🩺Pneumonia Diagnosis from Chest X-Ray Images : an end-to-end application🫁

• I used this dataset from kaggle.
• First, I built a custom CNN.
• Later, I built several Transfer Learning (Feature Extraction + Fine Tuning) models using:
  • VGG16
  • ResNet152V2
• Out of the 5 model that I built, ResNet152V2 model with feature extraction and a custom classifier gave the best accuracy of 89.7%. (For the metrics of all models, please refer to Summary.xlsx)
• All the models were built using TensorFlow.
• I used Google Colab (with its free GPU) for the entire ML workflow. I downloaded this notebook as Pneumonia_Diagnosis_From_Chest_X_Rays_Final.ipynb.
• I then used that best ResNet152V2 model for making inferences, by building a web application using Flask python framework, where a user can upload a chest x-ray image and get the diagnosis results.

⚠️NOTE: The best ResNet152V2 saved model is very large, hence is not present in this repo. Download it from here [Open this link and download the entire folder. You will download a ZIP file. Unzip it and inside it, you will find a folder called resnet152v2_feature_extraction_final_best_model. Place this folder in the root of your project. Note that the filename should be as it is]. This file is required otherwise by the flask server to perform inferences.

Description of files in the repo

➡️ app.py - Flask Server

➡️ Pneumonia_Diagnosis_From_Chest_X_Rays_Final.ipynb - IPython Notebook containing the code used to train various TensorFlow models.

➡️ predict_result.py - Script used to load the saved model and perform inference for an input image.

➡️ requirements.txt - Contains the python dependencies & the associated versions.

➡️ Summary.xlsx - Excel file containing the metrics (such as train|val|test loss|accuracy, precision, recall, etc) for the 5 models that I have built

➡️ static and templates - Flask uses these folder to serve static files, store uploaded images & render HTML pages.

Steps to run the application

• Clone the repo.
• Download the model from the link given above.
• Make sure your folder structure is as follows:

resnet152v2_model_folder (Download it from above link)
static
templates
app.py
predict_result.py
requirements.txt

• Within this directory, install the dependencies using the command: pip install -r requirements.txt
• Within this directory, run the following command to start the server: flask --app app run
• Navigate to localhost:5000 to access the web application.
• You can use the following images for testing: Normal Chest X-Ray Pneumonia Chest X-Ray

Screenshots

Home Page:

Upload Normal Chest X-Ray image (before prediction):

Diagnosis Results for Normal Chest X-Ray:

Diagnosis Results for Pneumonia Chest X-Ray:

Error message for grayscale image:

(Some images which you download from the internet, or kaggle dataset are grayscale images and they do not contain 3 color channels. This model specifically required 3 color channels for inference.)