The Use of Artificial Intelligence Models for Predicting the Dynamics of Acute Pancreatitis Progression

Introduction

Machine Learning (ML) is a field of computer science that investigates and develops models using algorithms that perform well-defined tasks based on data analysis. ML relies on algorithms that incorporate rules or patterns for analysis.

Models based on ML algorithms have experienced unprece­dented breakthroughs in data analysis over the past decade. The success of artificial intelligence (AI) in medicine has become possible due to the significant increase in computational power, the demand from large corporations and the availability of data, particularly with the widespread adoption of electronic medical records (EMRs) by hospitals. AI models utilising ML algorithms are actively used for identifying disease-specific biomarkers. Several AI models have demonstrated effectiveness in predicting the occurrence of gastrointestinal disease complications [1], including liver failure [2], sepsis [3] and other conditions.

Acute pancreatitis (AP) is one of the most common gastrointestinal diseases. While mild AP occurs in approximately 80% of patients and has a favorable course, severe acute pancreatitis (SAP) is characterised by concomitant complications, organ failure (OF) and mortality ranging from 6.1% to 30% [4, 5]. Patients with AP undergo a broad spectrum of exami­nations. Evaluating the condition, disease dynamics and prognosis requires analysing these extensive data. Unlike traditional approaches and techniques that may not adequately assess and interpret data, ML allows for a comprehensive analysis of patient records, visualisation of data and the results of clinical and laboratory tests. The use of ML algorithms opens up opportunities to discover new nonobvious interactions, including those occurring over time or depending on the correlation of parameter values. Additionally, ML models can compensate for the lack of variables in “input” data processing, thereby improving the accuracy, specificity and prognostic value of the models.

This review aimed to substantiate the use of ML in predicting AP progression, assessing the dynamics of complications and OF, determining the need for surgical interventions and determining mortality.

Method

This narrative study explored the use of AI and ML me­thods for predicting the progression of acute pancreatitis. PubMed^® and Google Scholar were searched using keywords such as «machine learning acute pancreatitis», «deep learning acute pancreatitis», «artificial intelligence acute pancreatitis», «artificial neural networks acute pancreatitis» and «machine learning acute pancreatitis — complications forecasting». Additionally, specific ML models, such as «recurrent neural network», «XGBoost» and «naive Bayes» were included. Special attention was given to sources that focused on predicting complications, OF, the timing of surgical interventions, and assessing mortality risk in AP patients.

Assessment of local changes using the ML models with visualisation

Currently, research is actively being conducted on the automatic identification of the individual structures of the pancreas in both 2D and 3D visualisation formats. Although current artificial neural network models have the ability to analyse data only approximately to the level of an expert’s opinion, with each passing year, the architecture of ML mo­dels continues to improve [6].

Radiomics involves transforming visualisation data into high-dimensional objects that can be used to obtain bio­markers for the classification and prediction of diseases, which in some cases can be assessed more accurately and earlier than when processed by an expert [7]. In pancreato­logy, radiomics is used as an auxiliary tool or automated interpreter of radiological images, where AI can generate accurate and reproducible visual diagnoses, reducing the workload on physicians. Several studies have demonstrated the advantages of radiomics as a promising direction for predicting oncological processes in the pancreas [8, 9]. AI can provide automatic or semiautomatic segmentation and registration of the pancreas and its lesions. These solutions have been applied to various visualisation methods commonly used for diagnosing liver and pancreatic diseases, such as ultrasound, endoscopic ultrasound, computed tomography (CT), magnetic resonance ima­ging and positron emission tomography/CT.

Several studies have demonstrated the role of visceral adi­pose tissue in predicting OF, SAP and pancreatic necrosis [10, 11]. Visceral adipose tissue has been demonstrated to outweigh body mass index in predicting the severity of AP [12]. Fat adjacent to acinar tissue is associated with damage to the pancreas parenchyma in AP. Fat in the pancreas has a direct toxic effect on surrounding tissues in AP [13].

Yang et al. optimised ConvNet with a naive Bayes mo­del and demonstrated that the quantitative determination of the pancreatic fat fraction reached expert-level accuracy [14]. In the study by Janssens et al., 3D iterative decomposition with echo asymmetry and least squares estimation (IDEAL) MRI was used to determine the proton density fat fraction for each pancreatic slice. A convolutional neural network based on the UNet architecture was trained to autonomously determine the proton density fat fraction to calculate the parameters for the entire pancreas. The average proton density fat fraction had a strong inverse correlation with the mean Hounsfield unit without contrast-enhanced CT, demonstrating the effectiveness of the methodology for assessing pancreatic obesity [15].

The use of ML to determine the window needed for decision-making in APs

Luo et al. proposed a deep learning model based on a recurrent neural network (RNN) for individually predicting the timing of surgical intervention for necrotising pancreatitis (AUC=0.71). A seven-day time window is the best prediction interval since, if the frequency of monitoring laboratory tests is low according to EMR data, it cannot be accurately used for forecasting in a narrow window, and some laboratory indicators may undergo significant value changes over one week, complicating precise predictions in an excessively long window [16]. Additionally, a decision tree model was suggested (based on the presence of pleural effusion, HCT value, K value, and blood serum amylase level), which predicted the need for surgical intervention with 67.56% accuracy [17].

The criteria for potential outcomes

Predictions of complications

Existing models for predicting complications in AP patients are underutilised by surgeons due to their complexity, lack of integration with clinical workflows, delay of 24 hours or more for score calculation and limited understanding of how risk prediction can be used to improve outcomes.

Complications related to respiratory function play a significant role in disease dynamics. Predicting acute lung injury (ALI) using artificial neural network (ANN) [18] and logistic regression (LR) models for SAP [19] patients demonstrated high accuracy compared to traditional scales. Accurate predictions were obtained for acute respiratory distress syndrome (ARDS) in SAP patients using ANN models [20] and naive Bayes mo­dels, among other ML models [21]. The XGBoost model showed the best performance (AUC=0.84) in ARDS prediction compared to the LR, random forest (RF), SVM and decision tree (DT) models [22]. Nomograms developed on the proposed models outperform the Ranson, SOFA, BISAP, etc., scales in terms of convenience and effectiveness [23].

A series of retrospective studies demonstrated high accuracy in predicting acute kidney injury (AKI) in AP patients. Models based on the extreme gradient boosting algorithms XGBoost [24], LR [25] and RF [26] showed high accuracy in identifying patient categories with potential AKI development. The RF model accurately predicts ARDS in patients with AP accompanied by AKI [27].

Infectious complications play a crucial role in AP progression. The gradient boosting decision tree model performed better at predicting sepsis development than did the other models. For predicting intra-abdominal infection, highly accurate models using ANNs, LRs and RFs have been proposed [28]. However, incorporating the APACHE II score, CTSI score for intra-abdominal pressure, hospitalisation in the intensive care unit (ICU) and severity level complicates the practical use of these models.

FireVoxel software for radiomic studies has been used to construct a 3D volume-of-interest (VOI) on multiple slices, and a histogram of voxel values within the VOI has been created. This model predicted the occurrence of new local fluid collections, pancreatic necrosis, venous thrombosis and arterial pseudoaneurysms [29].

A separate study was dedicated to assessing the prognosis of thrombocytopenia occurrence in patients with AP. The pre­sence of thrombocytopenia significantly increases the risk of intraoperative and postoperative bleeding. A nomogram constructed using multifactorial logistic analysis demonstrated excellent prognostic effectiveness in identifying thrombocytopenia [30].

Chang et al. used the LightGBM and XGBoost models to predict mortality, length of stay in the intensive care unit (ICU) and sepsis incidence with better accuracy than did the BISAP model [31].

ANN models demonstrated high accuracy in predicting the prognosis of patients with hospital stays longer than 7 [32] and 8.4 [33] days. Models based on radiographic and laboratory data obtained upon hospitalisation.

The SMART-CT index included two components, namely, those found in the mCTSI and points specified by experts (pancreatic necrosis, number of clusters, cluster size, pre­sence of gas in clusters, ascites, pleural effusion, celiac artery and liver steatosis). SMART-CT effectively predicted mortality, intensive care unit (ICU) stay, need for surgical intervention, hospitalisation duration ≥4 weeks, number of hospitalisations ≥2, intensive care unit stay ≥2 weeks, OF and multiorgan failure (MOF) and SAP. A model built using binomial LR was used to construct a nomogram [34].

Studies have shown that models based on MLs have better performance in terms of the area under the receiver operating characteristic (AUROC) curve, sensitivity and specificity than do individual scores from traditional systems for predicting complications in AP patients.

Predictions of OF

The 2012 revised Atlanta Classification recognises the pre­sence of an OF accompanying an AP as a key cha­racteristic of severe disease. Existing systems for defining OF based on clinical and biochemical parameters, scales and imaging-based assessment systems have been developed to identify groups of patients at risk of developing OF or clinically severe disease rather than to identify individual patients. In patients with AP, the varying number, type, and sequence of organ affects over time limit the possibility of an indivi­dualised approach to decision-making within a step-up stepwise treatment strategy. Discrepancies persist in the use of the degree of the OF assessment and biomarkers primarily used for their assessment. For example, in various studies assessing respiratory failure, criteria for ARDS or ALI have been used. Timely forecasting and assessment of the dynamics of OFs prevent the progression of complications and help surgeons determine appropriate step-up approaches for the treatment of patients with AP. To date, there are no universal, accurate or easy-to-use methods available for determining the dynamics of the course of AP in the early or late phases [35]. ML models are actively being studied for predicting OF in AP.

The use of the RF model based on five cytokines accurately predicted persistent OF in the early stages of AP within 24 hours of symptom onset. The accuracy of the 5-cytokine panel was significantly greater than that of each individual cytokine [36].

The multitask and time-aware gated recurrent unit RNN (MT-GRU) demonstrated superior results in predicting OF within the next 24 hours after hospitalisation in patients with AP [37].

The bidirectional encoder representations from transformers model, a deep learning natural language processing model for text records, showed good predictive ability in determining OF when processing EMR within the first 72 hours of hospitalisation [38].

The adaptive boosting (AdaBoost) model outperformed the other ML models in predicting MOF[39]. Additionally, the ANN model demonstrated high accuracy when using a smaller number of biomarkers for predicting MOF in patients with moderate to severe AP [40].

Death prognoses

Early detection of patients at high risk of mortality is extremely important for timely surgical intervention. Approximately one-fifth of patients with potentially fatal SAPs are misidentified using traditional assessment systems [35]. Ding et al. [41] used the Medical Information Mart for Intensive Care III (MIMIC-III) database to construct ML models to predict in-hospital mortality. Patients at high risk of death can be easily identified in the early stages of AP. ALT, total calcium, and WBC were identified as prognostic markers of mortality in acute pancreatitis. The prognostic efficacy based on the AUC was 0.769 for the ANN model, 0.607 for LR, 0.652 for the Ranson’s criteria, and 0.401 for the SOFA score, respectively.

In another study, the ANN was able to accurately predict mortality from AP (accuracy: 97.5%, with a discriminative po­wer of 3.58) and proved to be more accurate than the APACHE II scoring system (accuracy: 82.4%; with a discriminative power of 1.17; p<0.01) or the GS score at 48 hours (accuracy: 80.7%; with significant discriminative power, p<0.01) [42].

The most significant factors associated with patient mortality were identified as the modified Marshall score at hospitalization and the preoperative modified Marshall score. The combination of a Generative Adversarial Network (GAN) model with LR, SVM, and RF accurately assessed patient mortality at the early stage of surgical intervention. GAN-RF (0.99) and GAN-SVM (0.99) achieved better results in evaluating key factors than GAN-LR (0.90) [43].

The Gaussian Naive Bayes (GNB) [44] and XGBoost [45] models demonstrated greater accuracy than did the other ML models in predicting mortality in intensive care unit (ICU) patients with AP, and the ANN and LR models were compared with the Glasgow Coma Scale (GCS), Ranson, Multiple Organ Dysfunction Score and APACHE II scores [46].

ML models for evaluating indications for surgical interventions

Lan et al. assessed the risk of death based on the timing of surgical intervention in patients with AP complicated by infection and pancreatic necrosis. The RF model most accurately allowed for calculating the timing of surgical intervention, considering the mentioned mortality factors [43]. IL-6, infected necrosis, fever onset, and CRP were crucial factors for determining the timing of surgical intervention (either <4 or ≥4 weeks) in patients with necrotizing pancreatitis. The classified accuracy of RF (0.80) was higher than that of SVM (0.78) and LR (0.71). Thanks to modeling conducted by GAN, the classification accuracy for all three models was improved. GAN-RF (accuracy: 0.89) performed better than GAN-SVM (0.84) and GAN-LR (0.83) [43].

Discussion

For most studies, there is a lack of validation and prospective analysis of clinical data using ML models. The evaluation of EMR data to determine recommendations for infusion and predict the timing and need for surgical intervention is a promising area for AI research. In this field, ethical issues related to the limited involvement of physicians in patient treatment need to be addressed [47].

Identifying surgical landmarks and strategies for surgical intervention are topics for further AI research. In this field, standardisation does not always reflect the current scenario during surgery. ML models that utilise multimedia data and potentially process surgical procedure videos require preprocessing «input» data. This is due to the lack of a unified system for recording surgical procedures. It is also essential for surgeons to understand the logic of «output» data, which may not always be interpretable with existing tools. In other words, if the process leading to a recommendation is not clear, the «output» data of the model are likely to be overlooked [48].

Conclusion

The utilization of ML in patients afflicted with acute pancreatitis is increasingly being scrutinized. Research endeavours pertaining to ML, specifically those concerning prognostication and the evaluation of acute pancreatitis dynamics, have showcased superior precision compared to conventional methodologies. Nonetheless, a lack of understanding regarding contemporary ML models, coupled with predispositions toward the perceived complexity of their application, impedes their integration into clinical practice. However, further research on ML models is needed to identify local complications of AP using radiomic models. Exploring nonobvious interactions with ML tools for assessing systemic changes and individual criteria will expand the ability to determine treatment tactics and surgical interventions.

References

• 1. Kroner P.T., Engels M.M., Glicksberg B.S. et al. (2021) Artificial intelligence in gastroenterology: A state-of-the-art review. World J. Gastroent., 27: 6794–6824.
• 2. Przybyszewski E.M., Simon T.G., Chung R.T. (2023) Chapter 9 — Artificial intelligence in cirrhosis complications and acute liver failure. In: Su T.-H., Kao J.-H. (Eds.) Artificial Intelligence, Machine Learning, and Deep Learning in Precision Medicine in Liver Diseases (р. 179–194): Academic Press.
• 3. Schinkel M., Paranjape K., Nannan Panday R.S. et al. (2019) Clinical applications of artificial intelligence in sepsis: A narrative review. Computers Biol. Med., 115.
• 4. Wolbrink D.R.J., van de Poll M.C.G., Termorshuizen F. et al. (2022) Trends in Early and Late Mortality in Patients With Severe Acute Pancreatitis Admitted to ICUs: A Nationwide Cohort Study. Crit. Care Med., 50(10): 1513–1521.
• 5. Schepers N.J., Bakker O.J., Besselink M.G. et al. (2019) Impact of characteristics of organ failure and infected necrosis on mortality in necrotising pancreatitis. Gut, 68: 1044–51.
• 6. Paithane P., Kakarwal S. (2023) LMNS-Net: Lightweight Multiscale Novel Semantic-Net deep learning approach used for automatic pancreas image segmentation in CT scan images. Expert Systems with Applications, 234.
• 7. Gillies R.J., Kinahan P.E., Hricak H. (2016) Radiomics: Images Are More than Pictures, They Are Data. Radiology, 278(2): 563–577. DOI: 10.1148/radiol.2015151169.
• 8. Attiyeh M.A., Chakraborty J., Doussot A. et al. (2018) Survival Prediction in Pancreatic Ductal Adenocarcinoma by Quantitative Computed Tomography Image Analysis. Ann. Surg. Oncol., 25(4): 1034–1042. DOI: 10.1245/s10434-017-6323-3.
• 9. Cassinotto C., Chong J., Zogopoulos G. et al. (2017) Resectable pancreatic adenocarcinoma: Role of CT quantitative imaging biomarkers for predicting pathology and patient outcomes. Eur. J. Radiol., 90: 152–158. DOI: 10.1016/j.ejrad.2017.02.033.
• 10. Natu A., Stevens T., Kang L. et al. (2017) Visceral Adiposity Predicts Severity of Acute Pancreatitis. Pancreas, 46(6): 776–781. DOI: 10.1097/MPA.0000000000000845.
• 11. Patel K., Trivedi R.N., Durgampudi C. et al. (2015) Lipolysis of visceral adipocyte triglyceride by pancreatic lipases converts mild acute pancreatitis to severe pancreatitis independent of necrosis and inflammation. Am. J. Pathol., 185(3): 808–819.
• 12. Tian Y., Huang Q., Ren Y.T. et al. (2023) Visceral adipose tissue predicts severity and prognosis of acute pancreatitis in obese patients. Hepatobiliary Pancreat. Dis. Int.
• 13. Acharya C., Navina S., Singh V.P. (2014) Role of pancreatic fat in the outcomes of pancreatitis. Pancreatology, 14(5): 403–8. DOI: 10.1016/j.pan.2014.06.004.
• 14. Yang J.Z., Murphy R., Lu J. (2022) A fat fraction phantom for establishing new convolutional neural network to determine the pancreatic fat deposition. Heliyon, 8: e12478.
• 15. Janssens L.P., Takahashi H., Nagayama H. et al. (2023) Artificial intelligence assisted whole organ pancreatic fat estimation on magnetic resonance imaging and correlation with pancreas attenuation on computed tomography. Pancreatology, 23(5): 556–562.
• 16. Luo J., Lan L., Peng L. et al. (2020) Predicting Timing of Surgical Intervention Using Recurrent Neural Network for Necrotizing Pancreatitis. IEEE Access, 8: 207905–207913.
• 17. Husic-Selimovic A., Bijedic N., Sofic A. et al. (2023) Prediction of Surgical Treatment in Acute Pancreatitis Using Biochemical and Clinical Parameters. Med. Arch., 77: 29–33.
• 18. Fei Y., Gao K., Li W.Q. (2018) Artificial neural network algorithm model as powerful tool to predict acute lung injury following to severe acute pancreatitis. Pancreato­logy, 18: 892–899. DOI: 10.1016/j.pan.2018.09.007.
• 19. Jia M., Xu X., Zhou S. et al. (2023) Prediction of acute lung injury in severe acute pancreatitis by routine clinical data. Eur. J. Gastroenterol. Hepatol., 35(1): 36–44.
• 20. Zou K., Ren W., Huang S. et al. (2023) The role of artificial neural networks in prediction of severe acute pancreatitis associated acute respiratory distress syndrome: A retrospective study. Medicine (Baltimore), 102(29): e34399.
• 21. Zhang M., Pang M. (2023) Early prediction of acute respiratory distress syndrome complicated by acute pancreatitis based on four machine learning models. Clinics (Sao Paulo), 78: 100215. DOI: 10.1016/j.clinsp.2023.100215.
• 22. Zhang W., Chang Y., Ding Y. et al. (2023) To Establish an Early Prediction Model for Acute Respiratory Distress Syndrome in Severe Acute Pancreatitis Using Machine Learning Algorithm. J. Clin. Med., 12(5). DOI: 10.3390/jcm12051718.
• 23. Ding N., Guo C., Song K. et al. (2022) Nomogram for the Prediction of In-Hospital Incidence of Acute Respiratory Distress Syndrome in Patients with Acute Pancreatitis. Am. J. Med. Sci., 363(4): 322–332. DOI: 10.1016/j.amjms.2021.08.009.
• 24. Qu C., Gao L., Yu X.Q. et al. (2020) Machine Learning Models of Acute Kidney Injury Prediction in Acute Pancreatitis Patients. Gastroenterol. Res. Pract., 2020: 3431290.
• 25. Cheng Y., Yang J., Wu Q. et al. (2022) Machine learning for the prediction of acute kidney injury in patients with acute pancreatitis admitted to the intensive care unit. Chin. Med. J. (Engl.), 135(23): 2886–2887.
• 26. Yang Y., Xiao W., Liu X. et al. (2022) Machine Learning-Assisted Ensemble Analysis for the Prediction of Acute Pancreatitis with Acute Kidney Injury. Int. J. Gen. Med., 15: 5061–5072. DOI: 10.2147/IJGM.S361330.
• 27. Yang D., Kang J., Li Y. et al. (2023) Development of a predictive nomogram for acute respiratory distress syndrome in patients with acute pancreatitis complicated with acute kidney injury. Ren. Fail., 45(2): 2251591. DOI: 10.1080/0886022X.2023.2251591.
• 28. Qiu Q., Nian Y.J., Tang L. et al. (2019) Artificial neural networks accurately predict intra-abdominal infection in moderately severe and severe acute pancreatitis. J. Dig. Dis., 20(9): 486–494. DOI: 10.1111/1751-2980.12796.
• 29. Iranmahboob A.K., Kierans A.S., Huang C. et al. (2017) Preliminary investigation of whole-pancreas 3D histogram ADC metrics for predicting progression of acute pancreatitis. Clin. Imaging, 42: 172–177. DOI: 10.1016/j.clinimag.2016.12.007.
• 30. Deng H., Yu X., Gao K. et al. (2021) Dynamic Nomogram for Predicting Thrombocytopenia in Adults with Acute Pancreatitis. J. Inflamm. Res., 14: 6657–6667.
• 31. Chang C.H., Chen C.J., Ma Y.S. et al. (2023) Real-time artificial intelligence predicts adverse outcomes in acute pancreatitis in the emergency department: Comparison with clinical decision rule. Acad. Emerg. Med. DOI: 10.1111/acem.14824.
• 32. Pofahl W.E., Walczak S.M., Rhone E. (1998) Predicting length stay in acute pancreatitis with a neural network. Gastroenterol., 114.
• 33. Keogan M.T., Lo J.Y., Freed K.S. et al. (2002) Outcome analysis of patients with acute pancreatitis by using an artificial neural network. Acad. Radiol., 9(4): 410–419.
• 34. Gupta P., Kumar M.P., Verma M. et al. (2021) Development and validation of a computed tomography index for assessing outcomes in patients with acute pancreatitis: “SMART-CT” index. Abdom. Radiol. (NY), 46(4): 1618–1628.
• 35. Rau B.M. (2007) Predicting severity of acute pancreatitis. Curr. Gastroenterol. Rep., 9(2): 107–115. DOI: 10.1007/s11894-007-0004-5.
• 36. Langmead C., Lee P.J., Paragomi P. et al. (2021) A Novel 5-Cytokine Panel Outperforms Conventional Predictive Markers of Persistent Organ Failure in Acute Pancreatitis. Clin. Transl. Gastroenterol., 12(5): e00351.
• 37. Luo J., Lan L., Huang S. et al. (2023) Real-time prediction of organ failures in patients with acute pancreatitis using longitudinal irregular data. J. Biomed. Inf., 139: 104310.
• 38. Luo J., Lan L., Yang D. et al. (2021) Early Prediction of Organ Failures in Patients with Acute Pancreatitis Using Text Mining. Scientific Programming, 2021: 1–7.
• 39. Xu F., Chen X., Li C. et al. (2021) Prediction of Multiple Organ Failure Complicated by Moderately Severe or Severe Acute Pancreatitis Based on Machine Learning: A Multicenter Cohort Study. Mediators Inflamm., 2021: 5525118.
• 40. Qiu Q., Nian Y.J., Guo Y. et al. (2019) Development and validation of three machine-learning models for predicting multiple organ failure in moderately severe and severe acute pancreatitis. BMC Gastroenterol., 19(1): 118.
• 41. Ding N., Guo C., Li C. et al. (2021) An Artificial Neural Networks Model for Early Predicting In-Hospital Mortality in Acute Pancreatitis in MIMIC-III. Biomed. Res. Int., 2021: 6638919. DOI: 10.1155/2021/6638919.
• 42. Mofidi R., Duff M.D., Madhavan K.K. et al. (2007) Identification of severe acute pancreatitis using an artificial neural network. Surgery, 141(1): 59–66.
• 43. Lan L., Guo Q., Zhang Z. et al. (2020) Classification of Infected Necrotizing Pancreatitis for Surgery Within or Beyond 4 Weeks Using Machine Learning. Front. Bioeng. Biotechnol., 8: 541. DOI: 10.3389/fbioe.2020.00541.
• 44. Ren W., Zou K., Huang S. et al. (2023) Prediction of in-hospital Mortality of Intensive Care Unit Patients with Acute Pancreatitis Based on an Explainable Machine Learning Algorithm. J. Clin. Gastroenterol. DOI: 10.1097/MCG.0000000000001910.
• 45. Pang K., Li L., Ouyang W. et al. (2022) Establishment of ICU Mortality Risk Prediction Models with Machine Learning Algorithm Using MIMIC-IV Database. Diagnostics (Basel), 12(5). DOI: 10.3390/diagnostics12051068.
• 46. Halonen K.I., Leppaniemi A.K., Lundin J.E. et al. (2003) Predicting fatal outcome in the early phase of severe acute pancreatitis by using novel prognostic models. Pancreatology, 3(4): 309–315. DOI: 10.1159/000071769.
• 47. Satapathy P., Hermis A.H., Rustagi S. et al. (2023) Artificial intelligence in surgical education and training: opportunities, challenges, and ethical considerations-correspondence. Int. J. Surg., 109(5): 1543–1544.
• 48. Satapathy P., Pradhan K.B., Rustagi S. et al. (2023) Application of machine learning in surgery research: current uses and future directions-editorial. Int. J. Surg., 109(6): 1550–1551. DOI: 10.1097/JS9.0000000000000421.

Надійшла до редакції/Received: 19.04.2024
Прийнято до друку/Accepted: 24.04.2024