MIDRC-MetricTree; Cancer diagnosis.
Finding methods and metrics for AI/ML performance evaluation.
Last updated July 1, 2024
There are many oncology diagnostic tasks for which AI/ML algorithms may be helpful and for which performance will need to be assessed thoroughly and correctly. Here you will find examples of tasks and which branch of the MIDRC-MetricTree to investigate, including links. These tasks include:
Distinction between benign and malignant findings identified on radiologic images (branch: classification -> binary classification)
Distinction between cancer sub-types of imaged known cancers, such as the classification of hormone receptor-positive HER2-negative vs. hormone receptor-positive HER2-enriched vs. triple negative breast cancer (branch: classification -> multi-class classification)
Estimation of BI-RADS (breast cancer), Lung-RADS (lung cancer), PI-RADS (prostate cancer) and other ordinal suspiciousness ratings (branch: estimation)