& Treatment Response Prediction Platform — PhotoGuard AI is an AI-powered mobile imaging support tool that non-invasively assesses lesion planimetry, PpIX fluorescence, treatment response, outcome prediction and personalized PDT protocol from a standard lesion photograph. The system analyzes lesion area, PpIX distribution, erythema, crusting and epithelialization.

Post-PDT tissue color and geometric changes alter light reflection/scattering and fluorescence characteristics. The AI quantifies these optical signatures through image processing and machine learning to produce lesion area (cm²), PpIX score, response score and clearance probability.

Core Capabilities

• Planimetry: lesion area, perimeter, depth — with reference card calibration
• PpIX Fluorescence: intensity score (0-100), distribution pattern, incubation compliance
• Treatment Response: erythema, crusting, necrosis (%), epithelialization (%)
• Outcome Prediction: predicted sessions, clearance probability (%), recurrence risk
• Personalized Protocol: photosensitizer, light dose, wavelength, session interval
• Allergen & Light Protection: allergen-safe photosensitizer, SPF 50+ plan
• Teleconsultation: anonymized submission, urgency triage

The platform combines image processing and large language model (LLM)-based multimodal inference. From a standard RGB image, the lesion region is segmented; tissue, color, fluorescence and geometric features are extracted and clinical inference is performed with a multimodal LLM.

Processing Flow

• Image capture & quality audit (brightness, contrast, sharpness, reference card detection)
• Lesion segmentation and planimetric measurement (area, perimeter, depth)
• PpIX fluorescence assessment: intensity score, distribution pattern, incubation compliance
• Multimodal LLM inference: treatment response, outcome prediction, personalized protocol
• Summary metrics: lesion_area_cm2, fluorescence_score, response_score, clearance_probability

Software-Centric Architecture

The analysis provides multi-layer parameters across 7 specialized modules. Each parameter is reported as percentage (0-100), description, severity and risk level.

Output Format

• Track: lesion_area_cm2, lesion_perimeter_cm, lesion_depth_mm, area_reduction_pct
• Fluorescence: fluorescence_score, fluorescence_distribution, incubation_compliance
• Response: response_score, erythema_score, crusting_score, necrosis_pct, epithelialization_pct
• Predict: predicted_sessions, clearance_probability, recurrence_risk, stagnation_warning
• AI clinical commentary + personalized PDT protocol + printable full report (PDF)

The platform supports clinical workflows as a screening and pre-triage tool. Lesion parameter mapping provides structured, reproducible and comparable data to the physician. Final diagnosis and treatment decisions always rest with the specialist physician.

Safety Architecture (Human-in-the-loop)

• AI output is not a diagnosis; it is information supporting the physician's assessment
• AI commentary does not contain prescription/treatment recommendations
• Per-patient data isolation (RLS) and KVKK/GDPR-aligned sensitive-data approach

Integration Scenarios

The software-centric architecture aligns with European Union carbon-reduction and ESG (Environmental, Social, Governance) goals. Hardware independence brings medical consumable use close to zero and reduces the carbon footprint from patients' physical travel through telemedicine.

Roadmap

The following studies establish the clinical validity and methodological foundations of image-based AI assessment of photodynamic therapy and lesion evaluation. Our PhotoGuard AI platform draws inspiration from this literature but does not replace these methods.