Reporting on deep learning algorithms in health care

AI Generated Summary*

Commentary arguing that deep learning papers in health care lean too heavily on a narrow set of evaluation metrics. For continuous outcomes such as blood pressure, mean absolute error and scatter plots cannot reveal direction or proportional bias, so Bland-Altman plots and root mean square error are proposed as additions. In an example with diastolic pressure, mean absolute error was 7.54 mm Hg without outliers and 7.46 with them, while RMSE was 9.58 and 10.30. For binary outcomes, a dataset with 5% disease gave AUROC 0.957, sensitivity 98.6%, specificity 89.8%, yet area under the precision-recall curve was 0.348. Predictive values and a summary table of complementary methods are recommended.

*This summary was generated by AI and is published unedited. Oku does not alter these summaries. It may contain errors or omissions and is provided for general informational purposes only. Oku does not guarantee its accuracy, completeness, or reliability. For authoritative information, please refer to the original, peer-reviewed article.