From prediction to clinical decision-making: Explainable machine learning in acute suppurative cholecystitis

Table 1 Summary of recent studies on machine learning and radiomics in acute biliary disease and surgical risk prediction

Ref.	Year	Population	Modality	Model	Key findings	Translational value	Limitations	Gap identified
Demircioğlu[31]	2025	Radiomics methods	Imaging	ML frameworks	Reproducibility critical	High	Pipeline variability	No clinical translation
Ma et al[32]	2024	Acute biliary disease	CT radiomics	XGBoost + SHAP	Interpretable early severity prediction	High	Single-center	Lack of validation in cholecystitis-specific cohorts
Kenig et al[38]	2024	Surgical AI	Mixed	ML models	Improved preoperative prediction	Moderate	Limited validation	Lack of explainability integration
Cè et al[37]	2024	Radiomics workflow	CT/MRI	ML pipelines	Importance of pipeline standardization	High	Variability	No direct clinical application
Teng et al[36]	2024	Radiomics	Imaging	ML models	Focus on clinical integration challenges	High	Limited deployment	Lack of real-time decision tools
Sadeghi et al[40]	2024	Healthcare AI	Mixed	XAI models	Explainability improves trust	High	Misinterpretation risk	Limited disease-specific application
Borys et al[41]	2023	Imaging AI	Imaging	SHAP/XAI	Enables patient-level interpretation	High	Needs validation	Not integrated in surgical workflows
Erickson et al[9]	2019	Imaging review	Multimodal	ML overview	Highlights ML potential	Low-Moderate	Conceptual only	No clinical implementation
Bi et al[10]	2019	Imaging datasets	CT radiomics	ML classifiers	Quantifies imaging heterogeneity	Moderate	Feature instability	No integration with clinical variables
Hosny et al[8]	2018	Imaging datasets	CT radiomics	Deep learning	Demonstrates scalability	Moderate	Generalizability issues	Not tailored to acute surgical decision-making

SHAP: SHapley Additive exPlanations; CT: Computed tomography; ML: Machine learning; MRI: Magnetic resonance imaging; XAI: Explainable Artificial Intelligence; AI: Artificial intelligence; XGBoost: Extreme Gradient Boosting.

Table 2 Clinical decision framework integrating explainable machine learning in acute suppurative cholecystitis

Risk level (ML output)	Explainability insight (SHAP)	Clinical interpretation	Recommended action	Multidisciplinary involvement
High risk	Strong contribution from radiomic heterogeneity-inflammatory markers	Likely suppurative cholecystitis with high operative complexity	Early laparoscopic cholecystectomy; prioritize OR scheduling; consider senior surgical team	Surgeon, anesthesiologist, radiologist, ICU team
Intermediate risk	Mixed feature contribution; moderate radiomic signal with variable clinical parameters	Uncertain progression; potential for deterioration	Close monitoring; repeat imaging; optimize comorbidities; reassess within 12-24 hours	Surgeon, radiologist, ward team
Low risk	Minimal radiomic heterogeneity; low inflammatory contribution	Likely uncomplicated disease	Conservative management; antibiotics; delayed surgery if indicated	Surgeon, ward team
High risk with discordant clinical findings	High radiomic signal but mild clinical presentation	Early-stage suppuration or subclinical progression	Escalate monitoring; consider early intervention despite mild symptoms	Surgeon, radiologist
Low risk with severe clinical presentation	Low radiomic signal but strong clinical/Lab abnormalities	Possible alternative diagnosis or false-negative model output	Re-evaluate diagnosis; additional imaging; do not rely solely on model	Full team

SHAP: SHapley Additive exPlanations; ML: Machine learning; ICU: Intensive care unit; OR: Operating room.