Documentation

Complete guide to using Derma-AI for skin lesion classification

Getting Started

Derma-AI provides a simple web interface for skin lesion classification. Follow these steps to begin:

1

Upload Image

Upload a dermatoscopic image using the drag-and-drop interface

2

Run Prediction

Click the "Predict" button to analyze the image

3

View Results

Review classification results and confidence scores

API Reference

POST /api/predict

Submit an image for classification

Request
Content-Type: multipart/form-data

{
  "image": <file>
}
Response
{
  "prediction": {
    "class": "melanoma",
    "confidence": 0.87,
    "probabilities": {
      "melanoma": 0.87,
      "nevus": 0.08,
      "basal_cell_carcinoma": 0.03,
      "actinic_keratosis": 0.02
    },
    "grad_cam_url": "https://api.derma-ai.com/gradcam/abc123.png"
  },
  "processing_time": 1.2,
  "model_version": "v2.1.0"
}
Model Details

Architecture

  • ViT + EfficientNet Fusion
  • Input Size: 224×224
  • Parameters: 120M
  • Pre-trained on ImageNet

Performance

  • Accuracy: 89.9%
  • F1-Score: 0.9766
  • AUC-ROC: 0.9766
  • Training F1: 0.9999

Supported Classes

Melanoma
Melanocytic Nevus
Basal Cell Carcinoma
Actinic Keratosis
Benign Keratosis
Dermatofibroma
Vascular Lesion
Integration Guide

JavaScript/React Integration

const formData = new FormData();
formData.append('image', imageFile);
const res = await fetch('/api/predict', {
  method: 'POST',
  body: formData
});
const result = await res.json();

Python (requests) Integration

import requests

with open("lesion.jpg", "rb") as img:
  res = requests.post("https://api.derma-ai.com/predict", files={"image": img})
print(res.json())
Code Examples
Troubleshooting

Common Issues

Image Upload Fails

Ensure image is under 10MB and in JPG/PNG format

Low Confidence

Ensure good image lighting and focus

API Rate Limit

Free users are limited to 100 requests/day

Best Practices

  • Use dermatoscopic images when possible
  • Ensure lesion is well-lit and centered
  • Trim excessive hair or glare
  • Always consult a dermatologist

Support

Need help? Contact:

  • Email: support@derma-ai.com
  • GitHub: github.com/derma-ai
  • Docs: https://docs.derma-ai.com