Pneumonia Detection AI 🤖

🚧 There's a new version coming with PyTorch and a better GUI 🚧

This project uses a deep learning model built with the TensorFlow Library to detect pneumonia in X-ray images. The model architecture is based on the EfficientNetB7 model, which has achieved an accuracy of approximately 97.12% (97.11538%) on our test data. This high accuracy rate is one of the strengths of our AI model.

Important

The code that had achieved the highest acc is backup/V6/Model_T&T.ipynb.
And the codes with the light and super light models are backup/V7/Model_T&T.ipynb, backup/V8/Model_T&T.ipynb.

🚀 Usage

Tip

If you just want the model go to the Github Releases.

Note

This model was built using the EfficientNet library, which is a collection of state-of-the-art models for image classification. To use the model, you need to install the library and import it as follows:

import efficientnet.tfkeras

Or clone the project and use Utils.FixedDropout:

from Utils.FixedDropout import FixedDropout
from keras.models import load_model

# Load the model
model = load_model('PAI_model_T.h5', custom_objects={'FixedDropout': FixedDropout})

The project includes a Command Line Interface (CLI) and a (GUI) Graphical User Interface for easy use of the model. The CLI, which is based on the Python CLI template from the same author, provides a user-friendly, colorful interface that allows you to interact with the model. you can fined the CLI in

Caution

The last version of CLI is V0.8.9.3 and currently its has entered a minimal maintenance and not being updated and its going to be deprecated.

Interface\CLI

Additionally, a Graphical User Interface (GUI) is available. you can fined the GUI in

Interface\GUI

Example Image of the CLI (V0.8.9.3) ⤵

Example Image of the GUI (V0.8.9.6) ⤵

📦 Release

Newest release 📃

Go to newest release

📜 Training System Specifications

• Graphics Card (GPU): RTX 3090
• Memory (RAM): 64GB
• Operating System (OS): Windows 11 Pro
• Processor (CPU): Intel Core i7-12700KF

📜 Model

The model is a Convolutional Neural Network (CNN) trained on a dataset of 23681 X-ray images. The dataset is a combination of the following:

• Chest X-ray Pneumonia dataset from Kaggle
• Covid19-Pneumonia-Normal Chest X-Ray Images from Mendeley
• RSNA dataset

This combined dataset provides a comprehensive set of images for training the model.\

Model list:

Model	Base Model	Params	acc	Status	Grad Cam
V6	efficientnet-b7	65.4M	97.12%	✅	✅
V7 light	efficientnet-b4	29.7M	97.12%	✅	⚠️
V8 Super light	efficientnet-b0	4.8M	96.47%	✅	❌

🧪 Training Methods

The AI model supports two distinct training approaches:

• rev1: A straightforward method using Keras fit function for basic training.
• rev2: An enhanced training strategy incorporating data augmentation and subset training for improved accuracy and generalization.

rev2 Training Simplified:

• Memory Optimization: Begins with clearing system memory to ensure efficient resource utilization.
• Hyperparameter Setup: Configures essential training parameters such as epoch count and batch size.
• Data Enrichment: Utilizes data augmentation techniques to introduce variability in the training dataset.
• Focused Training: Implements training on data subsets to reduce overfitting and streamline the learning process.
• Adaptive Learning Rate: Applies a dynamic learning rate schedule to fine-tune the training progression.
• Training Supervision: Uses callbacks for monitoring training, saving the best model, and enabling early stopping.
• Progressive Learning: Trains the model iteratively on subsets, evaluating and adjusting after each epoch.
• Data Standardization: Normalizes image inputs to facilitate model training.
• Robustness Enhancement: Introduces random noise to training images to strengthen model robustness against unseen data.
• While rev1 is suitable for quick and simple model training, rev2 is tailored for those seeking a more sophisticated and potentially more effective training regimen.

📁 Repository Structure

Please note that due to the large size of some files and folders, they are not available directly in the repository. However, they can be found in the Releases page of the repository. This includes the model weights and the database, which are crucial for the functioning of the AI model.

🤝 Contribution

Any contributions to improve the project are welcome. You can submit a pull request or open an issue on GitHub. Please make sure to test your changes thoroughly before submitting. We appreciate your help in making this project better.

⚠️ WARNING

Caution

The model provided in this project should not be used for medical diagnosis without further validation. While the model has shown high accuracy in detecting pneumonia from X-ray images, it is not a substitute for professional medical advice. Please consult with a healthcare professional for medical advice.

📂 Other

Note

Please note that this code uses my:

• Python-CLI-template
  • for more info go to https://github.com/Aydinhamedi/Python-CLI-template.
• Python-color-print-V2
  • for more info go to https://github.com/Aydinhamedi/Python-color-print-V2.
• Python-color-print
  • for more info go to https://github.com/Aydinhamedi/Python-color-print.

📃 Results

Warning

Results were achieved using Rev2 training method and Rev1.2 model and with backup/V6/Model_T&T.ipynb code.

Acc:

📷 Grad cam:

Model	Grad-cam Ex
V6
V7 light	🚧None🚧
V8 super light	🚧None🚧

🚧 Other:

📚 Citing

@software{Hamedi_Pneumonia-Detection-Ai,
author = {Aydin Hamedi},
license = {MIT},
title = {Pneumonia-Detection-Ai},
url = {https://github.com/Aydinhamedi/Pneumonia-Detection-Ai}
}

License

This project is open-source and is licensed under the MIT License. See the LICENSE file for details.