Skin Disease Detection Using Convolutional Neural Networks

This is our submission for Electrothon 3.0 as a framework for Skin Disease Detection that analyses the skin pigmentation using Convolutional neural network for diagnosis of skin diseases. Detailed problem statement can be found here.

Authors

• Mukul Thakur
• Akash Rana
• Kashish Srivastava

Overview

We have trained a Convolutional Neural Network on an extensive training dataset of images of skin diseases and rashes to classify the input as one of the 8 categories of diseases. We achieved 81.68% training accuracy and 75.56% validation accuracy upon training the neural network.\ The 8 different catagories of skin diseases that we have selected are:

• Acne Cystic
• Eczema
• Impetigo
• Psoriasis
• Rosacea
• Schamberd
• Allergic Contact Dermatitis
• Tinea Ringworm Groin

Accuracy-Loss Curve

Installation

Build from Source

Clone the repository and checkout to stable commit

git clone https://github.com/Mukulthakur17/Skin-Pigment-Analysis.git
cd electrothon

Install Requirements

pip install -r requirements.txt

Preparing the model

• Download the trained model from here.
• Move it to the directory "Model Weights _ Json".

Starting the uvicorn server

After installing requirements

uvicorn main:app

This would start the ASGI server at http://127.0.0.1:8000.

Using Front-End

• After starting the server open http://127.0.0.1:8000 in your browser.
• Upload the image that you want to examine for diseases.
• Click Diagnose

• Detailed log of Disease predicted by the neural networks and its symptoms and duration will pop up

  

Using API Directly

Request URL:

 http://127.0.0.1:8000/predict

Using Swagger UI:

• After starting the server open http://127.0.0.1:8000/docs in your browser to open the swagger UI for APIs.
• Scroll down to Post /predict
• Click Try it Out
• Chose file to upload for classification
• Click Execute
• Under Responses section , a detailed log of disease, summary ,symptoms and duration will be displayed in json format as the response