PNEUMONIA X-RAY DETECTION

Brief Background on Pneumonia

• Well, for a long time, Pneumonia has been the leading cause of death to children under the age of five(under 5 years old).
• There exist a high possibility that pneumonia misdiagnosis leads to more severe cases of Pneumonua thus reducing chances of survival for patients diagnosed with the illness.
• In various reports, including the World Health Organization report on Pneumonia in 2017, it is reported that adults with chronic illnesses and those above 65 years are more prone to severe Pneumonia leading to many deaths of this vulnerable groups.
• However, with proper diagnosis and treatment of pneumonia, the chances of survival for a given patient will hence increase thus saving a larger population which would succumb to this severe illness.

Causes of Pneumonia

• There exists no specific causitive agent for pneumonia as various pathogens affecting the lungs may result into pneumonia.
• The causes of pneumonia range from:
  1. Bacterial infections of the lungs.
  2. Viral infections of the lungs.
  3. Fungal infections in the lungs(though rare).

About the DataSet

• The Dataset contains 5216 train images, 624 test images and 16 test images all belonging to two classes {Pneumonia, Normal} which totals to 5856 images.

• We also have randomized patient identification for all the patients.

• The Chest X-Rays are Posterior and Anterior, and have been obtained from retrospective cohorts of pediatric patients between 1 - 5 years old from Guangzhou Women and Children’s Medical Center, Guangzhou. However, it should be noted that the effects of pneumonia affect all vulnerable groups of individuals which include adults above 65 years and those living with chronic illnesses.

• For analysis, it is important to note that, all chest radiographs are initially screened for quality control by removing all low quality or unreadable scans. The diagnoses for the images were then graded by two expert physicians before being cleared for training the AI system.

• Be lest assured that our dataset is clean to use for the training of our model. Thank You👌🦾!

• More About the Dataset? Find it Here!

Brief Data Reporting

• We noted the distribution of the images for the Train set, Test set and validation set using a pie chart.

1. Pie Chart Showing Distribution of Images in The Data Batches

• It is good to note that in the pie-chart in our notebook, Train data has the highest share of the total images, followed by test data with a large margin and validation data tails the list.
• Therefore, from our data, we will have 87.2% of training images, 10.6% of test data and 2.1% of validation data.

2. Sample Images in the Dataset

• We can view the posterior and anterior images of the X-ray on the image in the sampled images in the notebook.

3. Showing Metrics and Charts of the Model

• Metrics: Accuracy, Precision, Recall, Specificity and F! Score.
• Charts:

  • Train Vs. Validation Loss

  • Training Vs. Validation AUC

  • Confusion Matrix

    1. Losses, AUC scores, Confusion Matrix
    • From the confusion matrix charts: The model is able to predict 50.80% of True Positives(Testing Pneumonia positive) and 34.62% of True negatives which results in the data being predicted having relatively high accuracy in the two classes. However, the percentage of false negatives is seemingly high which would result to an error.
    • The training loss drops gradually then slowly.
    • The validation loss is inconsistent.
    • The Training AUC is inconsistent while the validation AUC is gently increasing.

Directories Arrangement

pneumonia_x_rays-reading/
├── .gitattributes
├── best_model.hdf5
├── pneumonia_detection.ipynb
├── README.md
├── requirements.txt
├── Readme_Images/
│   ├── chest_x-ray.png
│   ├── cm.png
│   ├── dataset_pie.png
│   ├── roc.png
│   └── ...
├── Report/
│   ├── Pneumonia_detection.docx
└── webapp/
    ├── model/
    |  ├── pneumonia_x_rays_v1_0.keras
    |  └── pneumonia_x_rays_v2_0.keras
    ├── src/
    |   ├── static/
    |   └── templates/ 
    ├── templates/
    ├── virtual/
    ├── app.py
    └── .gitignore

Model Use

• To use this model, you'll need to input the X-ray image of the chest that will be evaluated.

• To access it locally:

  • Clone this repository

  git clone https://github.com/Dan-njuguna/pneumonia_X_rays_reading.git

  • Navigate to the repository

  cd pneumonia_X_rays_reading

  • Install project requirements:

  # On your Bash Terminal, Ensure you have python3 and pip installed
  pip install -r requirements.txt

  • Open the '.ipynb' file and run it to see your result.

  code .

Using the Web version

• Ensure you have saved the model version on your local machine.
• Navigate to webapp folder.

cd webapp

• Create a virtual environment.

python3 -m venv virtual

• Activate the virtual environment.

source virtual/bin/activate

• You will see the terminal change to something like this:

(virtual) dan@dan:~/pneumonia_X_rays_reading/webapp$ 

• Install requirements:

pip install -r requirements.txt

• Run the api script:

python3 app.py

• In the case you want the Deployed Version of the model find the link here!

UI Preview

Thank You!