Intelligent Medical Decision-Making Through Artificial Intelligence, Predictive Modeling, and Advanced Healthcare Analytics

AI and ML are becoming more common for supporting decisions in diagnostics, but the evaluation needs to be leakage safe, evidence needs to be interpretable, and there needs to be careful distinction between benchmark performance and claimable deployable clinical performance. The objective of this study was to create an AI based medical-intelligence framework for diagnosis of breast-cancer based on Wisconsin Breast Cancer (Diagnostic) dataset. Five hundred sixty-nine observations and 30 quantitative morphological predictors obtained from fine-needle aspirate images were analyzed. Logistic regression, radial-basis-function support vector machine (RBF SVM), random forest and histogram gradient boosting were compared by using a stratified 70/15/15 train-validation-locked-test design. The RBF SVM had the best cross-validated ROC-AUC of 0.9925 and after tuning, 0.9950. It correctly classified all 86 locked-test cases, with a threshold of 0.320 derived from validation that resulted in ROC-AUC and PR-AUC values of 1.000. The morphological separation was the largest for concave points, perimeter, radius, area and measures of concavity. The framework illustrates how predictive analytics and interpretation of models can be used to prioritize diagnostics, and highlights the importance of external validation for clinical use.

Hali tarjima qilinmagan