Covid19 Cxr Detection Eval

---
name: covid19-cxr-detection-eval
description: Evaluates a deep CNN's ability to classify chest X-ray images into COVID-19 positive and negative/healthy categories using region-edge and channel-boosted features. Use when the user wants to benchmark on Three datasets (names not specified in section), or asks about evaluating this task. Reports 95% Confidence Interval (CI).
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2012.05073
  bibtex_key: [REDACTED]
  confidence: medium
---

# covid19-cxr-detection-eval

> COVID-19 Detection in Chest X-Ray Images using a New Channel Boosted CNN — Saddam Hussain Khan et al. (2020) (arXiv:2012.05073, 2020)

## What this evaluates

Evaluates a deep CNN's ability to classify chest X-ray images into COVID-19 positive and negative/healthy categories using region-edge and channel-boosted features.

## Datasets

- **Three datasets (names not specified in section)** — total ?; splits: train (-1), val (-1), test (-1); repo https://github.com/PRLAB21/COVID-19-Detection-System-using-Chest-X-Ray-Images

## Metrics

- `95% Confidence Interval (CI)` **(primary)** — range: other
  - Statistical confidence interval computed over test-set evaluation results to quantify uncertainty around the primary performance metric.

## Input / output format

**Input**: Chest X-ray images resized to 224×224×3 pixels.

**Output**: Softmax class probabilities for COVID-19 vs. non-COVID/healthy classes.

## Scoring recipe

```python
predictions = model.predict(test_images)
labels = test_labels
accuracy = np.mean(predictions == labels)
# Compute 95% CI using standard error or bootstrap as per references [52], [53]
ci_lower, ci_upper = compute_95_ci(accuracy, n_bootstrap=1000)
return {'accuracy': accuracy, '95%_CI': (ci_lower, ci_upper)}
```

## Common pitfalls

- The paper uses the contradictory term 'Holdout cross-validation scheme' instead of a standard holdout or k-fold split.
- Data augmentation details (rotation angles, shear factors) are not specified, making exact reproduction difficult.
- Dataset names and exact sample counts are omitted from the experimental setup section.

## Evidence (verbatim from paper)

> Dataset was divided into train and test set with the ratio of 80:20%. From training dataset, 20% is reserved for model validation and hyperparameter selection. The final evaluation of the model was made on the test set, which was kept separate from training and validation dataset... For each of the CNN model, 95% confidence interval (CI) was computed [52], [53].

## Citation

```bibtex
@misc{khan2020covid19detection,
  title={COVID-19 Detection in Chest X-Ray Images using a New Channel Boosted CNN},
  author={Saddam Hussain Khan et al. (2020)},
  year={2020},
  note={arXiv:2012.05073}
}
```

- arXiv: 2012.05073