NeuralSight AI Documentation

What is in this Repository

This repository contains the entire codebase for NeuralSight AI solution backend code. Welcome to the NeuralSight Medical Image Detection GitHub repository! This repository contains code and resources for using deep learning techniques to detect specific features in medical images. The goal of this project is to improve the accuracy and efficiency of medical image analysis, with potential applications in diagnostics and treatment planning.

NeuralSight AI is an intelligent tool that is capable of detection and highlighting pathologies on chest X-Ray images at runtime. The following Python Scripts are part of the codebase:

Training directory

• Cleaning.py : Responsible for cleaning the x-ray image dataset.
• Extract.py : Creates Train and Validation Files that follow yolo dataset standards.
• Plotting.py: Allows you to visualize images and other exploratory data analysis graphs.
• Process.py: Enables you to create a csv file that include cleaned features generated from the initial dataset
• Training.py: Enables you to train a deep learning algorithm based on the x-ray images available and the clean csv file.

Application Directory

This directory contains directories and python & Fastapi Scrips used to develop the final API endpoint of the solution. You can find the API endpoints of the project Here and alternative documentation Here

For an an indepth understanding on how to setup the project, read through the project documentation available in the link below. You can find the project documentation online Here and it's github repository Here

Table of Contents:

• Important Resources
• Getting Started
• Pipeline
• Performance
• License
• Known Bugs
• Support and contact details

1. Important Resources

1. What is NeuralSight
2. NeuralSight Project Documentation
3. NeuralSight Frontend Repository
4. Company Charter
5. Have a new Feature or an Issue you want fixed?

Getting Started

1. Clone the repository
2. Install the required dependencies
3. Download the dataset and place it in the appropriate folder
4. Run the code
5. Prerequisites

Pipeline

Model training

1. Data Preparation:

• Image resize (640*640)
• Split data: training, testing, valid dataset(0.9, 0.1)
• Classes: Atelectasis, Infiltration, Emphysema, Mass, Nodule, Pleural Thickening, Effusion, Consolidation etc.

2. Training Configuration

    Before starting training the script, an environment file (.env) will need to be created with the following parameters.

    DATA_DIRECTORY=<path to root folder of the data>
    TRAIN_DIR=<path to a directory where the training data and information will be placed>
    WANDB_API_KEY=<wandb secret key>
    CURR_DIR=<path to current working directory>

  step1:
    pip install -r requirements.txt
  step2:
    python  cleaning.py
  step3:
    python process.py
  step4:
    python  extract.py
  step5:
    python -m training.py

2. Running Inferences.

   cd NeuralSight_AI

- `sudo docker build -t <image name> .`
- `docker run -d -p 80:80 <image name>`


OR

  sudo docker-compose build
  sudo docker-compose up

Performance

Table classification performance on the validation set.

License

The code in this repository is licensed under the GPL 3.0 license. This license allows for free use, distribution, and modification of the code, with the requirement that any derivative works must also be released under the GPL 3.0 license. You can find the full text of the GPL 3.0 license at the following link: https://www.gnu.org/licenses/gpl-3.0.en.html and a summary of the license can be found here: https://www.gnu.org/licenses/quick-guide-gplv3.html.

Please note that while this code is open-sourced, any medical data used in this code should be used only with permission and under strict compliance with patient privacy laws and regulations.

Known Bugs.

{There are no known bugs}

Support and contact details

{Any issues, questions, ideas, concerns or contributions to the code are highly encouraged. Contacts: info@neurallabs.africa, paul.ndirangu@neurallabs.africa & tom.kinyanjui@neurallabs.africa}