Healthcare
Utilizing deep learning from mobile phone photos for early detection of horizontal strabismus: a screening approach
This study proposes an AI-powered framework for the early detection of horizontal strabismus using smartphone-acquired facial images. It combines a Real-Time Detection Transformer (RT-DETR) for localizing nine ocular landmarks per eye across three gaze directions (left, center, right), with supervised machine learning classifiers. Five biometric ratios derived from landmark coordinates serve as features. Trained on 150 participants (96 with strabismus, 54 controls) and using SMOTE for class imbalance, the RT-DETR achieved an Intersection over Union (IoU) of 0.62 and a mean center-point error of 6.52 pixels. The Random Forest classifier demonstrated high performance with an accuracy of 0.95, sensitivity of 0.96, and specificity of 0.94. This proof-of-concept highlights the feasibility of smartphone-based strabismus screening under controlled conditions, offering an interpretable and accessible approach for early detection.
Executive Impact
This AI-driven screening tool for horizontal strabismus could revolutionize early detection in ophthalmology, particularly in underserved regions. By leveraging smartphone photography and explainable AI, it reduces reliance on specialized hardware and clinician expertise, enabling broader, more accessible screening. This could significantly decrease rates of amblyopia and related psychosocial issues by facilitating timely referrals and intervention.
0
Accuracy Rate
0
Sensitivity Rate
0
Specificity Rate
0
Mean Center-point Error
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Healthcare: This category focuses on applications and advancements in medical diagnostics, patient care, and public health. AI's role in healthcare is rapidly expanding, offering solutions for early disease detection, personalized treatment plans, and improved operational efficiency. This paper exemplifies how AI can make specialized medical screenings more accessible and cost-effective, particularly in areas with limited resources.
Enterprise Process Flow
Smartphone Image Acquisition (3 Gaze Angles)
→
Eye Landmark Annotation (9 Points/Eye)
→
RT-DETR Model Training & Landmark Detection
→
Feature Engineering (30 Geometric Ratios)
→
Random Forest Classification (Strabismus vs. Orthotropia)
0.62
Intersection over Union (IoU) for Landmark Detection
| Feature | Proposed Method (RT-DETR + RF) | Prior Art (General CNNs/Specialized Hardware) |
|---|---|---|
| Hardware Requirement |
|
|
| Interpretability |
|
|
| Gaze Directions Covered |
|
|
| Accessibility & Scalability |
|
|
97
Positive Predictive Value (PPV) for Strabismus Detection
StabSeen: A Mobile Screening Prototype
The framework was implemented as a proof-of-concept mobile application called StabSeen, developed with Flutter and deployed on Android. It interfaces with an AI model stored on a server via a Fast API, returning a binary screening output and probability score. This demonstrates the technical feasibility and potential for real-world application, guiding users to capture three necessary gaze images (center, left, right) and processing them in 15-20 seconds. Data privacy is maintained by not storing images persistently on the server.
Calculate Your Potential ROI
See how much time and cost your enterprise could reclaim with AI automation. Adjust the parameters below to get a personalized estimate.
Annual Cost Savings
Calculating...
Annual Hours Reclaimed
Calculating...
Your AI Implementation Roadmap
A strategic, phased approach to integrating advanced AI into your enterprise, ensuring maximum impact and seamless adoption.
Phase 1: Data Acquisition & Pre-processing Enhancement
Expand data collection to include diverse demographics, lighting conditions, and smartphone models. Implement semi-supervised or self-supervised learning for automated, efficient landmark annotation on larger datasets.
Phase 2: Model Optimization & On-Device Deployment
Refine RT-DETR for enhanced landmark detection accuracy and robustness in uncontrolled settings. Quantize and optimize models for on-device inference, ensuring real-time, offline screening capabilities for improved accessibility and privacy.
Phase 3: Multi-center Clinical Validation & Subtype Classification
Conduct extensive multi-center clinical trials to validate sensitivity and specificity across diverse patient populations. Develop advanced feature fusion techniques to enable reliable classification of specific strabismus subtypes (e.g., esotropia vs. exotropia, vertical components).
Phase 4: Telehealth Integration & Regulatory Approval
Integrate the validated system into existing telehealth platforms. Pursue necessary regulatory approvals for medical device software, preparing for widespread adoption in community screening and remote diagnostics.
Ready to Transform Your Enterprise with AI?
Unlock the full potential of artificial intelligence to drive efficiency, innovation, and competitive advantage.
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