SkinGuard – Skin Lesion Classification System

SkinGuard is an end-to-end machine learning project for classifying skin lesions as benign or malignant using deep learning.
The system includes model training, API-based inference, and a web dashboard for uploading images and viewing results.

This project was developed as part of my Project (University of London, 2023) and is intended for educational and research purposes only.

---

Project Overview

The application consists of:

• A MEAN stack web application (Angular + Node.js + MongoDB)
• A FastAPI service for model inference
• Multiple fine-tuned CNN models, including:
  • EfficientNetB7
  • Inceptionv3
  • MobileNetV3
  • ResNet50
  • VGG19
  • Xception
  • Custom CNN
• A dashboard to visualize predictions and stored results
• Other User Modules

---

How to Run the Application

MEAN Stack Module

The MEAN_STACK_MODULE folder contains:

• Angular frontend
• Node.js backend

Install Dependencies

All required dependencies are listed in: MEAN_STACK_MODULE/package.json

Install the dependencies and start the backend and frontend as per your local setup.

---

FastAPI Module

The FAST_API_MODULE folder contains the FastAPI backend used for model inference.

Follow the official FastAPI documentation to set up and run the API: https://fastapi.tiangolo.com/tutorial/first-steps/

---

Database Connection (Node.js Backend)

The MongoDB connection is configured in the Node.js backend (app.js) using Mongoose.
The password is read from an environment variable (MONGO_ATLAS_PW) for security.

---

Model Training Steps

Download the Dataset

• Download a skin lesion dataset (ISIC / Kaggle or equivalent)
• Organize the dataset into:
  • benign
  • malignant folders

---

Train and Fine-Tune Models

• Run the notebooks provided in the Notebooks/ folder
• Fine-tune the models using transfer learning

---

Add Models to FastAPI

• Save or download all fine-tuned models
• Copy the model files into: FAST_API_MODULE/models/

---

Custom CNN Model (JSON)

The folder below contains the JSON version of the custom CNN model: MEAN_STACK_MODULE/backend/ml_models/sg_model_1/

This model file will be made available on request.

---

Image Storage Setup

Create Images Folder

Create the following folders:

• FAST_API_MODULE/images/
• MEAN_STACK_MODULE/backend/patient_images

These folder is used to store uploaded images.

---

Add Test Images

For testing, add a few unseen images to:

• FAST_API_MODULE/skin_images/test/malignant/
• FAST_API_MODULE/skin_images/test/benign/

These images can be used to validate predictions via the API.

---

What the System Does

• Upload a skin lesion image
• Run inference using a trained CNN model
• Store predictions and metadata (persistence)
• Display results on the dashboard for later review
• Provides role-based access through separate user modules

---

User Modules

The application includes separate modules for different user roles:

Admin

• Manages users
• Controls system settings
• Monitors overall system activity

Doctor

• Manages consultations
• Reviews patient cases
• Reviews model predictions

Patients

• Book consultations
• Upload skin lesion images
• View prediction results

Lab Technicians

• Handle lab results
• Manage test-related data

---

License & Usage

This project is shared for learning, academic, and portfolio purposes only.
It is not approved for clinical or commercial use.