---
tags:
- skin-cancer-detection
- medical
- image-classification
- tensorflow
- keras
license: mit
datasets:
- HAM10000
metrics:
- accuracy
model-index:
  - name: Skin Cancer Detection Model
    results:
      - task:
          type: image-classification
          name: Image Classification
        dataset:
          name: HAM10000
          type: HAM10000
        metrics:
          - name: Accuracy
            type: accuracy
            value: 0.85
---

# Skin Cancer Detection Model

## Overview
This model was created as part of a final project for an AI bootcamp. It is a **skin cancer detection** model trained to classify skin lesions from dermatoscopic images using the **HAM10000 dataset**. The model is capable of predicting **seven different types of skin lesions**, each corresponding to various forms of skin cancer and other skin conditions.

The model has been trained using a **Convolutional Neural Network (CNN)** with **TensorFlow** and **Keras**. The goal of this project is to help in early detection of skin cancer by classifying images into seven distinct categories, which could assist healthcare professionals in diagnosis.

## Model Architecture
The model utilizes a CNN architecture fine-tuned for image classification tasks. Below is a brief description of the architecture:
- **Input size**: 224x224 RGB images
- **Base architecture**: Pretrained CNN (e.g., ResNet, VGG)
- **Output layer**: 7 softmax units, each corresponding to one of the skin lesion categories

## Model Performance
The model was trained on the HAM10000 dataset and achieved an accuracy of **85%** on the validation set. Further improvements could be made by additional fine-tuning and hyperparameter optimization.

## Datasets
The model was trained using the **HAM10000 dataset**, which consists of over 10,000 dermatoscopic images of skin lesions. The dataset includes seven types of lesions, described as follows:

| Label | Full Name | Description |
|-------|------------|-------------|
| **akiec** | Actinic Keratoses and Intraepithelial Carcinoma | A type of skin lesion that can develop into squamous cell carcinoma if left untreated. |
| **bcc**  | Basal Cell Carcinoma | A common form of skin cancer that rarely metastasizes. |
| **bkl**  | Benign Keratosis | Non-cancerous skin lesions like seborrheic keratosis. |
| **df**   | Dermatofibroma | A benign skin lesion usually found on the lower legs. |
| **nv**   | Melanocytic Nevus | Commonly known as a mole, usually benign but can develop into melanoma. |
| **vasc** | Vascular Lesions | Skin lesions that involve blood vessels, like angiomas. |
| **mel**  | Melanoma | The most dangerous form of skin cancer, often caused by UV radiation exposure. |

### Gradio Demo

You can try out the **skin cancer detection model** using the interactive demo hosted on Hugging Face Spaces [here](https://huggingface.co/spaces/syaha/skin-cancer-detection-gradio).

## Usage

To use this model for inference, you can load it using TensorFlow:

```python
from tensorflow.keras.models import load_model

# Load the model
model = load_model("path_to_model.h5")

# Preprocess input image and make predictions
image = preprocess_image("path_to_image.jpg")  # Custom image preprocessing function
prediction = model.predict(image)