation remain unclear. Here, we explored the role of complement and the therapeutic effect of complement inhibition in BD-induced liver injury and posttransplantation injury in a mouse BD and liver transplantation model. For complement inhibition, we used complement receptor 2 (CR2)-Crry, a murine inhibitor of C3 activation that specifically targets sites of complement activation. In the mouse model, BD resulted in complement activation and liver injury in donor livers and a cascade liver injury posttransplantation, mediated in part through the C3a-C3aR (C3a receptor) signaling pathway, which was ameliorated by treatment with CR2-Crry. Treatment of BD donors with CR2-Crry improved graft survival, which was further improved when recipients received an additional dose of CR2-Crry posttransplantation. Mechanistically, we determined that complement inhibition alleviated BD-induced donor liver injury and posttransplant cascade injury by regulating phosphoinositide 3-kinase (PI3K) signaling pathways. Together, BD induced donor liver injury and cascade injury post-transplantation, which was mediated by complement activation products acting on PI3K signaling pathways. Our study provides an experimental basis for developing strategies to improve the survival of BD donor grafts in liver transplantation.

CT-based Machine Learning for Donor Lung Screening Prior to Transplantation

Background

Assessment and selection of donor lungs remains largely subjective and experience based. Criteria to accept or decline lungs are poorly standardized and are not compliant with the current donor pool. Using ex vivo CT images, we investigated the use of a CT-based machine learning algorithm for screening donor lungs prior to transplantation.

Methods

Clinical measures and ex-situ CT scans were collected from 100 cases as part of a prospective clinical trial. Following procurement, donor lungs were inflated, placed on ice according to routine clinical practice, and imaged using a clinical CT scanner prior to transplantation while stored in the icebox. We trained and tested a supervised machine learning method called dictionary learning , which uses CT scans and learns specific image patterns and features pertaining to each class for a classification task. The results were evaluated with donor and recipient clinical measures.

Results

Of the 100 lung pairs donated, 70 were considered acceptable for transplantation (based on standard clinical assessment) prior to CT screening and were consequently implanted. The remaining 30 pairs were screened but not transplanted. Our machine learning algorithm was able to detect pulmonary abnormalities on the CT scans. Among the patients who received donor lungs, our algorithm identified recipients who had extended stays in the ICU and were at 19 times higher risk of developing CLAD within 2 years post-transplant.

Conclusions

We have created a strategy to ex vivo screen donor lungs using a CT-based machine learning algorithm. As the use of suboptimal donor lungs rises, it is important to have in place objective techniques that will assist physicians in accurately screening donor lungs to identify recipients most at risk of post-transplant complications.

An efficient edge/cloud medical system for rapid detection of level of consciousness in emergency medicine based on explainable machine learning models

Emergency medicine (EM) is one of the attractive research fields in which researchers investigate their efforts to diagnose and treat unforeseen illnesses or injuries. There are many tests and observations are involved in EM. Detection of the level of consciousness is one of these observations, which can be detected using several methods. Among these methods, the automatic estimation of the Glasgow coma scale (GCS) is studied in this paper. The GCS is a medical score used to describe a patient’s level of consciousness. This type of scoring system requires medical examination that may not be available with the shortage of the medical expert. Therefore, the automatic medical calculation for a patient’s level of consciousness is highly needed. Artificial intelligence has been deployed in several applications and appears to have a high performance regarding providing automatic solutions. The main objective of this work is to introduce the edge/cloud system to improve the efficiency of the consciousness measurement through efficient local data processing. Moreover, an efficient machine learning (ML) model to predict the level of consciousness of a certain patient based on the patient’s demographic, vital signs, and laboratory tests is proposed, as well as maintaining the explainability issue using Shapley additive explanations (SHAP) that provides natural language explanation in a form that helps the medical expert to understand the final prediction. The developed ML model is validated using vital signs and laboratory tests extracted from the MIMIC III dataset, and it achieves superior performance (mean absolute error (MAE) = 0.269, mean square error (MSE) = 0.625, R² score = 0.964). The resulting model is accurate, medically intuitive, and trustworthy.

Application of artificial intelligence in medical technologies: A systematic review of main trends

Objective

Artificial intelligence (AI) has been increasingly applied in various fields of science and technology. In line with the current research, medicine involves an increasing number of artificial intelligence technologies. The introduction of rapid AI can lead to positive and negative effects. This is a multilateral analytical literature review aimed at identifying the main branches and trends in the use of using artificial intelligence in medical technologies.

Methods

The total number of literature sources reviewed is n = 89, and they are analyzed based on the literature reporting evidence-based guideline PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) for a systematic review.

Results

As a result, from the initially selected 198 references, 155 references were obtained from the databases and the remaining 43 sources were found on open internet as direct links to publications. Finally, 89 literature sources were evaluated after exclusion of unsuitable references based on the duplicated and generalized information without focusing on the users.

Conclusions

This article is identifying the current state of artificial intelligence in medicine and prospects for future use. The findings of this review will be useful for healthcare and AI professionals for improving the circulation and use of medical AI from design to implementation stage.

Artificial Intelligence for Detecting and Quantifying Fatty Liver in Ultrasound Images: A Systematic Review

BACKGROUND

Non-alcoholic Fatty Liver Disease (NAFLD) is growing more prevalent worldwide. Although non-invasive diagnostic approaches such as conventional ultrasonography and clinical scoring systems have been proposed as alternatives to liver biopsy, their efficacy has been called into doubt. Artificial Intelligence (AI) is now combined with traditional diagnostic processes to improve the performance of non-invasive approaches.

OBJECTIVE

This study explores how well various AI methods function and perform on ultrasound (US) images to diagnose and quantify non-alcoholic fatty liver disease.

METHODOLOGY

A systematic review was conducted to achieve this objective. Five science bibliographic databases were searched, including PubMed, Association for Computing Machinery ACM Digital Library, Institute of Electrical and Electronics Engineers IEEE Xplore, Scopus, and Google Scholar. Only peer-reviewed English articles, conferences, theses, and book chapters were included. Data from studies were synthesized using narrative methodologies per Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria.

RESULTS

Forty-nine studies were included in the systematic review. According to the qualitative analysis, AI significantly enhanced the diagnosis of NAFLD, Non-Alcoholic Steatohepatitis (NASH), and liver fibrosis. In addition, modalities, image acquisition, feature extraction and selection, data management, and classifiers were assessed and compared in terms of performance measures (i.e., accuracy, sensitivity, and specificity).

CONCLUSION

AI-supported systems show potential performance increases in detecting and quantifying steatosis, NASH, and liver fibrosis in NAFLD patients. Before real-world implementation, prospective studies with direct comparisons of AI-assisted modalities and conventional techniques are necessary.

Interpretable prediction of mortality in liver transplant recipients based on machine learning

BACKGROUND

Accurate prediction of the mortality of post-liver transplantation is an important but challenging task. It relates to optimizing organ allocation and estimating the risk of possible dysfunction. Existing risk scoring models, such as the Balance of Risk (BAR) score and the Survival Outcomes Following Liver Transplantation (SOFT) score, do not predict the mortality of post-liver transplantation with sufficient accuracy. In this study, we evaluate the performance of machine learning models and establish an explainable machine learning model for predicting mortality in liver transplant recipients.

METHOD

The optimal feature set for the prediction of the mortality was selected by a wrapper method based on binary particle swarm optimization (BPSO). With the selected optimal feature set, seven machine learning models were applied to predict mortality over different time windows. The best-performing model was used to predict mortality through a comprehensive comparison and evaluation. An interpretable approach based on machine learning and SHapley Additive exPlanations (SHAP) is used to explicitly explain the model's decision and make new discoveries.

RESULTS

With regard to predictive power, our results demonstrated that the feature set selected by BPSO outperformed both the feature set in the existing risk score model (BAR score, SOFT score) and the feature set processed by principal component analysis (PCA). The best-performing model, extreme gradient boosting (XGBoost), was found to improve the Area Under a Curve (AUC) values for mortality prediction by 6.7%, 11.6%, and 17.4% at 3 months, 3 years, and 10 years, respectively, compared to the SOFT score. The main predictors of mortality and their impact were discussed for different age groups and different follow-up periods.

CONCLUSIONS

Our analysis demonstrates that XGBoost can be an ideal method to assess the mortality risk in liver transplantation. In combination with the SHAP approach, the proposed framework provides a more intuitive and comprehensive interpretation of the predictive model, thereby allowing the clinician to better understand the decision-making process of the model and the impact of factors associated with mortality risk in liver transplantation.

Crossroads in Liver Transplantation: Is Artificial Intelligence the Key to Donor-Recipient Matching?

Liver transplantation outcomes have improved in recent years. However, with the emergence of expanded donor criteria, tools to better assist donor-recipient matching have become necessary. Most of the currently proposed scores based on conventional biostatistics are not good classifiers of a problem that is considered "unbalanced." In recent years, the implementation of artificial intelligence in medicine has experienced exponential growth. Deep learning, a branch of artificial intelligence, may be the answer to this classification problem. The ability to handle a large number of variables with speed, objectivity, and multi-objective analysis is one of its advantages. Artificial neural networks and random forests have been the most widely used deep classifiers in this field. This review aims to give a brief overview of D-R matching and its evolution in recent years and how artificial intelligence may be able to provide a solution.