Abstract
BACKGROUND: Recent advancements in artificial intelligence have led to increased adoption of machine learning in disease identification, particularly for challenging diagnoses like necrotizing fasciitis and Vibrio vulnificus infections. This shift is driven by the technology's efficiency, objectivity, and accuracy, offering potential solutions to longstanding diagnostic hurdles in clinical practice. METHODS: This investigation incorporated 180 inpatients suffering from soft tissue infections. The participants were categorized into groups: cellulitis, non-Vibrio necrotizing fasciitis (NF), or V. Vulnificus NF. To predict the three relevant outcomes, we employed Light Gradient Boosting Machine (LightGBM) and 5-fold cross-validation methodologies for the development of a multi-class categorization model. Moreover, we applied the SHapley Additive exPlanations (SHAP) methodology to decipher the model's predictions. RESULTS: The multi-classification model possesses substantial predictive capacity, with a weighted-average AUC of 0.86, sensitivity of 87.2%, specificity of 74.5%, NPV of 81.6%, and PPV of 85.4%. The model's calibration was assessed using the Brier score, yielding a weighted mean of 0.084. This low value demonstrates a strong correlation between predicted probabilities and actual outcomes, indicating high predictive accuracy and reliability in the model's forecasts. CONCLUSIONS: We effectively developed a multiclassification model aimed at forecasting the occurrence of cellulitis, non-Vibrio NF, or V. Vulnificus NF in patients suffering from soft tissue infection, and we further described the model's predictions using the SHAP algorithm.