Skip to content

Initial Release

Latest
Latest
Compare
Choose a tag to compare
@harrycallahan117 harrycallahan117 released this 18 Sep 22:11
· 4 commits to master since this release
v1.0.0
f05af5a

Release Notes: Tumor Classification CNN Model

Version: v1.0.0
Title: Initial Release

Overview:
This release includes a Convolutional Neural Network (CNN) model designed for brain tumor classification using MRI scans. The model was built using TensorFlow and Keras, and is optimized for faster training through mixed precision. It leverages image augmentation and normalization techniques for enhanced model performance.

Key Features:

  • CNN Architecture:

    • Three convolutional blocks with batch normalization and dropout to prevent overfitting.
    • Fully connected layers for classification using softmax activation.
    • Supports multi-class classification.

  • Optimizations:

    • Mixed precision enabled for faster training.
    • ReduceLROnPlateau callback for adaptive learning rate adjustments.
    • ModelCheckpoint for saving the best model based on validation accuracy.

  • Data Augmentation:

    • Normalization (rescale=1./255) and data augmentation with ImageDataGenerator to improve generalization.
    • Training and validation splits to ensure robust evaluation.

Model Performance:

  • The model architecture has been fine-tuned to balance accuracy and efficiency.
  • Optimizer: Adam optimizer with a learning rate of 0.0001.
  • Training: The model was trained for 10 epochs using MRI scan images resized to 224x224 pixels.

Files Included:

  • new_model_saved/: The final saved model in TensorFlow SavedModel format.
    • This directory includes the optimized model for further inference.

How to Use:

  1. Download the Model:

    • Download the new_model_saved directory containing the trained model.
    • Copy the 'variables.data-00000-of-00001' file into the folder 'mri-tumor-analysis\backend\new_model_saved\variables'.

  2. Load the Model:

    • Use TensorFlow or Keras to load the model in your environment: 
      from tensorflow.keras.models import load_model
model = load_model('path_to_new_model_saved')

  3. Inference:

    • Pass MRI images of size (224, 224, 3) to classify tumor types.

Future Enhancements:

  • Fine-tuning with more epochs and larger datasets.
  • Further optimization for deployment in real-time healthcare environments.
Assets 3
Loading