Upload Skin Lesion Image
Drag and drop your image here or click to browse
For the best results, check the image quality guidelines below for tips and example images.
Supported formats: JPG, JPEG, PNG, HEIC
Skin Lesion AI
Advanced skin lesion analysis powered by a custom‑trained AI model
Detectable Lesion Types
Melanoma (MEL)
Most serious form of skin cancer
Melanocytic Nevus (NV)
Common benign skin lesions (moles)
Basal Cell Carcinoma (BCC)
Most common type of skin cancer
Actinic Keratosis (AKIEC)
Pre-cancerous skin condition
Benign Keratosis (BKL)
Non-cancerous skin growths
Dermatofibroma (DF)
Benign skin nodules
Vascular Lesion (VASC)
Blood vessel-related skin lesions
Image Quality Guidelines
How to Capture Quality Images
Proper Lighting
Use natural daylight or bright, even artificial lighting. Avoid shadows and harsh direct light that creates glare.
Proper Framing
Fill the frame with the lesion area. Crop out unnecessary features like side shadows, fingers, and background distractions.
Sharp Focus
Ensure the lesion is in sharp focus. Hold the camera steady and tap to focus on the lesion before taking the photo.
Appropriate Distance
Take the photo from close enough to see details clearly, but not so close that the image becomes blurry or distorted.
Image Quality Examples
✓ Excellent Quality
Clear, well-focused image with lesion prominently visible and good lighting
✓ Excellent Quality
Proper lighting and contrast with lesion details clearly visible
✓ Excellent Quality
Well-lit image with clear lesion boundaries and minimal background
✓ Excellent Quality
Good color accuracy and lesion detail visibility with proper framing
✓ Excellent Quality
Clear lesion with good contrast and proper lighting
✓ Excellent Quality
Well-defined lesion with sharp focus and appropriate distance
Additional Tips for Better Results
- Clean the area: Gently clean the skin around the lesion to remove any oils, creams, or debris that might affect image quality.
- Avoid flash: Use natural or ambient lighting instead of camera flash, which can cause glare and wash out important details.
- Multiple angles: If possible, take photos from slightly different angles to capture all aspects of the lesion.
- Steady hands: Use both hands to hold your device steady, or rest your elbows on a surface for stability.
- Check before uploading: Review your photo to ensure it's clear, well-lit, and the lesion is easily visible before analysis.
AI-Powered Analysis
Advanced EfficientNet model trained on thousands of dermatological images
Instant Results
Get analysis results in seconds with detailed confidence metrics
Privacy Protected
Your images are processed locally and not stored on our servers
Training Progress & Model Performance
Dataset Size
Trained on 4,015 dermatoscopic images from the ISIC dataset, this marks the first phase of training. The process is still ongoing, and the skin lesion detection model will be updated as training and validation progress. The current training phase uses 401,059 images, courtesy of the International Skin Imaging Collaboration.
Model training methodology
This project is a supervised deep learning approach for skin lesion classification using EfficientNet - a convolutional neural network (CNN) architecture with a custom head, fine‑tuned on the ISIC dataset to identify seven lesion types.
Training images undergo advanced augmentation (random crop, flip, rotation, color jitter) and ImageNet‑based normalisation, while validation/test sets are resized and normalised.
Class imbalance is addressed by computing class weights and applying Focal Loss (gamma=2) to prioritise hard examples. Optimisation uses AdamW (lr=0.001, weight decay=0.01) over 20 epochs with batch size 32.
Metrics including accuracy, loss, and UAR are tracked via Weights & Biases, with final evaluation on a held‑out test set.
Skin Lesion Detection AI – Phase 1 Training Results
Using EfficientNet - a convolutional neural network (CNN) architecture with a custom classification head, the model was trained on 4,015 ISIC dermatoscopic images.
Training accuracy reached ~89%, with validation accuracy at ~76% and UAR at ~77%, closely aligned — a strong indicator of good generalisation.
Loss decreased steadily with minimal overfitting, showing the model is learning robust patterns across lesion types.
These early results set a solid foundation for further tuning and scaling with the full 401k‑image ISIC dataset.
Frequently Asked Questions
What is Skin Lesion AI?
Skin Lesion AI is an advanced artificial intelligence system designed to analyze dermatoscopic images and assist in the identification of various skin lesions. Our AI model uses deep learning techniques trained on thousands of medical images to provide rapid, preliminary analysis of skin conditions.
How accurate is the AI analysis?
In the initial phase, this version of the AI model demonstrated a strong capacity to learn from data. While its accuracy on the training data reached 89%, a more important measure is its performance on new, unseen images. In this area, the model achieved an accuracy of approximately 77%.
This 77% accuracy represents a strong and promising baseline for Phase 1. We anticipate significant improvements in Phase 2 as we expand the dataset and continue to refine the model.
It's also important to note that accuracy can vary depending on factors like image quality and lighting conditions. To support interpretation, the AI also provides a confidence score with each analysis, indicating the reliability of the prediction.
Can this replace a doctor's diagnosis?
No, absolutely not. This AI tool is designed for educational and screening purposes only. It should never replace professional medical diagnosis or treatment. Always consult with a qualified dermatologist or healthcare provider for proper medical evaluation and treatment decisions.
What types of skin lesions can it detect?
The AI can identify 7 types of skin lesions: Melanoma (MEL), Melanocytic nevus (NV), Basal cell carcinoma (BCC), Actinic keratosis (AKIEC), Benign keratosis (BKL), Dermatofibroma (DF), and Vascular lesions (VASC).
Is my data secure and private?
Yes, your privacy is a priority. Images are processed locally in your browser and are not stored on the servers. No personal information or images are collected, transmitted, or retained by the system.
Who developed the Skin Lesion AI model?
This AI model was developed by Joseph Mondejar as part of the Master of Artificial Intelligence program at La Trobe University. It showcases the use of supervised deep learning for medical image analysis, specifically for classifying seven types of skin lesions.
The project is part of an ongoing academic journey, with a strong focus on ethics, accessibility, and the potential to support early detection of serious skin conditions like melanoma.
📬 For more info or collaboration, contact Joseph at 22687842@students.latrobe.edu.au or connect via LinkedIn.
How can I collaborate or provide feedback?
Your feedback and collaboration can help improve the Skin Lesion AI tool and increase its real-world impact. You can:
- 💬 Suggest improvements to usability or features
- 🧑⚕️ Contribute medical expertise or validation support
- 🧠 Collaborate on AI or data research
- 🌍 Partner on outreach to make the tool accessible in remote areas
To get involved or share ideas, please reach out via:
📧 Email: 22687842@students.latrobe.edu.au
🔗 LinkedIn: joseph-mondejar-666b6b30
Share Your Feedback
Help us improve the Skin Lesion AI Analyzer. Your feedback is valuable to us!
Important Medical Disclaimer
FOR EDUCATIONAL AND INFORMATIONAL PURPOSES ONLY
This AI-powered skin lesion analyzer is designed solely for educational and informational purposes. It is NOT a medical diagnostic tool and should NOT be used as a substitute for professional medical advice, diagnosis, or treatment.
- Not Medical Advice: The results provided by this AI system do not constitute medical advice and should not be relied upon for medical decision-making.
- Consult Healthcare Professionals: Always seek the advice of qualified healthcare professionals, particularly dermatologists, for proper evaluation of skin lesions.
- Emergency Situations: If you notice rapid changes in a mole or lesion, bleeding, itching, or other concerning symptoms, seek immediate medical attention.
- Limitations: AI systems have limitations and may produce false positives or false negatives. Results should never delay or replace professional medical consultation.
- No Liability: The developers and operators of this tool assume no responsibility for any medical decisions made based on the AI analysis results.
Remember: Early detection and professional medical evaluation are crucial for skin health. When in doubt, always consult a healthcare professional.