Abstract
Accurate classification of human teeth is a critical component of modern dental diagnostics and treatment planning. This paper proposes an automated system for identifying the four primary categories of human teeth—incisors, canines, premolars, and molars—using CT scan imagery. The methodology utilizes Wavelet Entropy (WE) for feature extraction and a Multilayer Perceptron (MLP) for classification. To enhance performance, the MLP is optimized using the Biogeography-Based Optimization (BBO) algorithm. Experimental results demonstrate that the proposed WE-MLP-BBO method achieves high classification accuracies, ranging from 83.50% to 84.75% across the four tooth types, significantly outperforming traditional optimization techniques like Genetic Algorithms (GA) and Firefly Algorithms (FA).
Introduction
Teeth are essential structures in vertebrates, serving as tools for chewing and, in many species, as weapons for defense. In humans, teeth also play a vital role in facial aesthetics and speech pronunciation. Human teeth are generally categorized into four types based on their morphology and function: incisors for cutting, canines for puncturing and tearing, and premolars and molars for grinding food.
Automating the classification of these teeth from medical imaging can reduce the workload of dental professionals and improve diagnostic precision. While previous studies often focused on binary classification (e.g., distinguishing only between molars and premolars), this research addresses the more complex task of four-way classification.
Methodology
Data Acquisition
The study utilized a dataset of 160 CT scan images, with 40 samples for each of the four tooth categories. CT scans were selected for their high resolution and clinical relevance in dental procedures.
Feature Extraction and Classification
The proposed pipeline consists of three main stages:
- Wavelet Entropy (WE): This technique is used to extract descriptive features from the CT images. By combining wavelet transforms with information entropy, the system captures the structural complexity of each tooth type in a compact mathematical format.
- Multilayer Perceptron (MLP): A feed-forward neural network serves as the core classifier, mapping the extracted WE features to the corresponding tooth category.
- Biogeography-Based Optimization (BBO): Instead of standard backpropagation, the BBO algorithm is used to optimize the weights and biases of the MLP. Inspired by the migration patterns of biological species between habitats, BBO effectively explores the solution space to find optimal network parameters, avoiding the local optima that often plague other evolutionary algorithms.
Results and Discussion
The performance of the WE-MLP-BBO model was evaluated using 10-fold cross-validation. The system achieved the following accuracies:
- Incisors: 83.75 ± 2.95%
- Canines: 83.50 ± 5.16%
- Premolars: 84.00 ± 5.16%
- Molars: 84.75 ± 3.43%
Comparative analysis showed that the BBO-optimized model significantly outperformed other swarm intelligence methods. The average accuracy for BBO was approximately 84%, compared to 75% for the Firefly Algorithm (FA) and 72% for the Genetic Algorithm (GA). The success of BBO is attributed to its unique migration and mutation mechanisms, which provide a more robust search for the optimal neural network configuration.
Conclusion
This research demonstrates that combining Wavelet Entropy with a BBO-optimized Multilayer Perceptron provides a reliable method for the automated classification of human teeth. The high accuracy achieved across all four categories suggests strong potential for clinical application. Future work will focus on expanding the dataset size and incorporating advanced image preprocessing techniques to further enhance classification reliability.
Summary updated