DevMesh

DERMATO.AI : Skin Disease Detection Using AI

Mourad Bouache

Mourad Bouache

California

0 0
  • 0 Collaborators

Derma diseases are common conditions with a variety of causes. Some of them can cause death like Melanoma, but with early diagnosis, we can save many lives. Dermato.AI is an AI tool that can identify, detect, and segment of 3 types of similar-looking skin diseases. ...learn more

Project status: Published/In Market

Artificial Intelligence, oneAPI

Intel Technologies
DevCloud, oneAPI, Intel Python, Intel vTune, Intel CPU, AI DevCloud / Xeon, Movidius NCS

Docs/PDFs [1]Code Samples [1]Links [2]

Overview / Usage

Skin conditions are common, Dermato.AI can be used by non-medical individuals for diagnosis. This tool can be used by doctors in order to confirm the correctness of their diagnosis of the disease.

This tool will detect the type of disease in a smooth way very smooth and report the result after a few clicks.

Methodology / Approach

Our AI gets an input image from the user and makes its predictions on it, and gives us bounding boxes, each with a class and a mask.

We made the AI on our dataset realized with LabelME with (03) three different methods:

  • Using oneAPI toolkit and Intel DevCloud using Keras with the TensorFlow backend to Train our UNET model,
  • Used the same API to train a CNN with GlobAveragePooling2D to be able to use the Convolutional part of the model, and get the features extracted from the image and identify the anomaly (the disease),
  • Mask RCNN provided by the Detectron2 library realized by Facebook, it is very accurate but slow.

Technologies Used

In order to compliment the project we went through several stages:

  1. The collection of the dataset
  2. **The verification of the dataset **
  3. Carrying out the models with different methods to choose the best way to satisfy the project.

The data collection part was the most difficult because it took several days and in it, we have (02) two parts, the images, and the annotations.

  • The Images: they were collected from different websites and they contained the disease (kaggle.com ..)

The annotation: They were created using LabelMe which is a graphical image annotation tool, we created for each image a .json file that has the class and the mask points.

Documents and Presentations

Dermato.AI.pdf

Repository

https://www.youtube.com/watch?v=3LboGde_1WM

Comments (0)