SkinGenics - An application to classify your cancerious lesions into benign and malignant

View Deployment

NOTE

Images should be stricltly benign or malignant, as it is a binary classification

Any random image (non cancerious lesion) will produce random output. 

Skin lesion images taken without applying oilment and opening pores will produce incorrect output.

SkinGenics helps you to predict your skin cancer between Benign and Malignant. It integrates Deep Learning technology with your device’s camera to help you capture and analyse skin lesions with more than 96% accuracy

Please Note: This application is built for learning purpose

How it Works

SkinGenics is a Skin-Cancer prediction application that uses CNN and Deep learning technology to calculate the probability that a skin lesions is Benign or Malignant, by comparing its appearance to thousands of other lesion images. SkinGenics has more than 96% accuracy

1. Upload Picture of Your Skin Lesion

2. SkinGenics will compare it with thousands of similar lesion images

3. SkinGenics will predict the class of cancer

Google Cloud

1. 'Google-Cloud-VM-Instance' is used to deploy React App
   Frontend URL - https://cancer-detection-deep-learning.de.r.appspot.com/


2. 'Google-Cloud-Function' is used to serve backend (GCP Service)
   Backend URL - https://us-central1-cancer-detection-deep-learning.cloudfunctions.net/predict


3. Model is built using Jupyter Notebook on 'GCP VM Instance - AMD Machine'
   Instance IP - http://34.168.91.133:5000 (SSH Commands)


4. GCP Buckets are used to store dataset and save models (.h5)

Folder Structure

1. Dataset : Contains ISIC 2018-HAM10000 Kaggle Dataset (nearly 33%)

2. Backend : Contains Fast-Api configuration to serve model

3. Frontend : Contains React.js codebase to serve UI

4. GCP : Contains Google-Cloud Configuration to serve backend

5. SavedModels : Stores model versions with various epochs and model architecture

6. Training : Stores notebook for current Convolutional Neural Network (CNN) Model