Table 2.

Classification results

A
Serum training cohort
	CM10			Q10
	accuracy	sens	spec	accuracy	sens	spec
DLDA	81	75	86	73	71	75
SVM	79	75	83	71	69	73
Serum validation cohort
DLDA				73	88	62
SVM				81	81	81
B
Tissue predicted (%)
	Benign		LMP		Cancer
Benign (true)	82.1		17.6		0.3
LMP (true)	10.6		64.5		24.9
Cancer (true)	3.4		10.6		86.0

A. Classification of serum samples on the training and independent validation cohort. Training cohort: average classification accuracy, sensitivity (sens), and specificity (spec) (in percentage) of discriminating ovarian cancer versus benign tumor for CM10 and Q10 arrays on 500 test sets (repeated random sampling; size training sets: 80, size test sets: 47 for CM10, 48 for Q10). Validation cohort: classification accuracy, sensitivity, and specificity (in percentage) of the classifiers trained on all serum training data for Q10 validation data only. Classification models: SVM (support vector machine) and DLDA (diagonal linear discriminant analysis) with feature selection.

B. Confusion matrix giving the percentage of cases (average over 500 test sets) from one class classified into each of the three classes. Rows correspond to the correct class, columns to the predicted class. Results are for DLDA with three different classes: benign tumors, tumors of low malignant potential (LMP) and cancer tissue (Q10 data).