Supervised Machine Learning
Most machine learning applications concern supervised learning ,
where a model is trained to predict a known outcome, called the target
variable, label or dependent variable. Supervised learning often
requires manual labeling of the target variable. Supervised learning can
be applied in almost all medical domains, such as disease diagnosis,
treatment outcome prediction, or classifications in medical imaging. One
such example is the FDA-approved Koios DS for Breast application.
This tool supports clinicians in breast cancer diagnosis by classifying
ultrasound images into benign, probably benign, suspicious, and probably
malignant. It has been shown to improve assessment performance compared
to the clinician’s assessment in a retrospective study. Supervised
machine learning has also surged in screening, predicting, contact and
tracing, and drug development. For example, during the COVID-19
pandemic, supervised ML was used to predict which potential drug
compounds could be effective against SARS-CoV-2 targets by developing
prediction models for the drug-likeliness of candidate compounds from
chemical libraries based on chemical descriptors. Within supervised ML,
gradient-boosted decision tree methods have been among the most popular
and performant, with the leading algorithms being Random Forest,
XGBoost, and LightGBM (Glossary).