Bacterial vaginosis is a significant public health concern affecting the reproductive health of sexually active women, underscoring the need for tools that enhance diagnostic accuracy. This study introduces a model-agnostic framework that provides explainability for black-box models used to diagnose bacterial vaginosis. The framework combines the predictive power of opaque models, such as artificial neural networks and random forests, with the transparency and interpretability of decision trees. By leveraging both the training data and the predictions from black-box models, the framework constructs decision trees without requiring access to the internal mechanics of the black-box models. Experimental results demonstrate that this approach generates decision trees more accurately than those created from the original training dataset.