Optimizing supernova classification with interpretable machine learning models

ObjectiveTo develop a computed tomography (CT) radiomics-based interpretable machine learning (ML) model to preoperatively predict human epidermal growth factor receptor 2 (HER2) status in bladder cancer (BCa) with multicenter validation.MethodsIn this retrospective study, 207 patients with pathologically confirmed BCa were enrolled and divided into the training set (n = 154) and test set (n = 53). Least absolute shrinkage and selection operator (LASSO) regression was used to identify the most discriminative features in the training set. Five radiomics-based ML models, namely logistic regression (LR), support vector machine (SVM), k-nearest neighbors (KNN), eXtreme Gradient Boosting (XGBoost) and random forest (RF), were developed. The predictive performance of established ML models was evaluated by the area under the receiver operating characteristic curve (AUC). The Shapley additive explanation (SHAP) was used to analyze the interpretability of ML models.ResultsA total of 1218 radiomics features were extracted from the nephrographic phase CT images, and 11 features were filtered for constructing ML models. In the test set, the AUCs of LR, SVM, KNN, XGBoost, and RF were 0.803, 0.709, 0.679, 0.794, and 0.815, with corresponding accuracies of 71.7%, 69.8%, 60.4%, 75.5%, and 75.5%, respectively. RF was identified as the optimal classifier. SHAP analysis showed that texture features (gray level size zone matrix and gray level co-occurrence matrix) were significant predictors of HER2 status.ConclusionsThe radiomics-based interpretable ML model provides a noninvasive tool to predict the HER2 status of BCa with satisfactory discriminatory performance.Critical relevance statementAn interpretable radiomics-based machine learning model can preoperatively predict HER2 status in bladder cancer, potentially aiding in the clinical decision-making process.Key PointsThe CT radiomics model could identify HER2 status in bladder cancer.The random forest model showed a more robust and accurate performance.The model demonstrated favorable interpretability through SHAP method.Graphical

ObjectivePostoperative complications remain a major concern in laryngeal cancer surgery, often requiring invasive interventions or intensive care. This study aimed to develop and validate an interpretable machine learning (ML) model to preoperatively predict Clavien-Dindo Grade ≥ III complications and support risk-informed perioperative decision-making.MethodsWe conducted a retrospective study using a temporally split cohort of laryngeal cancer patients. Postoperative complications were graded using the Clavien-Dindo (CD) classification. Eight ML algorithms were trained and evaluated using receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA). Model interpretability was assessed using SHapley Additive exPlanations (SHAP). A web-based calculator was deployed for clinical use.ResultThe random forest (RF) model achieved the best performance, with an area under the curve (AUC) of 0.935 in the training set and 0.842 in the test set. The model demonstrated robust sensitivity and specificity for both surgical and medical complications. Calibration curves indicated strong agreement between predicted and actual outcomes. SHAP analysis identified eight key predictors—such as vocal cord mobility, tumor subsite, and nutritional status—that contributed most to risk estimation. A user-friendly web calculator was developed and is accessible at: https://qilushiny.shinyapps.io/qilupredicate/.ConclusionWe developed a clinically interpretable ML model that accurately predicts major postoperative complications in patients undergoing laryngeal cancer surgery. This tool provides individualized risk assessments that can guide surgical planning, optimize perioperative strategies, and enhance shared decision-making. Prospective multicenter validation is needed to confirm its utility in routine practice.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12893-025-03213-z.

Interpretable vs. noninterpretable machine learning models for data-driven hydro-climatological process modeling

Studies reported that playing video games with harmful content can lead to adverse effects on players. Therefore, understanding the harmful content can help reduce these adverse effects. This study is the first to examine the potential of interpretable machine learning (ML) models for explaining the harmful content in video games that may potentially cause adverse effects on players based on game rating predictions. First, the study presents a performance analysis of the supervised ML models for game rating predictions. Secondly, using an interpretability analysis, this study explains the potentially harmful content. The results show that the ensemble Random Forest model robustly predicted game ratings. Then, the interpretable ML model successfully exposed and explained several harmful contents, including Blood, Fantasy Violence, Strong Language, and Blood and Gore. This revealed that the depiction of blood, the depiction of the mutilation of body parts, violent actions of human or non-human characters, and the frequent use of profanity might potentially be associated with adverse effects on players. The findings suggest the strength of interpretable ML models in explaining harmful content. The knowledge gained can be used to develop effective regulations for controlling identified video game content and potential adverse effects.

Artificial Intelligence (AI) is widely used in decision making procedures in myriads of real-world applications across important practical areas such as finance, healthcare, education, and safety critical systems. Due to its ubiquitous use in safety and privacy critical domains, it is often vital to understand the reasoning behind the AI decisions, which motivates the need for explainable AI (XAI). One of the major approaches to XAI is represented by computing so-called interpretable machine learning (ML) models, such as decision trees (DT), decision lists (DL) and decision sets (DS). These models build on the use of if-then rules and are thus deemed to be easily understandable by humans. A number of approaches have been proposed in the recent past to devising all kinds of interpretable ML models, the most prominent of which involve encoding the problem into a logic formalism, which is then tackled by invoking a reasoning or discrete optimization procedure. This paper overviews the recent advances of the reasoning and constraints based approaches to learning interpretable ML models and discusses their advantages and limitations.

BackgroundAnti-programmed cell death protein 1 (PD-1)/programmed cell death ligand 1 (PD-L1) immunotherapy has revolutionized cancer treatment. However, it can cause immune-related adverse events, including acute kidney injury (AKI). Such adverse events can interrupt treatment, affecting patient outcomes. Early prediction of AKI is essential for improved prognosis and personalized therapeutic strategies. Previous research has been constrained by significant limitations, underscoring the necessity for AKI risk prediction models for patients treated with PD-1/PD-L1 inhibitors. This study aimed to develop and validate an interpretable machine learning (ML) model for early AKI prediction in patients undergoing PD-1/PD-L1 inhibitor therapy using a retrospective cohort design.MethodsThis study collected data from patients treated with PD-1/PD-L1 inhibitors at Zhejiang Provincial People’s Hospital between January 2018 and January 2024. Nine ML models were evaluated. SHapley Additive exPlanations (SHAP) were employed to rank feature importance and interpret the final model. Additionally, a web-based calculator based on the model was developed.ResultsAmong the nine ML models evaluated, the Grandient Boosting Machine (GBM) model achieved the best predictive performance. In the validation set, the GBM model achieved an AUC of 0.850 (95%CI: 0.830–0.870). In the test set, the AUC was 0.795(95% CI: 0.747–0.844), demonstrating accurate AKI risk prediction. Calibration curves demonstrated a strong concordance between predicted and observed risk probabilities. An interpretable final GBM model with 13 features was developed after feature reduction based on feature importance ranking. A web-based calculator accessible at https://predicatingaki.shinyapps.io/PDmodel/ has been developed to assist clinicians in AKI risk assessment.ConclusionThis study developed and validated an interpretable ML model using a large dataset to predict AKI risk in patients receiving PD-1/PD-L1 inhibitor therapy. This model can assist clinicians in the early identification of high-risk patients, facilitating personalized treatment plans.Trial registrationThe study was conducted following the Declaration of Helsinki and was approved by the Ethics Committee of Zhejiang Provincial People’s Hospital (Approval No. KT2024116) in 3 Jan. 2025. As it was a retrospective study with anonymized data, informed consent was waived.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12911-025-03142-0.

BackgroundTo address the heterogeneity in treatment responses and the lack of robust prognostic tools for unresectable esophageal squamous cell carcinoma (ESCC) patients undergoing immunochemotherapy, this study aimed to develop and validate an interpretable machine learning (ML) model for survival prediction and risk stratification.MethodsA retrospective cohort of 323 unresectable ESCC patients treated with immunochemotherapy (2019–2025) was analyzed. Using the XGBoost algorithm, we integrated baseline clinical features (age, tumor location, TNM stage) and laboratory parameters (albumin, globulin, blood glucose) to construct a prognostic model. SHapley Additive exPlanations (SHAP) values were employed to quantify feature contributions, and external validation (n=48) was performed to assess generalizability. SHAP (SHapley Additive exPlanations) is a game theory-based framework that enables model interpretability by quantifying the contribution of each feature to predictions. The primary endpoint was overall survival (OS).ResultsThe model achieved AUC values of 0.794 (internal test) and 0.689 (external test), with calibration curves demonstrating strong concordance between predicted and observed survival rates. Key prognostic factors included tumor response, age, hypoalbuminemia, hyperglobulinemia and hyperglycemia. Risk stratification using a nomogram-derived cutoff (total score ≥50) revealed significantly inferior 2-year OS in high-risk versus low-risk patients (21.3% vs 58.6%, P<0.001).ConclusionThis interpretable ML model effectively predicts survival outcomes in unresectable ESCC patients receiving immunochemotherapy, offering a data-driven tool for personalized therapeutic decision-making. Multicenter prospective trials are warranted to validate its clinical utility.

ObjectivesThis study aimed to develop and validate an interpretable machine learning (ML) model based on structured preoperative CT features for non-invasive prediction of pancreatic neuroendocrine Tumors (PNETs) aggressiveness.MethodsThis retrospective study included 112 patients with PNETs who underwent contrast-enhanced abdominal CT. Patients were randomly assigned to training and validation cohorts. Clinical data and CT features were analysed using the Least Absolute Shrinkage and Selection Operator method and multivariate logistic regression to identify independent risk factors. Multiple ML models were evaluated to determine the optimal classifier. Model performance was assessed using receiver operating characteristic and calibration curves, and decision curve analysis. Shapley Additive Explanations (SHAP) quantified feature importance for interpretable risk prediction.ResultsA total of 112 patients were evaluated, including 80(mean age± standard deviation, 47 ± 13 years; 36 males)) in the training set and 32 (48 ± 15 years; 12 males) in the validation set. Tumour shape, necrotic changes, arterial relative enhancement ratio, and enhancement pattern independently predicted PNETs aggressiveness. The logistic regression model demonstrated excellent discrimination, achieving an area under the curve of 0.952 (95% CI: 0.952 (0.909–0.994) in the training cohort and 0.972 (95% CI 0.927–1.000) in the validation cohort. SHAP summary and force plots facilitated global and local model interpretation.ConclusionThe Interpretable ML model based on CT features could serve as a preoperative, noninvasive, and precise evaluation tool to differentiate aggressive and non-aggressive PNETs, facilitating personalized clinical management and potentially improving patient outcomes.

Interpretable Machine Learning Models for Severe Esophagitis Prediction in LA-NSCLC Patients Treated with Chemoradiation Therapy

BackgroundSplenomegaly serves as a crucial indicator for various diseases, particularly in hepatosplenomegaly and hematological disorders. Accurate assessment of splenomegaly is essential for improving diagnostic accuracy and treatment decisions, yet individualized diagnosis necessitates a standard reference for splenic volume. This study aimed to develop an interpretable machine learning (ML) model to evaluate standard splenic volume (SSV), enhancing personalized clinical decision-making.MethodsWe conducted a retrospective analysis of 1,186 volunteers from a multicenter cohort and evaluated 11 ML algorithms. SHapley Additive exPlanations (SHAP) were employed for feature selection and interpretation. Model performance was rigorously evaluated through key metrics such as root mean squared error (RMSE), coefficient of determination (R2), and additional validation parameters, further validated through comparisons with prior published formulas. We also developed free, open-access web-based calculators for the predictive model.ResultsModel development and internal validation involved 511 eligible volunteers, with external validation from an additional 111 volunteers. The random forest (RF) model (ML_SSV) integrating features such as age, body weight (BW), body height, body mass index (BMI), body surface area (BSA), red blood cell count, platelet count, total bilirubin, fibrinogen, and D-dimer, demonstrated exceptional predictive accuracy. In external validation, the model achieved an RMSE of 22.6 mL (R2=0.80), with residual analysis confirming normally distributed errors (range: −58.32 to 67.01 mL; P=0.201). Notably, a simplified RF model (ML_SSVa) utilizing only four non-invasive parameters (age, BW, BMI, BSA) retained robust performance, with an RMSE of 36.0 mL (R2=0.70) in external validation. Furthermore, both models outperformed all existing formulas in cross-validation analyses. The models were deployed as open-access calculators at https://mlssv.vip.cpolar.cn (ML_SSV) and https://mlssva.vip.cpolar.cn (ML_SSVa), enabling real-time estimation with SHAP-based interpretability.ConclusionsThis study establishes a novel interpretable ML model rigorously validated through statistical and clinical benchmarks. These models enable the assessment of SSV, providing a reference baseline for the individualized diagnosis of splenomegaly to enhance diagnostic accuracy and support data-driven clinical decision-making.

Gallbladder polyps (GBPs) are increasingly prevalent, with the majority being benign; however, neoplastic polyps carry a risk of malignant transformation, highlighting the importance of accurate differentiation. This study aimed to develop and validate interpretable machine learning (ML) models to accurately predict neoplastic GBPs in a retrospective cohort, identifying key features and providing model explanations using the Shapley additive explanations (SHAP) method. A total of 924 patients with GBPs who underwent cholecystectomy between January 2013 and December 2023 at Qilu Hospital of Shandong University were included. The patient characteristics, laboratory results, preoperative ultrasound findings, and postoperative pathological results were collected. The dataset was randomly split, with 80% used for model training and the remaining 20% used for model testing. This study employed nine ML algorithms to construct predictive models. Subsequently, model performance was evaluated and compared using several metrics, including the area under the receiver operating characteristic curve (AUC). Feature importance was ranked, and model interpretability was enhanced by the SHAP method. K-nearest neighbors, C5.0 decision tree algorithm, and gradient boosting machine models showed the highest performance, with the highest predictive efficacy for neoplastic polyps. The SHAP method revealed the top five predictors of neoplastic polyps according to the importance ranking. The polyp size was recognized as the most important predictor variable, indicating that lesions ≥ 18 mm should prompt heightened clinical surveillance and timely intervention. Our interpretable ML models accurately predict neoplastic polyps in GBP patients, providing guidance for treatment planning and resource allocation. The model's transparency fosters trust and understanding, empowering physicians to confidently use its predictions for improved patient care.

BackgroundPostoperative anastomotic leakage (AL) is a severe complication following esophageal cancer surgery, that often leads to a poor prognosis. This study aims to develop an interpretable machine learning (ML) model to predict AL occurrence and identify associated risk factors.MethodsA retrospective case‒control study analyzed clinical and laboratory data from esophageal cancer patients obtained via a case management system. Nine machine learning (ML) models were compared to identify the best-performing model and its optimal feature set. The selected LightGBM-based model underwent internal cross-validation and external validation. Performance was evaluated via metrics such as ROC, DCA, and PR curves. To enhance interpretability, the SHapley Additive exPlanations (SHAP) method was applied for feature analysis.ResultsData from a total of 406 esophageal cancer patients were collected, and the LightGBM-based model showed the best performance. The model included the following features: lesion length, McKeown surgery, gastrointestinal decompression drainage (GID) volume on postoperative day 1, and prealbumin difference. SHAP dependence plots were created for each variable to understand their impact on the outcome. The model achieved an AUC of 0.956 (95% CI: 0.934–0.978).ConclusionThis study successfully developed an interpretable ML model based on the LightGBM to predict postoperative AL in patients with esophageal cancer.

Early neurologic deterioration (END) is a common and feared complication for acute ischemic stroke (AIS) patients treated with mechanical thrombectomy (MT). This study aimed to develop an interpretable machine learning (ML) model for individualized prediction to predict END in AIS patients treated with MT. The retrospective cohort of AIS patients who underwent MT was from two hospitals. ML methods applied include logistic regression (LR), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost). The area under the receiver operating characteristic curve (AUC) was the main evaluation metric used. We also used Shapley Additive Explanation (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to interpret the result of the prediction model. A total of 985 patients were enrolled in this study, and the development of END was noted in 157 patients (15.9%). Among the used models, XGBoost had the highest prediction power (AUC = 0.826, 95% CI 0.781-0.871). The Delong test and calibration curve indicated that XGBoost significantly surpassed those of the other models in prediction. In addition, the AUC in the validating set was 0.846, which showed a good performance of the XGBoost. The SHAP method revealed that blood glucose was the most important predictor variable. The constructed interpretable ML model can be used to predict the risk probability of END after MT in AIS patients. It may help clinical decision making in the perioperative period of AIS patients treated with MT.

BackgroundAcute exacerbations of chronic obstructive pulmonary disease (AECOPD) are associated with accelerated lung function decline and increased mortality. However, early and accurate diagnosis remains clinically challenging due to nonspecific symptoms and limitations of existing diagnostic tools. This study aimed to develop an interpretable machine learning (ML) model integrating multimodal ultrasound indicators to facilitate real-time bedside diagnosis of AECOPD.MethodsIn this prospective, single-center study, 316 patients with COPD underwent standardized lung, diaphragmatic, and quadriceps ultrasound examinations upon hospital admission. Four ML algorithms were developed using a 7:3 training-to-test data split. Model performance was assessed by area under the receiver operating characteristic curve (AUC), and interpretability was enhanced using SHapley Additive exPlanations (SHAP).ResultsThe support vector machine (SVM) model achieved the best diagnostic performance, with an AUC of 0.9321 in the training set and 0.9302 in the test set. The final model incorporated six routinely obtainable variables, five of which were ultrasound derived. SHAP analysis identified elevated lung ultrasound scores, diaphragmatic dysfunction, and quadriceps atrophy as the most influential predictors.ConclusionsThis non-invasive and interpretable ML model, based on bedside ultrasound features, offers a clinically feasible tool for real-time AECOPD diagnosis. Further multicenter validation is warranted to confirm generalizability and explore integration with additional biomarkers or imaging modalities.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12931-025-03411-6.

BackgroundAlthough neoadjuvant immunochemotherapy (nICT) has revolutionized the management of locally advanced esophageal squamous cell carcinoma (ESCC), the inability to accurately predict pathological complete response (pCR) remains a major barrier to treatment personalization. We aimed to develop and validate an interpretable machine learning (ML) model using pretreatment multimodal features to predict pCR prior to nICT initiation.MethodsIn this retrospective study, 114 ESCC patients receiving nICT were randomly allocated into training (n=81) and validation (n=33) cohorts (7:3 ratio). Predictors of pCR were identified from pretreatment clinical variables, endoscopic ultrasonography, and hematological biomarkers via least absolute shrinkage and selection operator (LASSO) regression. Eight machine learning algorithms were implemented to construct prediction models. Model performance was assessed by area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Shapley Additive Explanations (SHAP) provided feature importance and model interpretability.ResultsFollowing feature selection, 17 variables were incorporated into model construction. The Random Forest (RF) model demonstrated perfect discrimination in the training cohort (AUC = 1.000, sensitivity = 1.000, specificity = 1.000, PPV = 1.000, NPV = 1.000), while maintaining robust predictive ability in the independent validation cohort (AUC = 0.913, sensitivity = 0.733, specificity = 0.889, PPV = 0.846, NPV = 0.800). Decision curve analysis (DCA) confirmed favorable clinical utility. SHAP analysis identified alcohol consumption, circumferential involvement ≥50%, elevated neutrophil-to-lymphocyte ratio (NLR), C-reactive protein (CRP), and alanine aminotransferase (ALT) as the key contributors to pCR prediction.ConclusionsWe established a clinically applicable, interpretable ML model that accurately predicts pCR to nICT in ESCC by integrating multimodal pretreatment data. This tool may optimize patient selection for nICT and advance precision therapy paradigms.