Brain-MRI-Tumor-Classification-Using-CNN

A test run on the dataset. Tasks- Image Augmentation, Feature Map, High Evaluation Metrics, Accuracy Graph

Experimental Setup:

• GPU: NVIDIA GeForce RTX 3080 Ti
• Compute Capability Memory: 8.6
• FP32 (float) Performance: 34.10 TFLOPS
• CUDA Version: 11.2
• TensorFlow(GPU) Version: 2.6.0
• Total Training Time: 1 Minute

Model Type

• CNN
• Layers Used: Conv2D, BatchNormalization, MaxPooling2D, Dropout, Flatten, Dense.
• Learning Rate: 0.0001
• Activation Layer: ReLu, Sigmoid(Final Classifier Layer)
• Dropout: 15%
• Parameters: 10,150,626
• Number of Epochs: 30

Dataset

• Real Dataset link: https://www.kaggle.com/navoneel/brain-mri-images-for-brain-tumor-detection
• Used Keras' ImageDataGenerator for augmentation
• Increased Data Size: 2007
• Image Type: Grayscale

Evaluation Metrics:

• Accuracy: 0.9900
• AUC(Area Under ROC Curve): 0.9998
• Recall: 0.9955
• Precision: 0.9913
• F1: 0.9934

Train vs Validation - Accuracy & Loss Graph:

Feature Map: