Signal Detection Theory has become the standard framework for evaluating diagnostic tests and clinical decision making. The SDT framework maps directly onto medical diagnosis: "signal" is disease present, "noise" is disease absent, "hits" are true positives, and "false alarms" are false positives. This mapping allows rigorous separation of a test's discriminative accuracy from the clinician's decision threshold.
Diagnostic Accuracy Measures
Specificity = P(negative test | no disease) = 1 − false alarm rate
PPV = P(disease | positive test)
NPV = P(no disease | negative test)
AUC = area under ROC = overall diagnostic accuracy
Clinical ROC Analysis
ROC curves in medicine compare diagnostic tests by plotting sensitivity against 1−specificity across different decision thresholds. A test with a higher AUC is better at discriminating diseased from healthy patients. The optimal threshold depends on disease prevalence and the relative costs of false positives versus false negatives — a distinction SDT handles naturally. For cancer screening, where missing a cancer (false negative) is costly, the threshold should be liberal; for invasive surgical decisions, where false positives lead to unnecessary surgery, the threshold should be conservative.
Modern medical applications include radiological image reading, pathology screening, dermatological assessment, and algorithmic diagnostic tools where SDT analysis helps calibrate human and machine performance.