AI in dermatology: a Munich specialist on triage, trust, and the limits of the algorithm

Author: Alexander Börve

July 23, 2026

The integration of artificial intelligence into dermatology has progressed from experimental validation to real-world clinical application. AI-assisted skin analysis tools, including systems such as Autoderm, are increasingly utilised for preliminary assessment and triage, with measurable effects on diagnostic efficiency, patient engagement, and the use of healthcare resources.

This article synthesises expert insights from Prof. Dr. Daniela Hartmann (Department of Dermatology, Allergology and Laser Medicine, Munich), who, in a recent interview with Medscape Germany, provides a measured and critical perspective on the evolving role of AI in dermatological practice. It focuses on clinical utility, validation standards, patient interaction, and system limitations. It also highlights the role of explainable AI (XAI) in human-machine collaboration.

Clinical utility: AI as a first point of orientation

AI-supported dermatological tools are increasingly recognised for their ability to provide rapid preliminary assessments of skin lesions. These systems function primarily as triage tools, offering initial orientation rather than definitive diagnoses.

Prof. Dr. Hartmann highlights that AI-supported tools can rapidly and automatically generate preliminary assessments of skin changes, providing patients with an initial point of orientation. This positions AI as an entry point into the diagnostic pathway, not a replacement for clinical evaluation.

Current evidence suggests that AI algorithms can achieve diagnostic performance comparable to experienced dermatologists in specific areas, particularly melanoma detection. Their principal value, however, lies in improving clinical workflow efficiency: helping prioritise cases, structure consultations, and make better use of specialist time. Prof. Dr. Hartmann is explicit that AI should be regarded as a decision-support tool, not a substitute for clinical expertise.

Clinical validation and regulatory standards

The deployment of AI in healthcare requires robust clinical validation and adherence to regulatory standards. In Europe, CE certification provides a baseline level of safety and reliability for medical AI systems.

Prof. Dr. Hartmann notes that systems such as Autoderm are CE-certified (Class I/IIa) and clinically validated, which provides a credible foundation. She cautions, however, against broad claims that may overstate real-world performance, such as asserting that a relatively limited condition set covers the vast majority of dermatological presentations. For B2B buyers evaluating AI infrastructure, this is an important distinction: evidence-based adoption demands honest scope.

Published evidence supports the clinical contribution of AI-assisted dermatology tools. Escalé-Besa et al. (2023) reported a 34% reduction in specialist referrals when AI-based skin assessment was integrated into primary care workflows, alongside a 92% satisfaction rate among general practitioners. These findings indicate that AI can reduce systemic burden while maintaining clinical effectiveness.

Impact on patient behaviour and clinical interaction

Freely available skin diagnostic tools are already shaping patient behaviour. As AI-based condition suggestion tools become more widely accessible, their influence on the doctor-patient relationship will grow. Patients who use such tools before a consultation are, in effect, performing a form of self-triage: the system generates several preliminary suggestions, which the patient arrives with in hand.

According to Prof. Dr. Hartmann, this presents both positive and negative implications. On the positive side, patients who have engaged with AI tools are often better informed, which can facilitate more focused consultations and support shared decision-making. On the negative side, AI-generated outputs may increase anxiety: systems tend toward worst-case presentations, and patients can misinterpret preliminary suggestions as confirmed findings.

This makes clinical contextualisation essential. AI outputs must be treated as supportive information and interpreted within a clinical framework rather than accepted or dismissed at face value. Prof. Dr. Hartmann is clear that addressing AI-generated hypotheses during consultation, explaining their basis, correcting misinterpretations, and resolving patient anxiety, is an increasingly important part of clinical workflow. It can be time-consuming, and that cost should not be underestimated.

Limitations, human oversight, and integration strategy

Despite its potential, AI in dermatology is subject to several important constraints. Models trained under controlled study conditions may perform differently in real-world clinical settings. Dataset bias, particularly the underrepresentation of darker skin tones in training data, can affect accuracy across patient populations. Image quality presents a further variable: poor lighting, blurred margins, insufficient contrast, or image artefacts can all reduce the reliability of automated assessment. And for rare or complex conditions, systems trained primarily on common presentations will naturally be less well-calibrated.

Prof. Dr. Hartmann is candid that these limitations reveal a gap between research conditions and real-world practice. More comprehensive validation is still needed across skin types, clinical settings, and condition ranges. This is a live development area, not a settled one.

Effective integration therefore requires a hybrid model: algorithmic support paired with professional clinical oversight. This is reflected in how systems such as Autoderm are designed, with outputs framed as informational condition suggestions rather than diagnoses, and clinical responsibility remaining with the healthcare professional. The clinician makes the decision; the AI informs it. This design principle is not a regulatory formality but a practical safeguard for patient safety.

Future directions: explainable AI and clinical adoption

A significant development in AI-driven dermatology is the emergence of explainable artificial intelligence (XAI). Unlike conventional “black-box” models, XAI systems provide insight into how a diagnostic output was generated, surfacing the reasoning and not just the result.

Prof. Dr. Hartmann regards XAI as an important step forward. Explainability enhances clinician trust in AI, reduces cognitive burden, supports accountability, and is more likely to achieve both clinical and regulatory acceptance. Research from the German Cancer Research Centre in Heidelberg has demonstrated measurable advantages of XAI over standard AI in dermatological settings, including improved diagnostic performance and greater clinician confidence.

The broader direction, in Prof. Dr. Hartmann’s view, is toward genuine human-machine collaboration: systems that communicate confidence levels, highlight uncertainty, and make their reasoning interpretable. That is not a vision of AI replacing clinical expertise. It is a vision of AI that earns the clinician’s trust precisely because it does not pretend to certainty it does not have.