Osteosarcoma Histopathology Eval

osteosarcoma-histopathology-eval

Advanced Hybrid Deep Learning Model for Enhanced Classification of Osteosarcoma Histopathology Images — Borji et al. (2024) (arXiv:2411.00832, 2024)

What this evaluates

Evaluates deep learning models on classifying osteosarcoma histopathology images into non-tumor, non-viable tumor, viable tumor, and non-viable ratio categories without prior segmentation. Probes the model's ability to capture local texture and global spatial patterns for medical image classification.

Datasets

• TCIA Osteosarcoma — total ?; splits: test (-1), val (-1)

Metrics

• accuracy (primary) — range: [0, 1]
  • Proportion of correctly predicted labels among all predictions.
• precision — range: [0, 1]
  • Ratio of true positive predictions to the total number of positive predictions.
• recall — range: [0, 1]
  • Ratio of true positive predictions to the total number of actual positives.
• F1-score — range: [0, 1]
  • Harmonic mean of precision and recall.

Input / output format

Input: H&E-stained histopathology image slides of osteosarcoma tissue.

Output: Class label: binary (VT vs NT), ternary (VT vs NVT vs NT), or quaternary (VT vs NVT vs NT vs NVR).

Scoring recipe

def compute_metrics(y_true, y_pred):
    accuracy = (y_true == y_pred).mean()
    precision = precision_score(y_true, y_pred, average='macro')
    recall = recall_score(y_true, y_pred, average='macro')
    f1 = f1_score(y_true, y_pred, average='macro')
    return accuracy, precision, recall, f1

Common pitfalls

• The dataset does not require prior segmentation, unlike many prior works that used segmentation techniques.
• Performance varies significantly across classification complexities (2-class vs 4-class), with CNNs dropping in the 4-class task while ViT maintains stability.
• Validation and test accuracies are reported separately; readers must distinguish between them when comparing results.

Evidence (verbatim from paper)

Among the tested models, the CNN + ViT hybrid model demonstrated the best performance across all tasks. In the four-class classification, which is the most challenging task, the hybrid model again outperformed all others, with 99.08% test accuracy and 99.70% validation accuracy.

Citation

@misc{borji2024advanced,
  title={Advanced Hybrid Deep Learning Model for Enhanced Classification of Osteosarcoma Histopathology Images},
  author={Borji et al. (2024)},
  year={2024},
  note={arXiv:2411.00832}
}

• arXiv: 2411.00832