CNN_Model

---

               

---

Requirments

Python

Modules Imported

• numpy
• pandas
• tensorflow
• sklearn
• matplotlib

The Current model was trained using Kaggle GPU P100 Accelerator.

---

Dataset

API Command

kaggle datasets download -d voidexiae/nih-chest-x-ray-voidex

Link

https://www.kaggle.com/datasets/voidexiae/nih-chest-x-ray-voidex

---

How to train the Model using Kaggle?

Visit

https://www.kaggle.com/

Next,

• After Successfull Login,

• Click on Create -> New Notebook

• Click on Add Data -> NIH | Chest-X-Ray | Voidex (Search for this)

• Now, select add dataset

• After Successfully Adding dataset,

• Copy the code present in train.py

• Include it in the code section of the kaggle notebook.

• goto More -> Accelerator -> GPU P100

• Click on Run.

• Wait till all the Epoch are Completed and check for the HDF(.h5) file in the output folder of the kaggle Notebook.

Use test and validate

import both test.py and validate.py python file along with Voidex.h5 and run it using,

python test.py

python validate.py

---

About the Dataset

The entire dataset is divided into 3 category namely test, train, validate in the ratio 2 : 7 : 1

• Total Images : 1,75,245
• Test : 26, 448
• Train : 1,33,896
• Validate : 14, 901

Origin Source:

https://www.nih.gov/news-events/news-releases/nih-clinical-center-provides-one-largest-publicly-available-chest-x-ray-datasets-scientific-community

---

About the Model

Input Layer

The input images are expected to have a shape of (img_width, img_height, 3), representing RGB images.

Convolutional Layers

• The first convolutional layer has 32 filters with a kernel size of (3, 3) and applies the ReLU activation function.
• MaxPooling is applied with a pool size of (2, 2) to reduce the spatial dimensions of the output.
• The second convolutional layer has 64 filters with a kernel size of (3, 3) and applies the ReLU activation function.
• MaxPooling is applied again with a pool size of (2, 2).
• The third convolutional layer has 128 filters with a kernel size of (3, 3) and applies the ReLU activation function.
• MaxPooling is applied once more with a pool size of (2, 2).

Flattening Layer

The output of the last convolutional layer is flattened into a 1-dimensional vector to be fed into the dense layers.

Dense Layers

• The flattened vector is passed through a dense layer with 128 units and the ReLU activation function.
• The final dense layer has 18 units (corresponding to the 18 classes) and applies the sigmoid activation function.
• The sigmoid activation function is used because the task is multi-label classification, where each class can be present or absent independently.

Model Compilation

• The model is compiled with the Adam optimizer and the binary cross-entropy loss function, suitable for multi-label classification.
• The accuracy metric is also specified to monitor the performance during training.

Data Generators

• The code sets up two ImageDataGenerators, train_generator and validation_generator, which perform data augmentation and normalization by rescaling the pixel values to the range [0, 1].
• The generators are configured to load images from the respective directories (dataset_path + '/train' and dataset_path + '/validation').
• The target size for the images is set to (img_width, img_height).
• The class_mode is set to 'categorical', indicating that the labels are one-hot encoded.

Model Training

• The model is trained using the fit function.
• The train_generator is used to provide training data, with the number of steps per epoch set to train_generator.samples // batch_size.
• The training is performed for the specified number of epochs.
• The validation_generator is used to provide validation data, with the number of validation steps set to validation_generator.samples // batch_size.

Model Saving

After training, the model is saved to a file named 'Voidex.h5' using the save function.

Output

The Model is capable of Classifiying the Images into 18 Different classes Namely,

• Atelectasis X-001
• Brain_Tumor X-002
• Cardiomegaly X-003
• Consolidation X-004
• Edema X-005
• Effusion X-006
• Emphysema X-007
• Fibrosis X-008
• Hernia X-009
• Infiltration X-010
• Mass X-011
• No_Brain_Finding X-012
• No_Lung_Finding X-013
• Nodule X-014
• Pleural X-015
• Pneumonia X-016
• Pneumothorax X-017
• Tuberculosis X-018

---

Test Results

After Successfully testing the Deep Learning Model with 26,448 Image we obtained an Accuracy of 84.69%.

You can find the Entire Model Test Report Here

Confusion Matrix:

Classwise Accuracy:

---

License

---

Feedback

If you have any feedback, please reach out to us at voidex.developer@gmail.com