Proposed Workflow Diagram

Brain Tumor MRI Dataset
(raw images)

Data Splitting
(Train, Validation, Test)

Image Standardization
(Normalization, Resizing)

Data Augmentation
(Rotation, Zoom, Shift, Flip, etc.)

Input: Prepared Images

Base Model Selection
(e.g. InceptionV3)

Transfer Learning
(Freeze Base, Fine-Tune Head)

Extracted Features (F1 to Fn)
High-Dimensional Feature Vectors

Classification Head Architecture

Custom Dense
Layers + Dropout

Softmax Output Layer
4 Nodes: Glioma, Meningioma, Pituitary, No Tumor

Predicted Tumor Classes
(The Model's Prediction)

Predicted Classes

Ground Truth Labels
(Test Set)

Core Metrics,
Recall, F1-Score

Diagnostic Insights
(Accuracy/Loss, Confusion Matrix, ROC Curves, Error Analysis)

Visual Explanations

GradCAM
(Class Activation Mapping)

GradCAM++
(Enhanced Localization)

Interpretability Analysis
(Feature Importance, Attention Regions)