Understanding
Every Lesion,
Every Skin
YourPersonalAILesionScreeningAgent—onephotoisallittakes.Multi-agentclinicalAIscreensskinlesionsacrosseightISIC2019classes,computesABCDEmorphometrics,assessespersonalrisk,andprovidesinteractivetriageguidance.
Powered by open dermatology research & autonomous clinical AI agents
Early Detection, Before it Spreads
Skincanceristhemostcommoncancerworldwide,yetsurvivalformelanomacaughtearlyisabove99%.DermaAgent—YourPersonalAILesionScreeningAgent—putstrainedclinicalAIineverybrowser:photographalesion,andinunderaseconditismatchedagainsteightclinicalclassesfromtheISIC2019dermoscopybenchmark.
Everypredictionshipswithaconfidencescore,aseveritytriageandpracticalnextsteps—andeverynumberwequoteismeasuredontheofficialISICtestsplit,notondatathemodeltrainedon.Thisisascreeningaid,notadiagnosis:anythingsuspiciousshouldalwaysbereviewedbyadermatologist.
Built on Benchmarked Science
Threetechniquescarryaphotographfromyourcameratoarankedclinicalmatch.
Transfer Learning
A MobileNetV3-Large vision network pretrained on ImageNet, then fine-tuned on 20,262 dermoscopy images from the ISIC 2019 benchmark to recognize eight lesion classes.
Balanced Training
Dermoscopy datasets are heavily skewed toward common moles. Class-balanced sampling, label smoothing and frozen batch statistics keep rare classes like SCC and VASC from being drowned out.
Real-time Inference
The model is exported to TorchScript and served by a FastAPI backend — top-three predictions with confidence scores in well under a second, even on CPU, with no image ever stored server-side.
What We Do Best
Everythingbetweennoticingaspotandknowingwhattodoaboutit.
AI Lesion Screening
Classify a skin lesion across eight clinical classes from a single photograph.
Severity Triage
Every prediction carries a concern level and the care steps that follow.
Risk Assessment
A structured questionnaire that scores your personal skin-cancer risk factors.
ABCDE Self-Check
The dermatologist's visual rule for spotting suspicious moles, built in.
On-Device Privacy
Your photos are analysed and discarded — nothing is stored on a server.
Open Method
Training code, model card and measured metrics are all inspectable.
Your Personal Risk Profile
Sevenquestionsdermatologistsaskfirst—phototype,UVhistory,moleburdenandfamilyhistory.Answeredlocallyinyourbrowser;nothingisuploaded.
How old are you?
Cumulative UV damage makes risk rise with age.
Smart Dermatology
Upload a photo of a skin lesion and our convolutional neural network classifies it across eight lesion types — trained on the ISIC 2019 dermoscopy dataset, with care guidance in seconds.
1 · Provide a photo
A clear, close-up, well-lit image works best.
Drag & drop an image here, or click to browse
JPG · PNG · WEBP — up to 12 MB
2 · Read the result
Top-3 predictions with confidence and care guidance.
Your diagnosis will appear here — with confidence scores, care tips and urgency guidance.
Model transparency
Connecting to the inference service…
How it works
- 1Your photo is resized to 128×128 and normalized — nothing is stored on our servers.
- 2A MobileNetV3-Large network, fine-tuned on 20K+ dermatologist-labelled dermoscopy images, scores all 8 lesion classes.
- 3You get the top-3 matches with confidence, severity guidance and care tips.
ABCDE self-check
- AAsymmetry — one half differs from the other.
- BBorder — irregular, scalloped or blurred edges.
- CColor — multiple shades of brown, black, red or blue.
- DDiameter — larger than 6 mm (pencil eraser).
- EEvolving — any change in size, shape or color.
Not a medical diagnosis. Smart Dermatology is an educational screening aid built on public research data. It cannot replace examination by a qualified dermatologist — always confirm any concerning lesion with a medical professional.
An Agent That Reasons Clinically
Multi-agent architecture coordinating neural image classification, computer vision ABCDE morphometrics, symptom parsing, and clinical knowledge retrieval with complete reasoning transparency.
How can I assist your skin health today?
Attach a lesion photo to trigger neural classification and ABCDE morphometrics, describe your symptoms, or explore dermatology questions with clinical guidelines.
Educational clinical triage screening aid. Integrates deep CNN inference with computer vision ABCDE morphometrics and clinical guideline synthesis. Does not replace histological biopsy.
Numbers That Matter
Most melanomas are found by patients themselves — a spot that changed, itched, or bled. Giving people a credible first screen, with the right nudges to see a clinician, is what early detection actually looks like.
That Spot You Keep
Noticing?
Photograph it, get a ranked match across eight lesion classes, and see whether it is worth booking a dermatologist. It takes about a minute, and nothing is uploaded to a server.