Region guided mask R-CNN with Haralick ResNet fusion for accurate coronary artery disease detection in computed tomography angiography images
Unlocking Advanced AI for Enterprise
This paper presents an integrated framework for accurate coronary artery disease (CAD) detection from CCTA images. It addresses challenges like noise, low contrast, and motion artifacts. The framework combines Region-Guided Mask R-CNN (RG-Mask R-CNN) for precise segmentation, Haralick-ResNet Fusion (HRF) for robust feature extraction, and a Deep Convolutional Network (DeepConvNet) for efficient classification. This hybrid approach significantly improves diagnostic performance by reducing structural complexity and capturing subtle variations, achieving 98.3% accuracy.
Key Performance Indicators
Drawing inspiration from the impressive results of the study, here are the core metrics that highlight the potential of advanced AI in real-world applications.
0
Accuracy
0
Loss
0
Dice Coefficient
0
FPS
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
0
Segmentation Accuracy
Region-Guided Mask R-CNN (RG-Mask R-CNN) achieved 96% accuracy for coronary artery segmentation by combining region growing for initial identification with Mask R-CNN for precise instance segmentation. This hybrid approach effectively handles complex structures, noise, and low contrast.
RG-Mask R-CNN Workflow
Input Processing (Image preprocessing and normalization)
→
Feature Extraction (Backbone Network extracts hierarchical features)
→
Region Proposal Network (Generates candidate object regions with confidence scores)
→
ROI Alignment (Precise feature alignment for accurate localization)
→
Non-Maximum Suppression (Eliminates redundant overlapping bounding boxes)
→
Segmented Output (Final segmented image with detected objects)
| Model | Accuracy | Dice Coefficient | IoU | Benefits |
|---|---|---|---|---|
| Proposed Work | 96% | 0.91 | 0.87 |
|
| U-Net | 93% | 0.89 | 0.85 |
|
| V-Net | 94% | 0.90 | 0.86 |
|
0
Highest Feature Importance (ResNet Feature 2)
Haralick-ResNet Fusion (HRF) effectively combines texture information from Haralick features (e.g., Angular Second Moment, Inverse Difference Moment) with high-level deep features from ResNet. ResNet Feature 2 (High-level) showed the highest importance score of 0.12, crucial for differentiating subtle variations in CCTA images.
Haralick-ResNet Fusion for Enhanced Feature Extraction
The study highlights that Haralick features (contrast, correlation, energy, homogeneity) capture important texture information from arteries and surrounding tissues, while ResNet extracts high-level deep features. This fusion strategy in the feature space, denoted as FHRF = WHH + WRR, significantly enhances the robustness and discriminative power of the feature extraction process, making it adept at identifying subtle variations often missed by traditional methods. This hybrid approach improves differentiation of normal from abnormal tissues based on illumination conditions, a critical aspect for accurate CAD diagnosis in complex CCTA images.
0
Overall Accuracy
DeepConvNet achieved an impressive 98.3% overall accuracy in classifying CAD. Its hierarchical feature representation, from low-level textures to pathological patterns, enables better generalization and reduces the need for manual feature extraction, outperforming VGG-16, ResNet-50, InceptionV3, and EfficientNet-B3.
| Model | Accuracy (%) | Loss | F1 Score | Key Advantage |
|---|---|---|---|---|
| Proposed DeepConvNet | 98.3 | 0.15 | 98.2 |
|
| VGG-16 | 87.2 | 0.4 | NA |
|
| ResNet-50 | 90.8 | 0.32 | NA |
|
| InceptionV3 | 91.5 | 0.29 | NA |
|
| EfficientNet-B3 | 93.0 | 0.26 | NA |
|
Quantify Your AI ROI
Estimate the potential return on investment for implementing advanced AI in your enterprise workflows, using metrics from this research.
Your AI Implementation Roadmap
A phased approach to integrating advanced AI, inspired by the systematic methodology of this research.
Phase 1: Discovery & Strategy
Assess current workflows, identify AI opportunities, and define clear objectives and success metrics based on similar research applications.
Phase 2: Data Preparation & Model Selection
Curate and normalize your enterprise data. Select and adapt AI models, leveraging insights from top-performing architectures like RG-Mask R-CNN, HRF, and DeepConvNet.
Phase 3: Customization & Integration
Tailor the chosen AI models to your specific enterprise context. Integrate them seamlessly into existing IT infrastructure and ensure robust data pipelines.
Phase 4: Validation & Deployment
Rigorously test the AI solution with enterprise-specific data, validate performance against defined metrics, and deploy in a controlled environment.
Phase 5: Monitoring & Iteration
Continuously monitor AI performance, gather feedback, and iterate on models to optimize for long-term efficiency and evolving business needs.
Ready to Transform Your Enterprise with AI?
Schedule a personalized consultation with our AI specialists to discuss how these cutting-edge techniques can be applied to your unique business challenges.
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