Fig. 7
(a) Confusion matrix for the training set. The model shows high class-wise accuracy with minor confusion between glioma and pituitary, reflecting their shared grayscale and boundary features. Diagonal dominance confirms strong initial learning. (b) Confusion matrix for the test set. Maintains high accuracy with sparse misclassifications, particularly between glioma and meningioma—consistent with real-world morphological overlaps in MRI. (c) Confusion matrix for the validation set. Performance is consistent with the training and test sets, confirming generalization and robustness. Misclassifications are minimal and class-specific patterns are well preserved.
