# ACO-optimized MobileNetV2-ShuffleNet hybrid model for automated dental caries classification

## TL;DR

This paper proposes an ACO-enhanced hybrid model combining MobileNetV2 and ShuffleNet for accurate and automated dental caries classification from X-ray images.

## Contribution

The novel contribution is the ACO-optimized hybrid architecture that improves classification accuracy in dental caries diagnosis.

## Key findings

- The hybrid model achieved 92.67% accuracy, outperforming standalone MobileNetV2 and ShuffleNet.
- ACO improved the model's performance by enabling efficient global search and parameter tuning.
- Preprocessing techniques like clustering and Sobel-Feldman edge detection helped balance data and highlight critical features.

## Abstract

Dental infections may result in severe health conditions when not diagnosed and responded to immediately. However, it is a difficult process that can take time and expertise to diagnose oral infections based on X-ray images. In this paper, a new method of dental caries classification based on the panoramic radiographic images is proposed, which is aimed at overcoming the class imbalance and weak anatomical differences. During the preprocessing stage, the clustering technique was used to form similar grouped data to balance the distribution of data, and the Sobel-Feldman edge technique was applied to emphasize critical features. MobileNetV2 and ShuffleNet models were also trained on the preprocessed set of data separately, but the classification ability was poor. A hybrid architecture was designed based on the combination of the strengths of the two models, so the level of precision increased. In a further effort to improve the performance of the model, Ant Colony Optimization (ACO) algorithm was incorporated to the hybrid framework. Addition of ACO made the classification highly accurate since it could perform an efficient global search and parameter tuning. The suggested ACO-enhanced hybrid approach showed better results with 92.67% accuracy than standalone networks which implies that the proposed model can be used on reliable and automated dental diagnosis.