Brain Tumor Cnn Eval

brain-tumor-cnn-eval

Brain Tumor Detection Through Diverse CNN Architectures in IoT Healthcare Industries: Fast R-CNN, U-Net, Transfer Learning-Based CNN, and Fully Connected CNN — Asghari Ilani et al. (2025) (arXiv:2509.05821, 2025)

What this evaluates

Evaluates the ability of various CNN architectures (custom, U-Net, Fast R-CNN, and transfer learning models) to accurately classify brain tumors (glioma, meningioma, pituitary) from MRI images. It probes architectural robustness, generalization across data splits, and performance under class imbalance conditions.

Datasets

• Kaggle Brain Tumor Dataset — total ?; splits: train (-1), val (-1), test (-1)

Metrics

• accuracy (primary) — range: percent
  • Percentage of correctly classified instances out of the total number of instances. Computed as (True Positives + True Negatives) / Total.
• f1-score — range: percent
  • Harmonic mean of precision and recall, computed per class and averaged (weighted).
• precision — range: percent
  • Ratio of true positive predictions to all positive predictions for each class.
• recall — range: percent
  • Ratio of true positive predictions to all actual positive instances for each class.
• auc — range: percent
  • Area under the Receiver Operating Characteristic curve, measuring the model's ability to distinguish between classes across all classification thresholds.

Input / output format

Input: Grayscale MRI images resized to 270×270 pixels.

Output: Softmax class probabilities for three tumor categories (glioma, meningioma, pituitary).

Scoring recipe

def compute_metrics(y_true, y_pred, y_proba):
    acc = (y_true == y_pred).mean() * 100
    prec = precision_score(y_true, y_pred, average='weighted') * 100
    rec = recall_score(y_true, y_pred, average='weighted') * 100
    f1 = f1_score(y_true, y_pred, average='weighted') * 100
    auc = roc_auc_score(y_true, y_proba, multi_class='ovr') * 100
    return {'accuracy': acc, 'precision': prec, 'recall': rec, 'f1-score': f1, 'auc': auc}

Common pitfalls

• The paper uses both a fixed 80/20 train-test split and 5-fold stratified cross-validation without clearly specifying which reported metric values correspond to which protocol.
• Class weights are mentioned for handling imbalance but the exact weighting scheme or formula is not provided, making exact reproduction difficult.
• The external cross-dataset validation excludes the 'normal' class, making direct comparison with the main 3-class results invalid.

Evidence (verbatim from paper)

The evaluation utilized a standardized dataset split into training (80%), validation (stratified cross-validation with k = 5), and testing (20%) subsets. The study also presents the accuracy, F-score, recall, precision, and AUC results obtained from these models in the study.

Citation

@misc{asghariilani2025braintumor,
  title={Brain Tumor Detection Through Diverse CNN Architectures in IoT Healthcare Industries: Fast R-CNN, U-Net, Transfer Learning-Based CNN, and Fully Connected CNN},
  author={Asghari Ilani et al. (2025)},
  year={2025},
  note={arXiv:2509.05821}
}

• arXiv: 2509.05821