Diagnostic Threshold Identification for Equine Laminitis Using Smoothed Receiver Operating Characteristic Analysis

Area of Science:

• Veterinary Radiology
• Equine Medicine
• Biostatistics

Background:

• Radiographic measurement parameters are crucial for diagnosing equine laminitis.
• Current diagnostic thresholds often rely on empirical receiver operating characteristic (ROC) analysis, which can be unstable with small or noisy datasets.

Purpose of the Study:

• To evaluate the effectiveness of nonuniform rational B-spline (NURBS)-based ROC smoothing for refining radiographic diagnostic thresholds in horses.
• To compare NURBS-derived thresholds against conventional empirical thresholds for laminitis detection.

Main Methods:

• Application of NURBS-based ROC smoothing to radiographic data from laminitic and healthy horses.
• Determination of diagnostic thresholds using Youden's index on smoothed ROC curves.
• Comparison with empirical thresholds derived from the index of union criterion.
• Assessment of diagnostic performance using area under the curve (AUC), sensitivity, specificity, and threshold agreement, with bootstrap resampling for confidence intervals.

Main Results:

• NURBS-smoothed ROC analysis yielded thresholds that matched or exceeded conventional methods in classification accuracy across multiple radiographic variables.
• NURBS-derived thresholds demonstrated reduced misclassification rates and better alignment with clinical diagnoses.
• The use of smoothing and resampling techniques enhanced the statistical reliability of the findings.

Conclusions:

• NURBS smoothing is a reliable complementary tool for improving the consistency of radiographic threshold selection in equine laminitis detection.
• This technique offers a more stable and accurate approach compared to traditional empirical methods, particularly for challenging datasets.