๐๐๐ ๐๐ซ๐จ๐ฃ๐๐๐ญ: ๐๐๐ฅ๐๐ซ๐ข๐ ๐๐๐ฅ๐ฅ ๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง ๐๐ฌ๐ข๐ง๐ ๐๐๐19
Malaria, a severe and widespread parasitic disease in humans, exhibits clinical symptoms as parasites infiltrate and multiply within human red cells. Transmission occurs when infected mosquitoes bite humans, releasing parasites into the bloodstream where they infect and destroy red blood cells.
- Google Colab
Link: https://www.kaggle.com/datasets/iarunava/cell-images-for-detecting-malaria
This dataset comprises 27,560 PNG images of malaria-infected and uninfected cells, grouped into two classes with 13,780 images each.
- Parasitized
- Uninfected
๐๐๐ฌ๐ค: ๐๐ข๐ง๐๐ซ๐ฒ ๐๐ฅ๐๐ฌ๐ฌ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง ๐๐ซ๐จ๐ฃ๐๐๐ญ ๐๐ก๐๐ฌ๐๐ฌ:
- Library Importation
- Data Subset Creation (Training, Testing, Validation)
- Data Processing (Prepare Training and Testing Data)
- Visualization of Training Images
- Architecture Development
- Model Compilation
- Training
- Evaluation
- Model Saving
- Prediction
- Testing
- Visualization of Test Images
Segregation into three subsets: training, testing, and validation using the split folder library. ๐๐๐ญ๐ ๐๐ซ๐จ๐๐๐ฌ๐ฌ๐ข๐ง๐ : Training: 22,069 images across 2 classes. Validation: 2,758 images across 2 classes. Testing: 2,760 images across 2 classes.
๐๐ข๐ฌ๐ฎ๐๐ฅ๐ข๐ณ๐๐ญ๐ข๐จ๐ง ๐จ๐ ๐๐ซ๐๐ข๐ง๐ข๐ง๐ ๐๐ฆ๐๐ ๐๐ฌ:
Parasitized:
Parasitized Image
Uninfected:
Uninfected Image
-
Utilization of VGG19 transfer learning technique to construct a malaria cell detection architecture.
-
Total params: 20,074,562
-
Trainable params: 50,178
-
Non-trainable params: 20,024,384
Implementation of a flatten layer to convert output into a 1D array.
Application of sigmoid activation function in the last FCNN layer due to the presence of only 2 classes.
๐๐จ๐๐๐ฅ ๐๐จ๐ฆ๐ฉ๐ข๐ฅ๐๐ญ๐ข๐จ๐ง, ๐๐ซ๐๐ข๐ง๐ข๐ง๐ , ๐๐ง๐ ๐๐ฏ๐๐ฅ๐ฎ๐๐ญ๐ข๐จ๐ง:
Utilization of binary cross-entropy and Adam optimizer for model compilation. Model trained over 50 epochs with plots for training and validation accuracy, loss, and validation loss.
Train Accuracy & Loss: [0.2546966075897217, 0.9057048559188843]
Testing Accuracy & Loss: [0.21056684851646423, 0.9217391014099121]
Model saved using h5 format.
Selection of image from testing data for prediction.
Conversion of the image into an array and expansion of its dimension.
Prediction made using if-else conditions based on maximum argument.
๐๐ข๐ฌ๐ฎ๐๐ฅ๐ข๐ณ๐๐ญ๐ข๐จ๐ง ๐จ๐ ๐๐ซ๐๐๐ข๐๐ญ๐ข๐จ๐ง๐ฌ:
Prediction Visualization
