ResNet-GA: genetic algorithm-driven optimization of ResNet for automated dental caries detection in medical imaging

Dental caries is one of the most prevalent oral diseases worldwide and remains a major public health concern because delayed or inaccurate diagnosis can lead to irreversible tooth destruction, pain, infection, and more invasive treatment. Early and reliable detection from dental images is therefore essential for improving preventive care and supporting clinical decision-making. However, manual interpretation of dental radiographs is time-consuming, subjective, and prone to inter-observer variability. Although deep learning has shown strong potential for automated caries detection, its performance often depends on manually selected hyperparameters and may be affected by image variability, low contrast, artifacts, and class imbalance. To address these limitations, we propose ResNet-GA, a hybrid deep learning framework that combines deep convolutional feature learning with a genetic algorithm to automatically optimize selected convolutional configurations, particularly the number of filters and kernel sizes in three convolutional blocks. The proposed framework was evaluated on a publicly available dental caries dataset obtained from Kaggle, comprising caries and non-caries images. Images were resized, normalized, and divided into training and validation subsets, with class balancing applied to reduce bias toward the majority class. The genetic algorithm iteratively evolved candidate convolutional configurations using classification accuracy as the fitness criterion. The optimized model achieved high diagnostic performance, with a mean test accuracy of 87.89% ± 1.41% (macro precision 89.74% ± 0.95%, macro recall 80.38% ± 3.04%, macro F1-score 83.32% ± 2.58%) across five independent runs on an independent test set. These results indicate that evolutionary optimization of convolutional configurations can improve robustness and classification performance for automated dental caries detection. By reducing dependence on manual architectural tuning and improving diagnostic consistency, the proposed framework offers a promising computer-aided decision-support approach for dental image analysis.