Histovit Cancer Classification Eval

histovit-cancer-classification-eval

HistoViT: Vision Transformer for Accurate and Scalable Histopathological Cancer Diagnosis — Ahmed et al. (2025) (arXiv:2508.11181, 2025)

What this evaluates

Multi-class histopathological image classification for cancer diagnosis across four tissue types (breast, prostate, bone, cervical). It probes the model's ability to extract robust morphological features from stained whole-slide image tiles without data augmentation.

Datasets

• ICIAR2018 — total 11794; splits: train (-1), test (-1)
• SIPAkMeD — total 4049; splits: train (-1), test (-1)
• SICAPv2 — total 11836; splits: train (-1), test (-1)
• UT-Osteosarcoma — total 10017; splits: train (-1), test (-1)

Metrics

• accuracy (primary) — range: [0, 1]
  • Fraction of correctly classified samples out of the total number of samples in the test set.
• precision — range: [0, 1]
  • Ratio of true positive predictions to the total number of positive predictions for each class, typically macro-averaged in multi-class settings.
• recall — range: [0, 1]
  • Ratio of true positive predictions to the total number of actual positives for each class, typically macro-averaged in multi-class settings.
• AUC — range: [0, 1]
  • Area Under the Receiver Operating Characteristic Curve, computed using a one-vs-rest strategy for multiclass classification.

Input / output format

Input: RGB histopathological image tiles/patches, resized to 224×224 pixels, with pixel values normalized to [0, 1].

Output: Class label prediction from a fixed set of C classes.

Scoring recipe

def compute_metrics(y_true, y_pred, y_prob, num_classes):
    accuracy = (y_true == y_pred).mean()
    precision = precision_score(y_true, y_pred, average='macro')
    recall = recall_score(y_true, y_pred, average='macro')
    auc = roc_auc_score(y_true, y_prob, multi_class='ovr', average='macro')
    return accuracy, precision, recall, auc

Common pitfalls

• No data augmentation is used, which is atypical for deep learning on small medical datasets and may limit generalizability.
• Split strategies vary by dataset (80:20, 70:30, 5-fold CV), making direct cross-dataset comparison difficult.
• AUC is computed using a one-vs-rest strategy for multiclass classification, not macro/micro averaging.

Evidence (verbatim from paper)

Compute test loss and accuracy metrics ... One-vs-rest AUC curve for one-vs-rest multiclass classification using the proposed model across three different cancer types

Citation

@misc{ahmed2025histovit,
  title={HistoViT: Vision Transformer for Accurate and Scalable Histopathological Cancer Diagnosis},
  author={Ahmed et al. (2025)},
  year={2025},
  note={arXiv:2508.11181}
}

• arXiv: 2508.11181