Enterprise AI Analysis
ACO-optimized MobileNetV2-ShuffleNet hybrid model for automated dental caries classification
This paper presents an elaborate method for categorizing dental caries using panoramic radiographic images. First clustering-based method was used to resolve the issue of class imbalance in the dataset, and then edge detection was used with Sobel-Feldman to increase the significance of specific features through preprocessing. Although MobileNetV2 and ShuffleNet as individual deep learning models were used, they did not perform well when used alone. To make the best use of the characteristics of both architectures, a hybrid model was built, uniting the advantages of the architectures. This was done by further optimization using the ACO algorithm within the hybrid framework that produced quite high performance values. The ACO-strengthened hybrid model showed better classification accuracy, due to its strong global search strategy as well as optimal parameter settings, hence confirming its promise in automated dental diagnosis systems.The accuracy scores of the individual models were moderate, where MobileNetV2 and ShuffleNet scored 76% and 82% respectively. Conversely, the proposed hybrid model has shown a better accuracy rate of 83%, and by combining the model with ACO, the level of accuracy was further improved to 92% indicating the efficiency of the hybrid model.
Executive Impact
Our AI solutions deliver quantifiable results across key business dimensions.
0
Accuracy Improvement
0
Precision Achieved
0
Recall Rate
0
AUC Score
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Preprocessing
Model Architecture
Optimization
Results
Preprocessing
The preprocessing stage is crucial for enhancing the quality of input data and balancing its distribution. A clustering-based selection method was applied to balance the dataset by selecting 3069 non-caries images from an initial 9,931 to match the 3069 caries images. This ensures equal representation for binary classification. The Sobel-Feldman edge detection technique was then used to emphasize critical features like tooth edges and caries regions in X-ray images, aiding in segmentation and classification tasks. This systematic approach ensures that the hybrid model receives high-quality, relevant data, significantly improving its ability to learn and classify dental caries accurately.
Model Architecture
The core of the proposed solution lies in its hybrid deep learning architecture, combining MobileNetV2 and ShuffleNet. MobileNetV2, known for its depthwise separable convolutions and inverted residual blocks, efficiently extracts intricate patterns and dense features from X-ray images with low computational cost. ShuffleNet contributes with its pointwise group convolutions and channel shuffling, ensuring rich feature map channels and improved information flow, even in small networks. By concatenating the outputs of these two branches, the model leverages the strengths of both, creating a robust and efficient feature extractor. This fusion layer then feeds into fully connected layers for final classification, enhancing the overall representational capability.
Optimization
To further boost the model's performance, the Ant Colony Optimization (ACO) algorithm was integrated into the hybrid framework. ACO, a nature-inspired metaheuristic, excels at parameter tuning and global search. By simulating ants traversing a solution space, guided by pheromone trails and heuristic data, ACO efficiently optimizes hyperparameters like learning rate, momentum, and batch size. This iterative optimization process helps the hybrid model converge to an optimal solution more effectively than traditional methods, preventing local minima and improving overall accuracy. The ACO-enhanced model demonstrates superior classification capabilities due to this intelligent tuning of its underlying deep learning components.
Results
The proposed ACO-optimized MobileNetV2-ShuffleNet hybrid model significantly outperforms standalone networks. With an achieved accuracy of 92.67%, a precision of 93.67%, a recall of 91.53%, and an F1-score of 92.58%, the model demonstrates robust and reliable performance. The AUC score of 96.79% highlights its strong discriminative capabilities. Compared to MobileNetV2 (76.60% accuracy) and ShuffleNet (82.90% accuracy) as individual models, the hybrid approach with ACO shows a substantial improvement, validating its effectiveness for automated dental caries classification from panoramic radiographs. The model's stability was further confirmed by achieving over 86% accuracy across all folds in 5-fold cross-validation.
Enterprise Process Flow
Clustering for Data Balancing
→
Sobel-Feldman Edge Detection
→
Hybrid MobileNetV2-ShuffleNet Feature Extraction
→
Ant Colony Optimization (ACO) for Parameter Tuning
→
Dental Caries Classification
Peak Classification Accuracy
The ACO-optimized hybrid model demonstrated a significant leap in classification accuracy, reaching a high point that ensures reliable diagnostic support for dental professionals.
92.67% Overall Accuracy for Caries Detection| Metric | MobileNetV2 | ShuffleNet | Proposed Hybrid Model (without ACO) | Proposed Hybrid Model (with ACO) |
|---|---|---|---|---|
| Accuracy |
|
|
|
|
| Precision |
|
|
|
|
| Recall |
|
|
|
|
| F1-Score |
|
|
|
|
| AUC |
|
|
|
|
Application in Automated Dental Diagnosis
The high accuracy and robustness of the ACO-optimized hybrid model demonstrate its strong potential for integration into automated dental diagnosis systems, providing reliable classification of dental caries from panoramic radiographic images. This can significantly reduce diagnosis time and increase efficiency for dental professionals.
Key Outcome: Improved diagnostic precision and faster identification of caries with minimal human intervention.
Calculate Your Potential ROI
Estimate the impact of AI automation on your operational efficiency and cost savings.
Your AI Implementation Roadmap
A clear path to integrating advanced AI into your operations.
Phase 1: Discovery & Strategy
Comprehensive assessment of current workflows, identification of AI opportunities, and tailored strategy development. Define KPIs and success metrics.
Phase 2: Solution Design & Development
Architecting the AI system, custom model training or fine-tuning, and robust software development. Focus on seamless integration with existing infrastructure.
Phase 3: Deployment & Integration
Pilot program implementation, user training, and full-scale deployment. Continuous monitoring and iterative adjustments for optimal performance.
Phase 4: Optimization & Scaling
Post-implementation analysis, performance optimization, and strategic scaling of AI solutions across other business units. Explore advanced features and new use cases.
Ready to Transform Your Enterprise with AI?
Connect with our experts to explore how tailored AI solutions can drive efficiency, innovation, and growth for your business.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat