Skip to Content

Explainability Assessment (XAI)

Physical Consulting Virtual

Service Description

This module evaluates whether an AI model's decisions are based on meaningful, domain-relevant features or whether the model relies on undesired shortcuts and spurious correlations. The customer-provided AI model is analysed on a supplied test dataset using established explainable AI techniques, including attribution methods (e.g., LRP, CRP), concept-based explanations, and feature importance analysis. The evaluation identifies which input features drive model predictions and assesses their alignment with domain knowledge.

Outcomes can be: Positive Indicators • Alignment of model explanations with established domain knowledge • Transparent and interpretable decision patterns • Consistent reliance on clinically/scientifically relevant features

Negative Indicators • Detection of spurious correlations or shortcut learning (e.g., “Clever Hans” behaviour) • Reliance on irrelevant, confounding, or artefactual input features • Inconsistent explanation patterns across similar cases

Keywords: Explainable AI model interpretability transparency Layer-wise Relevance Propagation Concept Relevance Propagation shortcut learning
Offerings: Research & Development Technological Solutions & Documentation Model & Algorithm (Development, Optimization & Evaluation, etc.)
Provider Logo

Provider & Contact

Provider Country Germany
Organisation Website https://hhi.fraunhofer.de
Published Email tefhealth@hhi.fraunhofer.de
Pricing Detail

Pricing is defined on a case‑by‑case basis and depends on the specific customer requirements, model characteristics, and validation scope. A detailed offer can be prepared upon request.

Operational Details

Service Inputs • A trained AI/deep learning model (with access to intermediate layers if applicable) • A representative test dataset • Domain knowledge documentation (optional, e.g., clinical guidelines, feature relevance expectations)
Service Outputs • Evaluation report documenting applied XAI methods, visual explanations (e.g., heatmaps, concept attributions), and findings on model decision behaviour • Summary of identified shortcut or spurious correlation risks