Related Experiment Videos
Whitening black boxes: Interpretable and explainable DL-based systems for trustworthy healthcare
Antonio Lo Faro1, Yves Grandvalet2, Luca Ulrich1
1Department of Management and Production Engineering, Politecnico di Torino, C.so Duca degli Abruzzi, 24, Torino, 10129, Italy.
Artificial Intelligence in Medicine
|June 5, 2026
Summary
Explainable AI (XAI) methods enhance deep learning (DL) in healthcare by making AI predictions interpretable. This review categorizes XAI techniques, highlighting their benefits and drawbacks for clinical use.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Explainable Artificial Intelligence (XAI)
Background:
- Deep learning (DL) models are increasingly used in healthcare for medical image analysis and clinical decision support.
- DL models often function as 'black boxes,' lacking transparency in their predictions, which hinders clinical trust and adoption.
- Explainable AI (XAI) offers methods to interpret DL model outputs, making them more accessible to developers and clinicians.
Purpose of the Study:
- To present a comprehensive taxonomy of widely used XAI methods for image classification.
- To analyze the benefits and drawbacks of different XAI techniques in the healthcare context.
- To investigate the influence of classifier type on XAI method selection and the impact of 'black box' models in healthcare.
Main Methods:
- Systematic literature review of papers published between January 2020 and July 2025 from Scopus and Google Scholar.
- Adherence to PRISMA guidelines for enhanced reporting and study selection.
- Classification of 69 identified XAI methods into four categories: backpropagation, perturbation, attention, and concept-based.
Main Results:
- Backpropagation-based techniques are increasingly utilized, offering intuitive heatmap explanations.
- Perturbation-based methods are common for validating model robustness but are computationally intensive.
- Concept-based and attention-based methods, though less common, show promise for human-semantic alignment and reflecting model behavior.
Conclusions:
- Future research should prioritize combined XAI approaches and concept-based methods for clinically relevant and computationally efficient explanations.
- Developing XAI techniques that align with clinical semantics is crucial for transparent and reliable DL systems in healthcare.
- Addressing the 'black box' nature of DL models through XAI is essential for advancing AI adoption in clinical practice.
Related Concept Videos
01:11Blinding
Blinding is a commonly used method of not telling participants which treatment a subject is receiving. Blinding is a critical part of a randomized control trial or RCT. It reduces the bias that affects the results. In an RCT, blinding is used in the form of a placebo. A placebo effect occurs when untreated subjects falsely believe they have received the treatment and report improved symptoms. A placebo or a dummy treatment is administered to subjects to negate the bias caused by such an effect.
01:30Health Information Technology and Healthcare Information System
Health Information Technology (HIT)
Health Information Technology, commonly called HIT, integrates advanced information systems and technology in healthcare settings. Its primary functions include:
Health Information Technology, commonly called HIT, integrates advanced information systems and technology in healthcare settings. Its primary functions include:
01:29Issues And Trends In Healthcare Delivery System
The issues and trends in healthcare delivery are constantly changing. The COVID-19 pandemic is one recent issue that wreaked havoc on healthcare systems, causing a shortage of healthcare workers, high demand for medicines and supplies, and increased medical expenditure due to a lack of insurance. Other issues include rising healthcare costs and care fragmentation.
Cost Containment
Payment for healthcare services has historically promoted adoption of costly and often unnecessary or inefficient...
Cost Containment
Payment for healthcare services has historically promoted adoption of costly and often unnecessary or inefficient...