Analysis
(quantitation)

Training

• Acquire a set of 100–200 samples with known levels for the analyte(s) of interest.
• Use a partial least-squares algorithm to derive quantification algorithm(s) used to relate spectra to analyte level(s).

Test

• Use quantitation algorithm(s) optimized for training samples to predict analyte level(s) for independent set of test samples.
• Compare predicted analyte level(s) to true values (acquired by standard analytical methods).

Diagnosis
(classification)

Training

• Acquire a set of 100–200 samples with known disease states. The samples should include both diseased and control specimens.
• Use pattern identification software to discover an algorithm that optimally distinguishes disease from control spectra.

Test

• Apply classification algorithm to predict diagnoses based upon an independent set of test samples.
• Compare predicted diagnoses to true diagnoses.

Copyright ©2003 IVD Technology