Urgent skin lesion referrals rose 170% in 10 years, and dermatology waitlists grew 82% since 2021. Long-term hiring plans aside, patients with chronic skin conditions must also have fair access to care.1
Given these growing pressures, integrating healthcare AI systems into dermatology presents a promising approach to enhance access, reduce waiting times, and ensure that all patients receive timely care.
Explore the top 5 AI in dermatology use cases with real-life examples.
Skin cancer screening in clinical settings
Artificial intelligence in dermatology is being increasingly utilized in skin cancer screening, particularly in primary care and real-world clinical settings where access to dermatologists is limited. AI algorithms help identify potential skin cancers early, particularly malignant skin lesions such as melanoma and non-melanoma skin cancers.
By analyzing skin lesion images, these systems can flag suspicious cases for further inspection, improving access to early detection and reducing unnecessary referrals.
The use of deep learning techniques, particularly convolutional neural networks (CNNs), has enabled the development of models that achieve dermatologist-level accuracy in classifying various skin lesions. This enhances diagnostic accuracy during skin cancer screening workflows and supports timely clinical decision-making.
For example, a meta-analysis of dermatologist performance set an NPV benchmark of 98.9%, while DERM, the NHS-approved AI system, demonstrated a superior NPV of 99.8% across real-world NHS data. This suggests AI can match or exceed clinician performance in safely excluding melanoma.2
Skin cancer diagnosis using deep learning models
AI technologies enable the automated dermatological diagnosis of skin cancer, particularly melanoma, using clinical and dermoscopic images. Deep learning algorithms trained on extensive clinical data and skin images now match or even exceed human performance in melanoma detection and the classification of other malignant cutaneous tumors.
Large annotated datasets, such as those provided through the International Skin Imaging Collaboration, support the development of these models. These initiatives ensure standardized data collection processes, enabling AI algorithms to generalize well to diverse populations and imaging conditions.
AI’s effectiveness in skin cancer diagnosis has been evaluated in randomized clinical trials, showing promise in increasing the sensitivity of skin tumor diagnosis and helping to identify malignancies in their early stages.
Differential diagnosis of skin diseases
Beyond cancer, artificial intelligence in dermatology also plays a role in distinguishing between various skin diseases with overlapping features. Using image and pattern recognition capabilities, machine learning algorithms analyze medical images to assist with differential diagnosis of conditions such as psoriasis, rosacea, and acne, as well as inflammatory skin diseases like eczema.
These tools function as computer vision diagnosis systems, augmenting human cognitive functions in clinical environments. Some models have demonstrated proficiency in detecting eczema skin lesions, contributing to better management of chronic conditions.
Classification and triage of skin lesions
AI is used to classify pigmented skin lesions and non-pigmented types, providing structured assessments for clinical practice. When integrated into healthcare systems, these tools can assist with triage by directing high-risk cases to specialists and managing benign cases conservatively.
Melanoma image classification using deep neural networks exemplifies how AI supports both the diagnostic and logistical aspects of care delivery. By optimizing the flow of patients in clinical settings, AI helps balance resource allocation while ensuring patient care is not compromised.
Decision support tools in primary care and teledermatology
Artificial intelligence (AI) has also become a cornerstone of medical AI devices designed for use by medical professionals in primary care settings. These tools support non-specialists in making decisions about whether to refer patients based on visual inspection of skin conditions.
In teledermatology, AI enhances image analysis when a dermatologist’s input is delayed or unavailable. Incorporating AI into clinical workflows helps bridge diagnostic gaps and standardize assessments across geographic and socioeconomic barriers.
Explore AI in healthcare use cases for more information.
AI in dermatology: Real-life examples
Diagnostics with the SkinVision app
The SkinVision app is a regulated medical device designed to help users assess skin spots for potential signs of skin cancer. By utilizing an AI-based algorithm, the app analyzes images of skin lesions taken with a smartphone and provides an immediate indication of risk. This tool serves as a self-examination aid, offering guidance on whether a professional medical consultation is advisable.
Key features are:
- Smart check camera: Allows users to capture images of skin spots and receive instant risk assessments.
- Body map: Enables tracking of skin spots over time to monitor changes.
- Risk profile assessment: Offers personalized advice tailored to individual skin types and risk factors.
- Reminders: Helps users establish and maintain a regular skin health monitoring routine.
The app is clinically validated and holds certifications, including the CE mark, TGA approval, and ISO certification. Healthcare professionals and organizations recommend it for its role in supporting early detection and promoting proactive skin health management.3
Figure 1: SkinVision’s skin cancer detection interface.
The German Cancer Research Center on diagnostic accuracy
The German Cancer Research Center (DKFZ) has developed an AI-based system to support skin cancer diagnosis, particularly for the early detection of melanoma. What sets it apart is its ability to explain its decisions by highlighting diagnostic features in skin lesion images, helping dermatologists understand and trust its assessments.
In a three-phase study, the system was shown to improve doctors’ diagnostic confidence and trust in AI recommendations.4
The SkinHealthMate App for skin lesion diagnosis
The SkinHealthMate App is a web and mobile application that uses ensemble deep learning (EfficientNetB1 and EfficientNetB5) to enhance dermatological diagnostics.
The system addresses limitations of traditional diagnosis, such as subjective biopsy site selection and delayed detection, by providing automated classification of skin lesions using the HAM10000 dataset.
Data augmentation and noise reduction techniques were applied to enhance model generalization, resulting in a 93% classification accuracy. The platform supports user-friendly workflows, patient record management, and treatment recommendations, and optimizes clinical decision-making.
It incorporates Grad-CAM visualizations for interpretability and offers advanced performance across various skin lesion types, though performance is weaker on melanoma due to class imbalance. Unlike earlier works, this platform integrates diagnosis, patient care, and usability into a single solution.5
Research on multimodal vision foundation in dermatology
A recent article on skin cancer diagnosis leveraged a multimodal AI model trained on over 500,000 clinical and dermoscopic images, patient metadata, and clinical notes from diverse populations across six continents. This model was benchmarked against human general practitioners, dermatologists, and other AI models in multiple settings, including primary care and teledermatology.
Key findings include:
- The AI achieved dermatologist-level classification accuracy in identifying over 540 skin conditions.
- It outperformed general practitioners in the detection and diagnosis of skin cancers and inflammatory skin diseases.
- The system maintained high diagnostic accuracy across different skin tones, a critical step toward reducing racial bias in dermatological AI tools.
- The AI model was evaluated in real-world clinical settings, including randomized clinical trials, showcasing its practical utility and generalizability.
Notably, the study emphasized the value of multimodal input (images combined with patient metadata) for enhancing diagnostic performance, particularly in cases where visual inspection alone is insufficient. The model demonstrated effectiveness in skin cancer screening, pigmented skin lesion classification, and early detection of malignant cutaneous tumors.6
Challenges of AI in dermatology
Generalizability of AI models
Artificial intelligence algorithms trained on data from specific populations or clinical settings often struggle to generalize across diverse environments. For example, an AI model developed using European datasets may underperform in primary care settings in Asia or Africa due to differences in pigmentation, lesion presentation, and imaging conditions.
This limitation is critical for skin cancer diagnosis and early detection, where incorrect generalization can result in missed malignant cutaneous tumors or unnecessary excisions of benign pigmented skin lesions. Ensuring diagnostic accuracy across varied populations remains a significant hurdle.
Ambiguities in the clinical context
AI in dermatology generally excels in image-based pattern recognition; however, clinical decision-making requires more than just visual inspection. Critical contextual factors, such as lesion evolution, patient history, and comorbidities, are often excluded from the training of neural network models.
This lack of context may lead to suboptimal differential diagnosis, particularly in cases involving inflammatory skin diseases or nonmelanoma skin cancer. Without structured integration of patient data, the diagnostic capabilities of medical AI devices remain limited when compared to the nuanced judgments made by human cognitive functions.
Benchmarking and validation constraints
Despite the promising results from AI algorithms in controlled environments, validation in clinical settings remains inconsistent. Many reported performance metrics rely on retrospective analysis rather than randomized clinical trials or prospective studies. As a result, there’s limited understanding of how artificial intelligence in dermatology performs under actual clinical pressure.
Moreover, studies comparing artificial intelligence algorithms often employ different test sets, metrics, or evaluation procedures, making it challenging to assess their relative strengths or weaknesses. A lack of standardized benchmarks hinders the ability to identify the most effective tools for tasks such as melanoma detection or eczema skin lesion detection.
Ethical and legal implications
AI applications that rely on patient data, including skin lesion images and other sensitive clinical data, must navigate complex ethical and legal landscapes. Data security, consent, and anonymization are crucial in protecting patient privacy, especially in healthcare systems that process large volumes of medical images.
External Links
- 1. NHS England » Exploring the future – artificial intelligence (AI) and dermatology.
- 2. https://www.edgehealth.co.uk/wp-content/uploads/2024/08/Evaluating-Pathways-for-AI-Dermatology-in-Skin-Cancer-Detection.pdf
- 3. Getting Started - SkinVision App.
- 4. AI-based support system for skin cancer diagnostics explains its decisions - German Cancer Research Center.
- 5. ScienceDirect.
- 6. A multimodal vision foundation model for clinical dermatology | Nature Medicine.
Comments
Your email address will not be published. All fields are required.