Configuring the AI Engine
The Neolens AI engine is highly configurable, allowing you to tailor detection sensitivity, analysis depth, and output verbosity to your clinical or research needs.
Understanding Key Configuration Parameters
Neolens AI models provide several configuration parameters to let you tailor how the system interprets, analyzes, and outputs results. The most frequently adjusted parameters include:
1. Detection Sensitivity (sensitivity
)
Defines how aggressively the model should flag findings.
low
(0.3): Reduces false positives but may miss subtle anomaliesmedium
(0.5): Balanced behavior (default)high
(0.8): Detects even minor anomalies, with higher risk of noise
{
"sensitivity": "medium",
"confidence_threshold": 0.5
}
2. Analysis Depth (depth
)
Controls how deeply the AI should analyze each image.
shallow
: Faster analysis, surface-level insights (~2-5 seconds)standard
: Good tradeoff for most use cases (~5-15 seconds, default)deep
: Slower but richer and more exhaustive evaluation (~15-45 seconds)
3. Output Verbosity (verbosity
)
Specifies the amount of detail included in the response.
minimal
: Essential data only (findings, confidence scores)standard
: Includes explanations and context (default)verbose
: Detailed results including intermediate stepsdebug
: Full raw output with model internals (development only)
4. Model Selection (model_version
)
Choose specific AI model versions for different use cases:
Model Version | Specialization | Best For |
---|---|---|
v3.2.1 | General radiology | X-rays, CT scans (default) |
v3.1.5 | Neuroimaging | MRI brain scans |
v2.8.3 | Cardiology | Cardiac imaging, ECG |
experimental | Latest features | R&D, testing only |
5. Region of Interest (roi
)
Focus analysis on specific anatomical regions:
{
"roi": {
"region": "thorax",
"coordinates": [100, 50, 400, 300],
"auto_detect": true
}
}
⚡ Performance vs Quality Trade-offs
Configuration | Speed | Accuracy | Resource Usage | Best For |
---|---|---|---|---|
shallow + low | ⚡⚡⚡ | ⭐⭐ | Low | Screening, high-volume |
standard + medium | ⚡⚡ | ⭐⭐⭐ | Medium | General clinical use |
deep + high | ⚡ | ⭐⭐⭐⭐ | High | Critical cases, research |
🎯 Configuration by Use Case
Emergency Department Triage
{
"sensitivity": "high",
"depth": "shallow",
"verbosity": "minimal",
"priority_filter": "critical_only",
"response_time_target": "< 5s"
}
Research & Development
{
"sensitivity": "high",
"depth": "deep",
"verbosity": "debug",
"model_version": "experimental",
"uncertainty_estimation": true
}
Routine Screening
{
"sensitivity": "medium",
"depth": "standard",
"verbosity": "standard",
"batch_processing": true,
"false_positive_tolerance": "low"
}
🛠️ Advanced Configuration Options
Custom Thresholds
{
"thresholds": {
"anomaly_detection": 0.7,
"classification": 0.8,
"measurement_precision": 0.95
}
}
Multi-Modal Analysis
{
"multi_modal": {
"enabled": true,
"primary_modality": "ct",
"secondary_data": ["clinical_notes", "lab_results"],
"fusion_strategy": "weighted_ensemble"
}
}
Quality Assurance
{
"qa_settings": {
"image_quality_check": true,
"minimum_resolution": "512x512",
"artifact_detection": true,
"preprocessing": "auto"
}
}
📊 Configuration API
Set Configuration
curl -X PUT "https://api.neolens.ai/v1/config" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"sensitivity": "high",
"depth": "standard",
"verbosity": "verbose",
"model_version": "v3.2.1"
}'
Get Current Configuration
curl -X GET "https://api.neolens.ai/v1/config" \
-H "Authorization: Bearer YOUR_API_KEY"
Validate Configuration
curl -X POST "https://api.neolens.ai/v1/config/validate" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"sensitivity": "ultra_high",
"depth": "experimental"
}'
⚠️ Configuration Warnings
Performance Impact
deep
analysis can increase processing time by 3-5xdebug
verbosity generates large response payloads (>10MB possible)experimental
models may have unstable behavior
Clinical Considerations
- Higher sensitivity increases false positive rates
- Lower sensitivity may miss subtle but critical findings
- Always validate configuration changes with clinical workflows
🧪 Testing Your Configuration
A/B Testing Framework
{
"test_config": {
"baseline": {"sensitivity": "medium", "depth": "standard"},
"variant": {"sensitivity": "high", "depth": "deep"},
"sample_size": 100,
"metrics": ["accuracy", "false_positive_rate", "processing_time"]
}
}
Configuration Profiles
Save commonly used configurations:
{
"profiles": {
"emergency": {...},
"research": {...},
"screening": {...}
}
}
📈 Monitoring Configuration Impact
Track how configuration changes affect:
- Accuracy metrics: Sensitivity, specificity, AUC
- Performance: Response time, throughput
- Clinical outcomes: False positive/negative rates
- User satisfaction: Radiologist feedback, workflow efficiency
Best Practices
- Start with default settings and adjust incrementally
- Test configuration changes on validation datasets before production
- Document configuration rationale for regulatory compliance
- Monitor performance metrics after configuration updates
- Use profiles to quickly switch between use cases