COVID-19 Pneumonia Detection Using kNN-SVM and VGG16 Features

Overview

This repository implements a novel COVID-19 Pneumonia detection system using a combination of deep feature extraction from chest X-rays via the VGG16 architecture and a kNN-SVM hybrid model for classification. The project is designed to scale easily for large datasets and features a flexible architecture that allows hyperparameter tuning via command-line arguments.

---

Model Pipeline

  +------------------------------+
  | 1. Data Preprocessing        |
  |    and Augmentation          |
  |  - Load Images               |
  |  - Normalize                 |
  |  - Data Augmentation         |
  +------------------------------+
              |
              v
  +------------------------------+
  | 2. Exploratory Data Analysis |
  |    (EDA)                     |
  |  - Visualize Class Distributions|
  |  - Visualize Sample Images    |
  +------------------------------+
              |
              v
  +------------------------------+
  | 3. Feature Extraction        |
  |    (VGG16)                   |
  |  - Extract Features           |
  |  - Transfer Learning          |
  +------------------------------+
              |
              v
  +------------------------------+
  | 4. Dimensionality Reduction  |
  |    (Autoencoder)             |
  |  - Compress Features          |
  |  - Preserve Key Patterns      |
  +------------------------------+
              |
              v
  +------------------------------+
  | 5. Classification            |
  |    (kNN-SVM)                 |
  |  - Local Sensitivity (kNN)    |
  |  - Global Stability (SVM)    |
  +------------------------------+
              |
              v
  +------------------------------+
  | 6. Evaluation                |
  |  - Accuracy                  |
  |  - Precision                 |
  |  - Recall                    |
  |  - F1-Score                  |
  +------------------------------+

---

Features

• Transfer Learning: Utilizes the pre-trained VGG16 model for deep feature extraction.
• Autoencoder: Reduces feature dimensionality to optimize classification.
• kNN-SVM Hybrid Model: Combines the strengths of k-Nearest Neighbors (kNN) and Support Vector Machines (SVM) for robust classification.
• Command-line Interface: Supports command-line arguments for batch size, number of epochs, kNN neighbors, SVM regularization parameter, and more.

---

Dependencies

To install the necessary dependencies, run the following command:

pip install -r requirements.txt

---

Usage

To run the pipeline, use the following command:

python main.py --data_dir <path_to_data> --batch_size 64 --epochs 20 --k_neighbors 7 --svm_c 0.5 --eda

---

Arguments:

--data_dir: Path to the directory containing the dataset (train/test split).
--img_size: Image size (default: 224).
--batch_size: Batch size for feature extraction and training (default: 32).
--epochs: Number of epochs for autoencoder training (default: 10).
--k_neighbors: Number of neighbors for kNN (default: 5).
--svm_c: SVM regularization parameter C (default: 1.0).
--eda: Flag to perform EDA (visualizations).

---

Results

Upon training, the model outputs classification metrics such as precision, recall, F1-score, and accuracy.

---

---

Application

Before running the Streamlit application, ensure that the FastAPI server is running. If it's not running, start it with:

uvicorn backend.api:app --reload --host 0.0.0.0 --port 8000

---

Then, launch the Streamlit app:

streamlit run app.py

---

Dockerization

• Build the Docker Image:

docker build -t covid19-detection-app .

• Run the Docker Container:

docker run -d -p 8000:8000 -p 8501:8501 covid19-detection-app

---

Citation

If you use this code for your research, please cite:

@article{covid_knn_svm,
  title={COVID-19 Pneumonia Detection Using kNN-SVM and VGG16 Features},
  author={A Bahuguna, D Yadav, A Senapati, BN Saha},
  journal={International Journal of Machine Learning and AI Research},
  year={2024}
}