Understanding Machine Learning Applications in Lung Transplantation: A Narrative Review

Abstract

Lung transplantation (LTx) offers life-saving therapy for patients with end-stage lung disease but remains limited by donor shortages, complex postoperative management and graft failure. Machine learning (ML) enables opportunities to address these challenges by identifying patterns in complex, high-dimensional data, thereby providing novel insights and improving outcomes. This review outlines ML studies in LTx and explains the methodologies. ML has demonstrated promising results in organ allocation and outcome prediction. Techniques such as support vector machines, and deep learning are useful in risk stratification, while methods like random forests improve interpretability and transfer learning supports model development in data-scarce settings. ML has a growing role in multi-omics data and imaging-based diagnostics. Despite promising results, barriers such as small datasets, cross-center inconsistency, poor interpretability, and limited external validation, hinder clinical adoption. Future progress requires multicenter collaborations, transparent methodologies, and integration within clinical workflows. ML should serve as complementary tool that enhances decision-making, rather than replacing clinical judgement. With careful implementation, it holds the potential to improve transplant outcomes.

Introduction

Lung transplantation (LTx) is a life-saving treatment for end-stage lung disease. Despite surgical and perioperative advances, challenges remain, including donor shortage, primary graft dysfunction (PGD), and chronic lung allograft dysfunction (CLAD). As clinical data expand and pathophysiology is better understood, these challenges also increase in complexity. Traditional decision-making and predictive modelling is therefore limited.

Machine learning (ML), can identify complex, non-linear patterns, supporting outcome prediction and personalized care [1–5]. In solid organ transplantation, ML is increasingly used to predict survival and improve organ allocation [6]. Nonetheless, integration in LTx lags behind due to small, heterogeneous datasets and complex pathways [7].

The aim of this narrative review is twofold. First, to provide clinicians with a conceptual foundation that fosters understanding of ML. Second, to explore ML applications in LTx, covering outcome prediction, organ allocation, imaging, omics, and other applications.

Principles of Machine Learning

ML enables mathematical models to learn from data, identify patterns, and make predictions with minimal human intervention. By leveraging algorithms, ML models extract insights and predict outcomes [1]. ML is a central component of artificial intelligence (AI) and closely connected to data science and computer science. These domains overlap (Figure 1) in methodologies, applications, and objectives, making clear distinction difficult [1, 3–5].

FIGURE 1

ML employs datasets specific for the task. In medical datasets, clinical factors (e.g., age, smoking) serve as dimensions (features), while individual observations (e.g., patients, images) represent samples (data points). Based on whether labeled data (samples with known outputs) are used, ML approaches can be classified as supervised, unsupervised, and semi-supervised [1–5, 8].

Supervised ML uses labeled data to train predictive models [1–5, 8]. To ensure generalizability, datasets are divided into training, validation, and testing subsets. Models first learn patterns from the training set. The validation set aids in hyperparameter tuning (e.g., batch size, learning rate). It detects underfitting and overfitting, meaning that the model is too simple to capture the true patterns, or learns the noise in the data, respectively (Figure 2) [1–3, 8]. Cross-validation is used to ensure generalizability by partitioning the dataset into training and validation subsets. An approach is k-fold cross-validation, which divides data randomly into k (a number) folds. The model is trained on k-1 folds and validated on the remaining one, repeating this process k times so each subset serves as validation once [1, 2, 5, 8]. Cross-validation ensures the model outcomes are robust and not dependent on a single random split of the dataset [1, 2, 5, 8]. Finally, the test set, an unseen portion of data, is used to evaluate the final model performance [1, 2, 8].

FIGURE 2

Supervised ML is used for classification and regression. Both utilize labeled datasets, but differ in output: classification predicts categories, regression predicts continuous values [1, 2, 5, 8].

Conversely, unsupervised ML analyzes unlabeled data to identify patterns [1–3, 5, 8]. Choosing between supervised and unsupervised learning can be difficult, particularly when labeled data are scarce. Semi-supervised ML bridges this gap by combining limited labeled data alongside many unlabeled samples, useful in medical research where data annotation is resource-intensive [1, 2, 8]. Commonly used ML methods, shown in Figure 3, are evaluated and compared using diverse metrics (Table 1).

FIGURE 3

State-of-the-Art of Machine Learning in Lung Transplantation

LTx involves a heterogeneous, limited patient population with extensive data. LTx recipients have worse outcome than other solid organ transplant recipients, highlighting persistent gaps. ML could contribute to personalized treatment and improved outcomes, as seen in other transplants [5, 9, 10].

The following section reviews key studies, as far as we know (2004–2025), organized into: (1) outcome prediction, (2) organ allocation, and (3) imaging, omics, and other applications. A summary is presented in Table 2. Studies using simpler, borderline-ML methods are excluded from the main text but included in Table 2 and Figure 3.

Outcome Prediction

Survival and Quality of Life

In a series of studies, Oztekin, Delen, Amini and colleagues demonstrated the value of ML for outcome prediction. Initially, they showed that ML outperformed expert–selected variables and traditional statistical models in predicting 9-year graft survival after heart–lung transplantation, identifying more relevant variables and relationships [12]. They applied logistic regression (Supplementary Text, Figure 3A.2), decision trees (DTs), and artificial neural networks (ANNs). DTs (Figure 3A.6) are interpretable models that recursively split data to form rule-based trees. They are sensitive to noise and require pruning (removing unnecessary parts) to improve generalizability [1, 2, 4, 5, 8]. ANNs (Figure 3A.8) are algorithms inspired by the brain (Figures 4A–D). The simplest form, a single-layer perceptron, mimics a biological neuron. Adding hidden layers, referring to synaptic connections creates a multilayer perceptron (MLP) [1, 2, 4, 5, 8]. Unlike DTs, ANNs lack interpretability and rely on large datasets, therefore, the United Network for Organ Sharing (UNOS) cohort of 16,604 patients was crucial for this approach [5, 8].

FIGURE 4

Later, their work was extended to survival estimation, again comparing ML with expert-selected and literature-based variables. ML outperformed both approaches by retaining important predictors overlooked in traditional methods. They applied DTs and ANNs, and additionally introduced support vector machines (SVMs) [13]. SVMs (Figure 3A.5) are algorithms that maximize the margin between classes (distance between the decision boundary and the nearest data points from each class). An innovation is the kernel trick, which enables SVMs to classify nonlinearly separable data by mapping it into higher-dimensional space (Figure 5) [1–5, 8]. Model performance was compared using Cox regression (Supplementary Text, Figure 3A.3). Subsequently, k-means clustering, two-step cluster analysis, and conventional heuristic approaches were used to determine the optimal number of patient risk groups. Unsupervised k-means clustering (Figure 3B.1) groups data into a predefined number of clusters based on feature similarity by iteratively assigning samples to the nearest centroid (center of a cluster) and updating centroids as the mean of assigned samples. It offers an unbiased way to explore risk groups [1–5, 8]. In this study, three clusters were optimal [13].

FIGURE 5

In 2011, a DT–based hybrid model was designed to provide an interpretable ML approach. However, its accuracy remained low. Moreover, using variables predefined from previous studies biased the model, potentially missing important interactions [14]. To predict quality of life, Genetic Algorithm (GA)-based approaches for feature selection were introduced [16], particularly useful for complex, feature-rich domains with limited samples as in LTx. GAs (Figure 3C.1) are optimization techniques inspired by biological evolution, using selection, crossover, and mutation to find optimal solutions, e.g., determining representative variables [5, 59]. The GA was combined with three classification algorithms: SVM, ANN and k-Nearest Neighbors (kNN) (Figure 3A.7). Unlike other algorithms, kNN predicts without training, by averaging outcomes of the k most similar samples to unseen input. Performance depends on data quality, choice of distance metric, and k. In high-dimensional data, kNN’s accuracy can degrade [1, 2, 5, 8], therefore, combining it with GA is appropriate.

Subsequent research performed classification of post-LTx survival (≤1 year vs. ≥10 years), incorporating additional methods, namely ensemble models such as random forests (RF) and gradient boosting trees [21]. Ensemble learning combines multiple models to improve predictive accuracy, reduce overfitting, and enhance robustness [1, 2, 5, 8]. Bagging (bootstrap aggregating) (Figure 3A.10.1) improves stability by training on different data subsets [1, 2, 5, 8]. RF is a common bagging method that aggregates DTs [1, 5, 8]. Boosting (Figure 3A.10.2) builds models sequentially, each correcting errors of the previous one [1, 5, 8]. Among all models, RF achieved the best performance. To improve model transparency, the authors employed an explainable AI (XAI) method: SHapley Additive Explanations (SHAP), a model-agnostic framework that quantifies each feature’s contribution to a prediction by considering all possible feature combinations [60]. SHAP identified Hepatitis B surface antibody and forced expiratory volume in one second (FEV1) as predictors of long-term survival. However, methodological limitations warrant consideration. The use of binary classification (≤1 year vs. ≥10 years) excluded nearly half of the cohort [21]. This neglects intermediate survival, arguably the most challenging to predict, which makes the modest performance noticeable.

Moro et al. created a DT for survival predictions. Using UNOS data, 47 features were identified via stepwise logistic regression, assuming linear relationships. Consequently, meaningful nonlinear interactions may have been missed, and reducing 60 to 47 variables offered minimal dimensional or computational benefit. The final DT used six key predictors, including three postoperative variables, limiting the model’s preoperative prognostic utility, despite its interpretability. Eight subgroups (decision nodes) showed distinct survival curves. As expected, best outcomes occurred in younger recipients with short hospital stays, limited ventilation support, and no reintubation [25].

To compare survival between increased risk for disease transmission (IRD) organ recipients versus non-IRD organ recipients, Mark et al. applied RF and Cox regression. As Cox regression performed best, it was selected for further analysis, which somewhat diminished the novelty of ML implementation. Nevertheless, the study offered a data-driven perspective to expand the donor pool, demonstrating a 7.2% improvement in 5-year survival for IRD lung transplant recipients [17].

Unlike the prior study, Tian et al. demonstrated that RF can outperform Cox regression, for survival prediction under standard conditions, achieving high predictive accuracy. Generalizability across subgroups with different diagnoses and treatments was reported. However, the single-center design and limited sample size may question this [22].

The effectiveness of RF, combined DTs, was also shown by Fessler et al., analyzing 284 variables across 12 perioperative stages to predict one-year mortality. As presumed, the accuracy went up by including information of later stages. Lung allocation score (LAS) emerged as top predictor [18].

Primary Graft Dysfunction

A subsequent study by Fessler et al. used gradient boosting to predict PGD3, a syndrome linked to adverse outcomes [61]. Extracorporeal membrane oxygenation use, along with recipient factors, were revealed as top predictors [20]. Due to the short length of these papers [18, 20], the information provided on the ML implementation is limited. In their most recent paper [28], predicting PGD3 at 72h, they offer more information about logistic regression and XGBoost, an efficient gradient boosting variant, that improves computational memory usage, well-suited for large datasets [62]. Fessler’s studies introduce an innovative approach by progressively incorporating data from successive transplant phases, allowing the prognosis to be refined at each stage.

Michelson et al. similarly predicted PGD3 using pretransplant data, enabling potential application in patient selection and pretransplant counseling. From 100 features, Least Absolute Shrinkage and Selection Operator (LASSO) (Supplementary Text) selected 11 predictors. Among four models, kNN performed best and was released as open-access risk calculator [26].

With data from 802 patients, Xia et al. evaluated nine algorithms. RF classified PGD3 best. SHAP identified blood loss as important, but prior feature selection, based on linear relation assumption, may have introduced selection bias [27].

Other Outcome Parameters

Using a small, unbalanced dataset, Tian et al. developed eight ML models combined with seven feature selection methods to predict airway stenosis requiring clinical intervention. Key predictors in RF included postoperative 6-minute walk test and indication for LTx. This model could guide postoperative follow-up [24].

Braccioni et al. assessed how clinical parameters relate to symptom severity during exercise testing after LTx. Boruta, a feature selection method based on RF [63], revealed associations for limited exercise capacity: dyspnea correlating with peak ventilation and work rate, muscle effort with breathing reserve, and muscle pain with VO₂ peaks. These findings linked reduced aerobic capacity and high ventilatory cost to symptom severity. DT visualizations offered interpretable insights to guide exercise prescriptions [19]. Despite the small dataset (n = 24), the authors justified using ML, noting the method performs well in small, high-dimensional datasets without assuming normality or independence. Nonetheless, small cohorts increase overfitting risk and limit generalizability of the findings.

To analyze repeated FEV1 measurements after LTx, Pande et al. developed a longitudinal model, handling challenges as within-subject correlation, unequal time intervals, and unbalanced designs. Although FEV1 typically declines over time, patterns vary with individual factors. The method was clearly described and implemented in an R package [15].

Overall, the studies reviewed above show the potential of ML in LTx, but the applications stay rather limited. Stronger tools, e.g., deep learning (DL), could be implemented, as seen in section Organ Allocation [33].

Organ Allocation

LTx faces suboptimal organ allocation, causing long wait times and significant candidate mortality [64]. Varying donor selection criteria across centers limits organ availability. Allocation studies suffer from bias, as unaccepted organs are absent in training datasets. Unlike other transplants with comprehensive donor-recipient risk stratification, LTx allocation largely neglects the combined influence of factors [30].

To address these challenges, Zafar et al. developed the LTx Advanced Prediction Tool (LAPT). Based on 15,124 UNOS cases, LAPT grouped patients into low-, medium-, and high-risk subsets. LAPT outperformed LAS by predicting 1-, 5-, and 10-year survival and graft half-life for donor-recipient matches. This web-based tool enables data-driven allocation beyond recipient-centric systems [30].

Dueñas-Jurado et al. combined logistic regression with ANNs for donor-recipient matching. They incorporated donor, recipient and perioperative variables to predict 6-month graft survival, claiming to outperform traditional methods, although metrics were not reported. Key predictors included low pre-transplant CO₂, while prolonged donor ventilation, older donor and recipient age were linked to poorer outcomes [29].

To assess the suitability of donor lungs, Sage et al. created InsighTx, a RF model integrating ex vivo lung perfusion (EVLP) and other variables, offering a quantitative approach to evaluate and improve lung utilization [32]. However, its primary endpoint, extubation time, serves only as a short-term proxy for success and does not fully capture longer-term outcomes.

Pu et al. developed eight ML models using donor demographics to predict lung, heart, and thoracic cavity volumes, to improve donor-recipient size matching [33]. The performance of these approaches was benchmarked against convolutional neural network (CNN)-based image segmentation models, which were used to generate the volumetric ground truth. CNNs are a class of DL (Figure 3A.9), referring to ANNs with multiple hidden layers, designed to process structured grid-like data like images. They use filters to detect local structures (e.g., edges) and combine them to recognize shapes. Like other DL models, it requires large labeled datasets and significant processing power [1–3, 8]. The best-performing model was a MLP for individual lungs and thoracic cavity estimates. These non-imaging-based volume predictions may enhance allocation [33].

In contrast to these optimistic findings, Brahmbhatt et al. concluded that LAS, clinician-based models, LASSO, and RF are not sufficiently accurate to predict post-LTx survival. LAS overestimated mortality in high-risk patients and the AUROC of the Houston Methodist model was not achieved, highlighting challenges of reproducibility and possible overfitting in earlier literature. Predictive performance was not improved by ML, disease-specific models, or donor variables [31].

Similarly, Dalton et al. reported that LAS refinement and advanced techniques did not improve performance. Seven models were evaluated with waitlist and post-transplant data to predict waitlist mortality or post-transplant survival. While waitlist models showed strong discrimination, all post-transplant models performed poorly [34]. A possible solution is integrating images or biological markers. Studies employing these approaches are examined in section Imaging, Omics and Other Applications.

Imaging, Omics and Other Applications

Barbosa et al. investigated quantitative CT (qCT) to diagnose bronchiolitis obliterans syndrome (BOS), a form of CLAD. Logistic regression and SVM were used to compare qCT metrics, pulmonary function tests (PFT), and semi-quantitative imaging scores as input. To reduce qCT dimensionality, principal component analysis (PCA) (Figure 3B.2) was applied, projecting the data onto components capturing the highest variance while minimizing information loss [1, 2, 5, 8]. PCA of qCT together with PFT outperformed all models. However, BOS diagnosis relied solely on chart-reviewed PFT decline, creating circularity, lacking pathological confirmation, and potentially biasing comparisons between qCT- and PFT-based models [36]. In a subsequent study, qCT features including lobar volumes, airway volumes, and airway resistance differed significantly in BOS patients, even at baseline. Using SVM, they constructed classifiers in one-, two-, and three-dimensional feature spaces. Remarkably, with only three qCT parameters, the model achieved 85% accuracy in predicting BOS [38]. Bartholmai et al. also used qCTs, to develop the CALIPER platform for interstitial lung diseases. They applied different ML methods to categorize lung parenchyma into five patterns, challenging even for expert readers to distinguish. CALIPER provided 3D visualizations for tracking of disease burden [35]. Later, McInnis et al. tested CALIPER to distinguish CLAD phenotypes and predict graft survival. Both CALIPER and radiologist scores independently predicted graft failure, with CALIPER enabling reproducible phenotyping and early prognostication without requiring expiratory CT [45]. An XGBoost model based on X-rays and perfusion data from EVLP was developed to predict transplant suitability and ventilation duration post-LTx. Abnormalities were scored per lobe and correlated with oxygenation, compliance and edema. SHAP ranked consolidation and infiltrates as strongest associated with function and transplantability [54]. These studies illustrate how ML-driven imaging analysis can overcome interobserver variability, provide objective and reproducible quantification, reduce human workload, and enable more accurate, scalable assessment of graft injury.

Tran-Dinh et al. developed a model to predict acute cellular rejection using soluble CD31 (sCD31) as biomarker. From only forty recipients, sCD31 levels were combined with recipient haematosis in a CNN model [46]. The authors claim their model uses concepts similar to transfer learning (Figure 3C.2), where a model trained on one task is adapted to another, valuable in data-scarce settings [1, 2]. However, this is questionable, as their network was trained from scratch rather than optimized from a pretrained model. In another study, a topological autoencoder (Taelcore) was created to improve these predictions by capturing underlying data structures. Applied to the same dataset, dimensionality reduction with Taelcore achieved more accurate predictions than methods like PCA [55]. Likewise, features extracted by Taelcore lack biological interpretability.

To predict tacrolimus trough levels (TTLs) in LTx patients, Choshi et al. developed a long short-term memory–based Recurrent Neural Network (RNN), a DL model handling sequential data. This approach relied on clinical inputs identified by SHAP, including previous TTLs and tacrolimus doses. The model captured temporal patterns in dosing and drug response, enabling individualized immunosuppressant management [58]. Yet, its accuracy may diminish in real-world patient settings where missed doses and irregular timing are common.

A gene expression–based DL classifier by Cantu et al. used preprocurement donor lung biopsies to predict PGD3. Their Toll-like receptor model outperformed clinical covariates [40], demonstrating strong discriminative ability and indicating donor innate immune activation as a key driver of PGD, though the analysis was limited to two pathways. Gao et al. also used transcriptomic data in different algorithms. Four neutrophil extracellular traps-related hub genes were identified as drivers of ischemia-reperfusion injury. Three of these were validated in clinical samples, related with PGD development [56]. Furthermore, transcriptomic data were used to explore cuproptosis, a form of cell death, as a potential mechanism in ischemia-reperfusion injury. Three methods (LASSO, SVM, RF) recognized critical biomarkers, with good performance. Functional enrichment linked these genes to immune regulation and cell death, while immune infiltration analysis revealed associations with distinct immune cell subsets [51].

Using unsupervised ML on LTx transbronchial biopsies, Halloran et al. defined four rejection archetypes. PCA linked T-cell mediated rejection (TCMR) and injury to T cell and macrophage transcripts, and antibody-mediated rejection-like to endothelial markers [39]. They also showed that this method worked for mucosal biopsies [41]. However, because mucosal biopsies were obtained only during protocol or clinically indicated bronchoscopies, the sampling may be biased toward unwell patients, limiting generalizability to asymptomatic recipients. Molecular TCMR was associated with future graft loss. Molecular scores outperformed clinical variables in RF and remained robust even in low-surfactant or mucosal samples [42]. Across these studies, Halloran et al. demonstrate that molecular profiling of biopsies provides a more biologically coherent assessment of rejection than histology, although the work remains limited by sampling bias, nonspecific injury signals, and small sample size. Using previously reported mucosal biopsy data [41], Zhang et al. classified recipients into four rejection-related subgroups. Supervised classification achieved high accuracies (likely overfitted: more features than samples) and lacked external validation. Predictive genes were linked to T cell signaling and innate immunity [47].

In another study, lymphocytic bronchitis gene signature in transbronchial biopsies and small airway brushings were used to predict graft failure and differentiate CLAD from controls. Gene expression profiling with RF showed superior diagnostic performance for brushings over biopsies, but because brushings contain mixed epithelial and leukocyte populations, cell-type–specific interpretation remains limited. The lymphocytic bronchitis score was elevated in CLAD and associated with 2.4-fold increased risk of graft loss [43].

Su et al. analyzed 181 sputum samples from 59 recipients using 16S rRNA sequencing, classifying samples into “stable”, “infection”, and “rejection”. Differences in microbial composition appeared, with six genera enriched during acute rejection, suggesting immune-modulatory roles. Integrating these genera and clinical data in a RF classified well, though repeated samples per patient may cause biased results [48]. A study by Weigt described that gene expression profiling of cells in bronchoalveolar lavage (BAL) revealed an immune activation signature preceding clinical CLAD diagnosis. Forty genes were differentially expressed in incipient CLAD versus CLAD-free samples, enriched for cytotoxic lymphocyte markers. SVM achieved 94.1% accuracy in distinguishing only seventeen cases [37]. Berra et al. also used BAL samples to predict CLAD and investigate the association with the renin–angiotensin system. Although single proteins could not discriminate, combinations in ML classifiers can, reflecting ML’s strength in modelling beyond human assessment [44].

Another study predicted PGD using volatile organic compounds (VOCs) from BAL fluid and bronchial aspirate samples. VOC profiling with SVM modeling achieved 83% accuracy in distinguishing PGD3 from lower grades. Twenty VOCs, associated with lipid peroxidation and oxidative stress, were top predictors. Additional analyses linked VOC patterns to clinical variables, including donor BMI and Organ Care System, indicating potential confounding. Recipient and intraoperative factors did not significantly influence VOC profiles [50].

Key Insights, Future Directions and Conclusion

A consistent strength of ML is its ability to integrate many weak or noisy features into a meaningful signal, where human interpretation or single-variable analyses fail. ML can capture complex, nonlinear interactions, reveal hidden patterns, and offer early risk stratification that traditional clinical or statistical methods miss. Yet, the limitations across studies are strikingly uniform. Most studies are small, single-center, only internally validated, and based on imbalanced datasets. Sampling bias, missing confounders, and heterogeneous data quality further reduce generalizability. Compared with kidney, liver, and heart transplantation, where ML-based tools are more mature, ML approaches in LTx research remains largely underexplored [65–74]. Reporting is often insufficient: many papers provide limited mathematical detail about model design, preprocessing, hyperparameter tuning, or validation, making replication difficult and hindering fair comparison across studies. More transparent, standardized reporting following frameworks like MI-CLAIM (Minimum Information about Clinical Artificial Intelligence Modeling) and TRIPOD-AI (Transparent Reporting of a Multivariable prediction model for individual Prognosis Or Diagnosis) should be strongly encouraged.

Future Directions & Underused Advanced Methods

Future directions should include more multimodal datasets, true external validation, and the careful use of advanced ML methods. Stacking, an ensemble model, could improve performance by combining diverse base learners and a meta learner. Generative Adversarial Networks (GANs) could augment datasets. They consist of a generator that creates synthetic data and a discriminator that evaluates authenticity. Through adversarial training, based on unlabeled data, both networks iteratively improve, allowing to generate realistic data. Although the information content does not increase, it enhances model flexibility and generalization. These are only two examples of underused ML methods, that could strengthen model performance. Post-hoc explanation tools such as Local Interpretable Model-agnostic Explanations [5] and SHAP will remain essential to ensure that predictions are clinically interpretable.

Conclusion

ML holds major potential in LTx, from improving outcome prediction and organ allocation, to imaging and omics-based insights. Yet, clinical adoption remains limited due to small, single-center datasets and insufficient external validation. Enhancing generalizability and building trust requires large multicenter studies, XAI, and standardized reporting. Additionally, ethical considerations remain important when using ML in medicine [2]. Progress in other solid organ transplants highlights opportunities for LTx, with techniques still unexplored, offering room for future innovation. Crucially, ML should complement clinical decision-making, and not replace clinical judgement. Its success relies on collaboration among clinicians, data scientists, ethicists, and regulators. Overcoming current barriers will enable ML to meaningfully improve transplant outcomes.

References

• 1.

  Burkov A . The Hundred Page Machine Learning Book.

  • Google Scholar

• 2.

  Goodfellow I Bengio Y Courville A . Deep Learning. The MIT Press (2016).

  • Google Scholar

• 3.

  Lim CP Vaidya A Chen YW Jain V Jain LC . Artificial Intelligence and Machine Learning for Healthcare: Emerging Methodologies and Trends, 2. Springer (2023).

  • Google Scholar

• 4.

  Lim CP Vaidya A Chen YW Jain T Jain LC , editors. Artificial Intelligence and Machine Learning for Healthcare: Vol. 1: Image and Data Analytics, 228. Springer International Publishing (2023). 10.1007/978-3-031-11154-9

  • CrossRef
  • Google Scholar

• 5.

  Simon GJ Aliferis C , editors. Artificial Intelligence and Machine Learning in Health Care and Medical Sciences: Best Practices and Pitfalls. Springer International Publishing (2024). 10.1007/978-3-031-39355-6

  • CrossRef
  • Google Scholar

• 6.

  Gholamzadeh M Abtahi H Safdari R . Machine Learning-Based Techniques to Improve Lung Transplantation Outcomes and Complications: A Systematic Review. BMC Med Res Methodol (2022) 22(1):331. 10.1186/s12874-022-01823-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 7.

  Rampolla R . Lung Transplantation: An Overview of Candidacy and Outcomes. Ochsner J (2014) 14(4):641–8.

  • Pubmed Abstract
  • Google Scholar

• 8.

  James G Witten D Hastie T Tibshirani R Taylor J . An Introduction to Statistical Learning: With Applications in Python. Springer International Publishing (2023). 10.1007/978-3-031-38747-0

  • CrossRef
  • Google Scholar

• 9.

  Connor KL O’Sullivan ED Marson LP Wigmore SJ Harrison EM . The Future Role of Machine Learning in Clinical Transplantation. Transplantation (2021) 105(4):723–35. 10.1097/TP.0000000000003424

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 10.

  He J Baxter SL Xu J Xu J Zhou X Zhang K . The Practical Implementation of Artificial Intelligence Technologies in Medicine. Nat Med (2019) 25(1):30–6. 10.1038/s41591-018-0307-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11.

  Troiani JS Carlin BP . Comparison of Bayesian, Classical, and Heuristic Approaches in Identifying Acute Disease Events in Lung Transplant Recipients. Stat Med (2004) 23(5):803–24. 10.1002/sim.1651

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 12.

  Oztekin A Delen D Kong Z . Predicting the Graft Survival for Heart–Lung Transplantation Patients: An Integrated Data Mining Methodology. Int J Med Inf (2009) 78(12):e84–e96. 10.1016/j.ijmedinf.2009.04.007

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13.

  Delen D Oztekin A Kong Z . A Machine Learning-Based Approach to Prognostic Analysis of Thoracic Transplantations. Artif Intell Med (2010) 49(1):33–42. 10.1016/j.artmed.2010.01.002

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14.

  Oztekin A Kong ZJ Delen D . Development of a Structural Equation Modeling-Based Decision Tree Methodology for the Analysis of Lung Transplantations. Decis Support Syst (2011) 51(1):155–66. 10.1016/j.dss.2010.12.004

  • CrossRef
  • Google Scholar

• 15.

  Pande A Li L Rajeswaran J Ehrlinger J Kogalur UB Blackstone EH et al Boosted Multivariate Trees for Longitudinal Data. Mach Learn (2017) 106(2):277–305. 10.1007/s10994-016-5597-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16.

  Oztekin A Al-Ebbini L Sevkli Z Delen D . A Decision Analytic Approach to Predicting Quality of Life for Lung Transplant Recipients: A Hybrid Genetic Algorithms-Based Methodology. Eur J Oper Res (2018) 266(2):639–51. 10.1016/j.ejor.2017.09.034

  • CrossRef
  • Google Scholar

• 17.

  Mark E Goldsman D Keskinocak P Sokol J . Using Machine Learning to Estimate Survival Curves for Patients Receiving an Increased Risk for Disease Transmission Heart, Liver, or Lung Versus Waiting for a Standard Organ. Transpl Infect Dis (2019) 21(6):e13181. 10.1111/tid.13181

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18.

  Fessler J Gouy-Pailler C Fischler M Guen ML . Machine Learning in Lung Transplantation. J Heart Lung Transpl (2020) 39(4):S385. 10.1016/j.healun.2020.01.497

  • CrossRef
  • Google Scholar

• 19.

  Braccioni F Bottigliengo D Ermolao A Schiavon M Loy M Marchi MR et al Dyspnea, Effort and Muscle Pain During Exercise in Lung Transplant Recipients: An Analysis of Their Association with Cardiopulmonary Function Parameters Using Machine Learning. Respir Res (2020) 21(1):267. 10.1186/s12931-020-01535-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20.

  Fessler J Vallee A Gouy-Pailler C Davignon M Fischler M Guen ML . Machine-Learning for Primary Graft Dysfunction in Lung Transplantation. J Heart Lung Transpl (2021) 40(4):S380. 10.1016/j.healun.2021.01.1069

  • CrossRef
  • Google Scholar

• 21.

  Amini M Bagheri A Delen D . An Explanatory Analytics Model for Identifying Factors Indicative of Long-Versus Short-Term Survival After Lung Transplantation. Decis Anal J (2022) 3:100058. 10.1016/j.dajour.2022.100058

  • CrossRef
  • Google Scholar

• 22.

  Tian D Yan HJ Huang H Zuo YJ Liu MZ Zhao J et al Machine Learning–Based Prognostic Model for Patients After Lung Transplantation. JAMA Netw Open (2023) 6(5):e2312022. 10.1001/jamanetworkopen.2023.12022

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23.

  Melnyk V Xu W Ryan JP Karim HT Chan EG Mahajan A et al Utilization of Machine Learning to Model the Effect of Blood Product Transfusion on Short‐Term Lung Transplant Outcomes. Clin Transpl (2023) 37(6):e14961. 10.1111/ctr.14961

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 24.

  Tian D Zuo YJ Yan HJ Huang H Liu MZ Yang H et al Machine Learning Model Predicts Airway Stenosis Requiring Clinical Intervention in Patients After Lung Transplantation: A Retrospective Case-Controlled Study. BMC Med Inform Decis Mak (2024) 24(1):229. 10.1186/s12911-024-02635-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 25.

  Moro A Janjua HM Rogers MP Kundu MG Pietrobon R Read MD et al Survival Tree Provides Individualized Estimates of Survival After Lung Transplant. J Surg Res (2024) 299:195–204. 10.1016/j.jss.2024.04.017

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 26.

  Michelson AP Oh I Gupta A Puri V Kreisel D Gelman AE et al Developing Machine Learning Models to Predict Primary Graft Dysfunction After Lung Transplantation. Am J Transpl (2024) 24(3):458–67. 10.1016/j.ajt.2023.07.008

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 27.

  Xia W Liu W He Z Song C Liu J Chen R et al Machine Learning for Predicting Primary Graft Dysfunction After Lung Transplantation: An Interpretable Model Study. Transplantation (2025) 109:1458–70. 10.1097/TP.0000000000005326

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28.

  Fessler J Gouy-Pailler C Ma W Devaquet J Messika J Glorion M et al Machine Learning for Predicting Pulmonary Graft Dysfunction After Double-Lung Transplantation: A Single-Center Study Using Donor, Recipient, and Intraoperative Variables. Transpl Int (2025) 38:14965. 10.3389/ti.2025.14965

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 29.

  Dueñas-Jurado JM Gutiérrez PA Casado-Adam A Santos-Luna F Salvatierra-Velázquez A Cárcel S et al New Models for Donor-Recipient Matching in Lung Transplantations. Plos One (2021) 16(6):e0252148. 10.1371/journal.pone.0252148

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 30.

  Zafar F Hossain MM Zhang Y Dani A Schecter M Hayes D Jr et al Lung Transplantation Advanced Prediction Tool: Determining Recipient’s Outcome for a Certain Donor. Transplantation (2022) 106(10):2019–30. 10.1097/TP.0000000000004131

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31.

  Brahmbhatt JM Hee Wai T Goss CH Lease ED Merlo CA Kapnadak SG et al The Lung Allocation Score and Other Available Models Lack Predictive Accuracy for Post-Lung Transplant Survival. J Heart Lung Transpl (2022) 41(8):1063–74. 10.1016/j.healun.2022.05.008

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32.

  Sage AT Donahoe LL Shamandy AA Mousavi SH Chao BT Zhou X et al A machine-Learning Approach to Human Ex Vivo Lung Perfusion Predicts Transplantation Outcomes and Promotes Organ Utilization. Nat Commun (2023) 14(1):4810. 10.1038/s41467-023-40468-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 33.

  Pu L Leader JK Ali A Geng Z Wilson D . Predicting left/right Lung Volumes, Thoracic Cavity Volume, and Heart Volume from Subject Demographics to Improve Lung Transplant. J Med Imaging (2023) 10(05):051806. 10.1117/1.JMI.10.5.051806

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34.

  Dalton JE Lehr CJ Gunsalus PR Mourany L Valapour M . Refining the Lung Allocation Score Models Fails to Improve Discrimination Performance. CHEST (2023) 163(1):152–63. 10.1016/j.chest.2022.08.2217

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 35.

  Bartholmai BJ Raghunath S Karwoski RA Moua T Rajagopalan S Maldonado F et al Quantitative Computed Tomography Imaging of Interstitial Lung Diseases. J Thorac Imaging (2013) 28(5):298–307. 10.1097/RTI.0b013e3182a21969

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 36.

  Barbosa EM Simpson S Lee JC Tustison N Gee J Shou H . Multivariate Modeling Using Quantitative CT Metrics May Improve Accuracy of Diagnosis of Bronchiolitis Obliterans Syndrome After Lung Transplantation. Comput Biol Med (2017) 89:275–81. 10.1016/j.compbiomed.2017.08.016

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 37.

  Weigt SS Wang X Palchevskiy V Gregson AL Patel N DerHovanessian A et al Gene Expression Profiling of Bronchoalveolar Lavage Cells Preceding a Clinical Diagnosis of Chronic Lung Allograft Dysfunction. Plos One (2017) 12(1):e0169894. 10.1371/journal.pone.0169894

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 38.

  Barbosa EJM Lanclus M Vos W Van Holsbeke C De Backer W De Backer J et al Machine Learning Algorithms Utilizing Quantitative CT Features May Predict Eventual Onset of Bronchiolitis Obliterans Syndrome After Lung Transplantation. Acad Radiol (2018) 25(9):1201–12. 10.1016/j.acra.2018.01.013

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39.

  Halloran KM Parkes MD Chang J Timofte IL Snell GI Westall GP et al Molecular Assessment of Rejection and Injury in Lung Transplant Biopsies. J Heart Lung Transpl (2019) 38(5):504–13. 10.1016/j.healun.2019.01.1317

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40.

  Cantu E Yan M Suzuki Y Buckley T Galati V Majeti N et al Preprocurement in Situ Donor Lung Tissue Gene Expression Classifies Primary Graft Dysfunction Risk. Am J Respir Crit Care Med (2020) 202(7):1046–8. 10.1164/rccm.201912-2436LE

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 41.

  Halloran K Parkes MD Timofte IL Snell GI Westall GP Hachem R et al Molecular Phenotyping of Rejection-Related Changes in Mucosal Biopsies from Lung Transplants. Am J Transpl (2020) 20(4):954–66. 10.1111/ajt.15685

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 42.

  Halloran K Parkes MD Timofte I Snell G Westall G Havlin J et al Molecular T-Cell‒Mediated Rejection in Transbronchial and Mucosal Lung Transplant Biopsies Is Associated with Future Risk of Graft Loss. J Heart Lung Transpl (2020) 39(12):1327–37. 10.1016/j.healun.2020.08.013

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 43.

  Dugger DT Fung M Hays SR Singer JP Kleinhenz ME Leard LE et al Chronic Lung Allograft Dysfunction Small Airways Reveal a Lymphocytic Inflammation Gene Signature. Am J Transpl (2021) 21(1):362–71. 10.1111/ajt.16293

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44.

  Berra G Farkona S Mohammed-Ali Z Kotlyar M Levy L Clotet-Freixas S et al Association Between the Renin–Angiotensin System and Chronic Lung Allograft Dysfunction. Eur Respir J (2021) 58(4):2002975. 10.1183/13993003.02975-2020

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45.

  McInnis MC Ma J Karur GR Houbois C Levy L Havlin J et al Chronic Lung Allograft Dysfunction Phenotype and Prognosis by Machine Learning CT Analysis. Eur Respir J (2022) 60(1):2101652. 10.1183/13993003.01652-2021

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 46.

  Tran-Dinh A Laurent Q Even G Tanaka S Lortat-Jacob B Castier Y et al Personalized Risk Predictor for Acute Cellular Rejection in Lung Transplant Using Soluble CD31. Sci Rep (2022) 12(1):17628. 10.1038/s41598-022-21070-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47.

  Zhang YH Li ZD Zeng T Chen L Huang T Cai YD . Screening Gene Signatures for Clinical Response Subtypes of Lung Transplantation. Mol Genet Genomics (2022) 297(5):1301–13. 10.1007/s00438-022-01918-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 48.

  Su J xi LC yue LH et al The Airway Microbiota Signatures of Infection and Rejection in Lung Transplant Recipients. Microbiol Spectr (2022) 10 (2):e00344–21. 10.1128/spectrum.00344-21

  • CrossRef
  • Google Scholar

• 49.

  Watzenboeck ML Gorki AD Quattrone F Gawish R Schwarz S Lambers C et al Multi-Omics Profiling Predicts Allograft Function After Lung Transplantation. Eur Respir J (2022) 59(2):2003292. 10.1183/13993003.03292-2020

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 50.

  Stefanuto PH Romano R Rees CA Nasir M Thakuria L Simon A et al Volatile Organic Compound Profiling to Explore Primary Graft Dysfunction After Lung Transplantation. Sci Rep (2022) 12(1):2053. 10.1038/s41598-022-05994-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51.

  Qin J Xiao X Li S Wen N Qin K Li H et al Identification of Cuproptosis-Related Biomarkers and Analysis of Immune Infiltration in Allograft Lung Ischemia-Reperfusion Injury. Front Mol Biosci (2023) 10:1269478. 10.3389/fmolb.2023.1269478

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 52.

  Wijbenga N Hoek RAS Mathot BJ Seghers L Moor CC Aerts JGJV et al Diagnostic Performance of Electronic Nose Technology in Chronic Lung Allograft Dysfunction. J Heart Lung Transpl (2023) 42(2):236–45. 10.1016/j.healun.2022.09.009

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 53.

  Ram S Verleden SE Kumar M Bell AJ Pal R Ordies S et al Computed Tomography–Based Machine Learning for Donor Lung Screening Before Transplantation. J Heart Lung Transpl (2024) 43(3):394–402. 10.1016/j.healun.2023.09.018

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 54.

  Chao BT McInnis MC Sage AT Yeung JC Cypel M Liu M et al A Radiographic Score for Human Donor Lungs on Ex Vivo Lung Perfusion Predicts Transplant Outcomes. J Heart Lung Transpl (2024) 43(5):797–805. 10.1016/j.healun.2024.01.004

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 55.

  Gouiaa F Vomo-Donfack KL Tran-Dinh A Morilla I . Novel Dimensionality Reduction Method, Taelcore, Enhances Lung Transplantation Risk Prediction. Comput Biol Med (2024) 169:107969. 10.1016/j.compbiomed.2024.107969

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 56.

  Gao J Zhang Z Yu J Zhang N Fu Y Jiang X et al Identification of Neutrophil Extracellular Trap-Related Gene Expression Signatures in Ischemia Reperfusion Injury During Lung Transplantation: A Transcriptome Analysis and Clinical Validation. J Inflamm Res (2024) 17:981–1001. 10.2147/JIR.S444774

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 57.

  Chen Y Li E Yang Q Chang Z Yu B Lu J et al Predicting Time to First Rejection Episode in Lung Transplant Patients Using a Comprehensive Multi-Indicator Model. J Inflamm Res (2025) 18:477–91. 10.2147/JIR.S495515

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 58.

  Choshi H Miyoshi K Tanioka M Arai H Tanaka S Shien K et al Long Short-Term Memory Algorithm for Personalized Tacrolimus Dosing: A Simple and Effective Time Series Forecasting Approach Post-Lung Transplantation. J Heart Lung Transpl (2025) 44(3):351–61. 10.1016/j.healun.2024.10.026

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 59.

  Ahmad F Mat-Isa NA Hussain Z Boudville R Osman MK . Genetic Algorithm-Artificial Neural Network (GA-ANN) Hybrid Intelligence for Cancer Diagnosis. In: 2010 2nd International Conference on Computational Intelligence, Communication Systems and Networks (2010). p. 78–83. 10.1109/CICSyN.2010.46

  • CrossRef
  • Google Scholar

• 60.

  Štrumbelj E Kononenko I . Explaining Prediction Models and Individual Predictions with Feature Contributions. Knowl Inf Syst (2014) 41(3):647–65. 10.1007/s10115-013-0679-x

  • CrossRef
  • Google Scholar

• 61.

  Van Slambrouck J Van Raemdonck D Vos R Vanluyten C Vanstapel A Prisciandaro E et al A Focused Review on Primary Graft Dysfunction After Clinical Lung Transplantation: A Multilevel Syndrome. Cells (2022) 11(4):745. 10.3390/cells11040745

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 62.

  Hakkal S Lahcen AA . XGBoost to Enhance Learner Performance Prediction. Comput Educ Artif Intell (2024) 7:100254. 10.1016/j.caeai.2024.100254

  • CrossRef
  • Google Scholar

• 63.

  Kursa MB Jankowski A Rudnicki WR . Boruta – A System for Feature Selection. Fundam Informaticae (2010) 101(4):271–85. 10.3233/FI-2010-288

  • CrossRef
  • Google Scholar

• 64.

  Pierson RN Barr ML McCullough KP Egan T Garrity E Jessup M et al Thoracic Organ Transplantation. Am J Transpl (2004) 4:93–105. 10.1111/j.1600-6135.2004.00401.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 65.

  Yoon J Zame WR Banerjee A Cadeiras M Alaa AM Schaar M . Van Der. Personalized Survival Predictions via Trees of Predictors: An Application to Cardiac Transplantation. PLOS ONE (2018) 13(3):e0194985. 10.1371/journal.pone.0194985

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 66.

  Kampaktsis PN Moustakidis S Tzani A Doulamis IP Drosou A Tzoumas A et al State-Of-The-Art Machine Learning Improves Predictive Accuracy of 1-Year Survival After Heart Transplantation. ESC Heart Fail (2021) 8(4):3433–6. 10.1002/ehf2.13425

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 67.

  Ayers B Sandholm T Gosev I Prasad S Kilic A . Using Machine Learning to Improve Survival Prediction After Heart Transplantation. J Card Surg (2021) 36(11):4113–20. 10.1111/jocs.15917

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 68.

  Hsich EM Thuita L McNamara DM Rogers JG Valapour M Goldberg LR et al Variables of Importance in the Scientific Registry of Transplant Recipients Database Predictive of Heart Transplant Waitlist Mortality. Am J Transpl (2019) 19(7):2067–76. 10.1111/ajt.15265

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 69.

  Seraphin TP Luedde M Roderburg C van Treeck M Scheider P Buelow RD et al Prediction of Heart Transplant Rejection from Routine Pathology Slides with Self-Supervised Deep Learning. Eur Heart J - Digit Health (2023) 4(3):265–74. 10.1093/ehjdh/ztad016

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 70.

  Esteban C Staeck O Yang Y Tresp V . Predicting Clinical Events by Combining Static and Dynamic Information Using Recurrent Neural Networks. Arxiv Preprint Posted Online (2016) 93–101. 10.48550/arXiv.1602.02685

  • CrossRef
  • Google Scholar

• 71.

  Marsh JN Matlock MK Kudose S Liu TC Stappenbeck TS Gaut JP et al Deep Learning Global Glomerulosclerosis in Transplant Kidney Frozen Sections. IEEE Trans Med Imaging (2018) 37(12):2718–28. 10.1109/TMI.2018.2851150

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 72.

  Yoo D Divard G Raynaud M Cohen A Mone TD Rosenthal JT et al A Machine Learning-Driven Virtual Biopsy System for Kidney Transplant Patients. Nat Commun (2024) 15(1):554. 10.1038/s41467-023-44595-z

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 73.

  Bertsimas D Kung J Trichakis N Wang Y Hirose R Vagefi PA . Development and Validation of an Optimized Prediction of Mortality for Candidates Awaiting Liver Transplantation. Am J Transpl Off J Am Soc Transpl Am Soc Transpl Surg (2019) 19(4):1109–18. 10.1111/ajt.15172

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 74.

  Shao W Ding H Wang Y Shi Z Zhang H Meng F et al Key Genes and Immune Pathways in T-Cell Mediated Rejection Post-Liver Transplantation Identified via Integrated RNA-Seq and Machine Learning. Sci Rep (2024) 14(1):24315. 10.1038/s41598-024-74874-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar