This project focuses on detecting malaria-infected red blood cells (RBCs) using a Convolutional Neural Network (CNN).
It is built as a hands-on learning project .
The goal of this project is to understand:
- How image data is processed
- How CNN models are trained
- How medical image classification works in practice
- Python
- TensorFlow / Keras
- NumPy
- Pandas
- Matplotlib
- Seaborn
- Google Colab
The dataset contains microscopic images of red blood cells divided into two classes:
- Parasitized (Infected)
- Uninfected (Healthy)
The dataset is loaded directly using TensorFlow Datasets (TFDS).
- Dataset loading & preprocessing
- Image resizing & normalization
- CNN model building
- Model training with callbacks
- Evaluation using:
- Accuracy
- Confusion Matrix
- ROC-AUC
- True vs Predicted visualization
- Model saving & loading
The model uses a custom CNN built with:
- Convolution Layers
- Batch Normalization
- Max Pooling
- Fully Connected (Dense) Layers
- Dropout for regularization
- Sigmoid activation for binary classification
The model was trained and evaluated earlier using GPU on Google Colab.
Currently, due to GPU limitations, the notebook is being re-run for final clean outputs.
Evaluation includes:
- Training & Validation Accuracy
- Confusion Matrix
- ROC Curve
- True vs Predicted image visualization
Final results will be updated soon.
- Clone the repository
- Open the notebook in Google Colab
- Enable GPU runtime
- Run all cells from top to bottom
Two methods are used:
- Saving the full model (architecture + weights)
- Saving only the weights
This helps in resuming training and using the trained model later without retraining.
Through this project I learned:
- How CNN works internally
- How medical image classification is implemented
- How to evaluate models using:
- Confusion Matrix
- ROC-AUC
- Precision, Recall, F1-score
- How to save and reload trained models
Digambar
GitHub: https://github.com/Digam-hue