Fig. 7.

Comparative performance analysis of hybrid CNN-based models for dental condition classification. a Macro-averaged performance metrics (accuracy, precision, recall, and F1-score) for three hybrid models across 5-fold cross-validation, with error bars indicating standard deviation. b Confusion matrices for each hybrid model showing the classification results across four dental conditions (fillings, cavity, implant, and impacted tooth), with numerical values representing the count of samples in each prediction category. c t-SNE visualization of CNN feature space structure across model architectures. Data points represent individual dental X-ray samples projected into 2D space, color-coded by condition (blue: fillings, purple: cavity, orange: implant, red: impacted tooth)