Quadratic discriminant analysis in anatomical research: A scoping review and critical methodological appraisal of morphometrics, shape analysis and tissue classification

Background Quadratic discriminant analysis (QDA) accommodates class-specific covariance structures but may be unstable with small samples, high dimensionality or biologically dependent observations. Purpose and review questions This review evaluated anatomical applications of QDA, their limitations, and when classical or stabilised forms may be appropriate. Scope and review approach Database and citation searches identified 766 records; 101 were screened and 29 full texts assessed. Twelve publications met the eligibility criteria: eight direct anatomical applications and four methodological, comparative or contextual publications. Principal methodological findings Applications covered skeletal, morphometric, cytomorphometric, tissue and imaging data. The heterogeneous evidence did not show consistent superiority of QDA over LDA or regularised alternatives. Recurrent limitations included small class-specific samples, unstable or incompletely reported covariance estimation, possible preprocessing leakage, inadequate control of biological dependence and limited external validation. None of the eight direct applications reported calibration as a principal outcome. Key recommendations QDA should be treated as a candidate rather than a default classifier. Preprocessing and tuning should occur within resampling, validation should preserve the independent biological unit, and QDA should be compared with simpler and regularised alternatives. Conclusion QDA may be useful, but credible anatomical applications require stable covariance estimation and biologically appropriate validation.