SkinGlance-ModelAPI

An intelligent and lightweight backend for skin cancer detection — combining multiple deep learning models into a robust ensemble, accessible via a FastAPI interface.

---

🧠 About the Project

SkinGlance-ModelAPI is the backend engine for the SkinGlance Android application — a mobile-first solution for early detection of skin cancer using deep learning. This repository contains:

• 🧪 Code to train and save multiple CNN models (DenseNet169, InceptionV3, MobileNetV2)
• 🤖 A fuzzy ensemble prediction strategy combining model outputs
• 🚀 A FastAPI-based RESTful API to serve predictions in real-time
• 🧹 Can be run locally or cloud environments like Kaggle Notebooks

---

🗂️ Project Structure

SkinGlance-ModelAPI/
├── model_training/        # Model development notebooks scripts
├── api_service/           # FastAPI app + saved models
└── README.md

🧪 Use Cases

✅ Case 1: Run on Kaggle

Ideal for experimenting without setting up a local environment

Follow these steps:

1. 🔁 Clone this repo to understand the full project structure:

   git clone https://github.com/sandip-mondal-0248/SkinGlance-ModelAPI

2. 📅 Import the training notebook (model_training/skinglance_model.ipynb) into Kaggle Notebooks.

3. 📂 Add a skin cancer dataset: Use ISIC Dataset(Preprocess version) or any other similar dataset of your choice. Structure it like:

   dataset/
   ├── train/
   │   ├── benign/
   │   └── malignant/
   └── test/
   

4. 🧠 Train the models by running the notebook in Kaggle. It will save:

   • DenseNet169_best.keras
   • InceptionV3_best.keras
   • MobileNetV2_best.keras

5. ✅ Test the ensemble prediction logic in the same notebook.

6. ⬇️ Download the trained models from Kaggle into your local computer.

7. 📦 Place them into:

   api_service/app/models/
   

8. 🚀 Run the FastAPI server locally using the command in main.py (Replace with actual host and port number):

   uvicorn app.main:app --host 0.0.0.0 --port 8000

9. 🌐 Test the API:

   • Using Postman or any HTTP client
   • Or simply send an image request from browser

---

💻 Case 2: Run Locally

If you're setting up the full system on your machine follow these steps:

1. 🔁 Clone the repo:

   git clone https://github.com/sandip-mondal-0248/SkinGlance-ModelAPI
   cd SkinGlance-ModelAPI

2. 🔧 Modify paths (if needed):

   • Replace dataset paths in notebooks or scripts with local ones
   • Set up any file I/O based on your directory structure

3. 🧠 Train the models by running the notebooks. It will save:

   • DenseNet169_best.keras
   • InceptionV3_best.keras
   • MobileNetV2_best.keras

4. 🧱 Set up the api environment:

   cd api_service
   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   pip install -r requirements.txt

5. ⏩ Continue with the same process as described in Case 1:

   • Train models
   • Save them to api_service/app/models/
   • Run FastAPI server
   • Test the API

---

📱 Try the Android App

You can also try our Android app, which uses this same API.

📱 Android App GitHub Repo: SkinGlance Android App

---

🔁 API Endpoints

Method	Endpoint	Description
GET	/	Health check
POST	/predict	Upload an image for prediction

---

🧠 Models Used in Ensemble

• ✅ DenseNet169 (models/DenseNet169_best.keras)
• ✅ InceptionV3 (models/InceptionV3_best.keras)
• ✅ MobileNetV2 (models/MobileNetV2_best.keras)
• 📊 Combined using a fuzzy rank-based fusion strategy

---

📂 Datasets

You can use public datasets like:

• ISIC preprocess version available on kaggle: skin-cancer-malignant-vs-benign
• Or your own dataset formatted as: train/benign, train/malignant

---

📜 License

This project is licensed under the MIT License

---

🤝 Contributing

Pull requests are welcome. For major changes, please open an issue first. If you find this helpful, please ⭐ star the repo!

---

📬 Contact

For questions, suggestions, or collaborations: Sandip Mondal — https://sandipmondal.bio.link/ 🔗 LinkedIn