Skip to content
DE 🇩🇪 – EN 🇬🇧
Contact
Intelligence Lab

Where data becomes intelligence.

The Stroke Helper Lab is the core of our learning system.
Here, structured and clinically validated datasets are analyzed, patterns are identified, and algorithms are continuously refined.
What begins in the app and is clinically evaluated in the Pro version converges here — to optimize future screenings.
Not a static product.
A system that improves with every validated data point.

Contact

Quality, transparency, and control

tructured data foundation

Changes in heart rate provide important indicators of acute stress and potential neurological events.

Clinical validation

Algorithms evolve based on clinically evaluated cases. Pattern analysis, optimization, and evaluation are conducted in a structured and traceable manner. Every improvement is documented.

Version control & transparency

Models are versioned. Changes are traceable. Deployment states remain reproducible. This ensures regulatory clarity.

Interoperable architecture

The lab operates with FHIR-compliant datasets based on HL7 standards. Structured data enable study readiness, system integration, and international interoperability.

The learning loop

App → Pro → Lab → optimized models → improved app. A closed-loop system that becomes more precise with every validated data point.

The physician decides. The AI supports.

The AI does not operate autonomously. Physicians review, assess, and can correct results. The Lab is designed as a supportive system — not a replacement for clinical expertise.

Intellligence Lab (laboratory)

The lab overview displays the current status of all analyses – from uploaded and automatically processed datasets to cases pending manual review. It also shows the AI models in use – StrokeFaceAsymmetry, StrokeFaceCNN, and FusionCalib – as well as the current CPU and GPU utilization.

New Case – Start of Training

In this step, new datasets are created. In addition to before-and-after images, optional voice recordings and risk factors such as high blood pressure, diabetes, or atrial fibrillation can be added. These data form the foundation for training the models and help identify patterns and changes more accurately over time.

Review – Evaluation and Assessment

During the review process, submitted datasets are analyzed and assessed. The AI evaluates image quality, facial asymmetries, and changes in facial features between the before-and-after images. A manual review by medical experts follows to validate the results and further optimize the models.

Landmark Analysis – Detecting the Smallest Changes

In the landmark analysis, characteristic points on the face are marked — for example around the eyes, eyebrows, mouth corners, nasal folds, and facial contour. These reference points allow the AI to detect even the smallest asymmetries or subtle changes that may indicate a potential stroke. The collected data is directly integrated into model training, continuously improving the precision of stroke detection.

Manual Validation – Human and AI in Collaboration

While the AI can analyze and learn autonomously, humans remain a crucial part of the process. Medical experts and reviewers can manually inspect, correct, or supplement datasets. This creates a valuable feedback loop: the AI supports the analysis – humans validate and refine the results. Together, accuracy and reliability continuously improve.

Intelligence Lab

Central platform for AI training
Start: with human involvement (doctors, experts validating data)
Goal: continuous improvement of detection accuracy

Clinical Pro

Rollout version for hospital staff
Used for data collection in real environments
Provides diagnostic capabilities
Sends data & results back to the Lab → strengthens training

AI MIRA – in-house development

The core of the system

StrokeFaceAsymmetry: Analysis of facial asymmetries (mouth, eyes, cheeks, eyebrows)
StrokeFaceCNN: Deep learning model (Convolutional Neural Network) for image & 3D FaceID recognition (TrueDepth data, facial scans)
FusionCalib: Fusion of image, speech, and checklist data for higher precision
Pattern recognition & algorithm development → continuous improvement of early detection in new app versions

Stroke Helper App

For patients & relatives
Detects symptoms in time (face, speech, checklists)
Supports in emergencies: action guidance & emergency call option