qhjqhj00/research-skills-pool/alzheimer-mri-4class-eval

alzheimer-mri-4class-eval

An Efficient Medical Image Classification Method Based on a Lightweight Improved ConvNeXt-Tiny Architecture — Xia et al. (2025) (arXiv:2508.11532, 2025)

What this evaluates

Evaluates multi-class classification performance on Alzheimer's disease MRI scans to assess a model's ability to distinguish between different stages of dementia and healthy controls under resource-constrained hardware conditions.

Datasets

• Alzheimer MRI 4 Classes Dataset — total ?; splits: train (-1), val (-1), test (-1)

Metrics

• Accuracy (primary) — range: [0, 1]
• Proportion of correctly classified instances out of the total number of instances.
• F1-score — range: [0, 1]
• Harmonic mean of precision and recall: 2 (precision recall) / (precision + recall). Reported per class and macro-averaged.
• AUC — range: [0, 1]
• Area under the Receiver Operating Characteristic curve, measuring the trade-off between true positive rate and false positive rate across classification thresholds.

Input / output format

Input: MRI image scans of Alzheimer's disease patients.

Output: Predicted class label from four categories: Non Demented, Mild Demented, Moderate Demented, Very Mild Demented.

Scoring recipe

def compute_metrics(y_true, y_pred, y_proba):
    accuracy = np.mean(y_true == y_pred)
    f1 = f1_score(y_true, y_pred, average='macro')
    auc = roc_auc_score(y_true, y_proba, multi_class='ovr')
    return {'accuracy': accuracy, 'f1_score': f1, 'auc': auc}

Common pitfalls

• Different baseline models were trained with different learning rates and epoch counts, making direct performance comparisons potentially confounded by training schedule differences.
• Class imbalance is noted, particularly for the Moderate Demented class, which may skew accuracy and affect generalization metrics.
• Evaluation was conducted on CPU-only hardware, which impacts inference speed but the reported metrics focus solely on accuracy/F1/AUC.

Evidence (verbatim from paper)

The improved CN-T model achieved the best overall classification performance, with F1-scores ranging from 0.83 to 1.00.

Citation

@misc{xia2025efficient,
  title={An Efficient Medical Image Classification Method Based on a Lightweight Improved ConvNeXt-Tiny Architecture},
  author={Xia et al. (2025)},
  year={2025},
  note={arXiv:2508.11532}
}

• arXiv: 2508.11532