SpectraClassifier 1.0: a user friendly, automated MRS-based classifier-development system

BMC Bioinformatics

Table 1 Comparing results for the validation of SC, using PCA prior to LDA.

	Long TE - [20]	Long TE - SC	Short TE - [21]	Short TE - SC
Classes	AUC ± SE	AUC ± SE	AUC ± SE	AUC ± SE
1 vs. 2	0.953 ± 0.031 ₍₈₎	0.977 ± 0.016 ₍₈₎	0.956 ± 0.028 ₍₄₎	0.923 ± 0.028 ₍₄₎
1 vs. 3	0.593 ± 0.104 ₍₆₎	0.757 ± 0.054 ₍₆₎	0.591 ± 0.097 ₍₄₎	0.688 ± 0.056 ₍₄₎
1 vs. 4	0.918 ± 0.063 ₍₇₎	0.941 ± 0.025 ₍₇₎	0.966 ± 0.029 ₍₃₎	0.962 ± 0.019 ₍₃₎
2 vs. 3	0.961 ± 0.038 ₍₅₎	0.970 ± 0.028 ₍₅₎	0.954 ± 0.044 ₍₄₎	0.972 ± 0.021 ₍₄₎
2 vs. 4	0.931 ± 0.073 ₍₁₀₎	0.999 ± 0.003 ₍₁₀₎	0.997 ± 0.009 ₍₁₁₎	1.000 ± 0.000 ₍₁₁₎
3 vs. 4	0.961 ± 0.053 ₍₄₎	0.995 ± 0.010 ₍₄₎	0.986 ± 0.025 ₍₂₎	0.979 ± 0.025 ₍₂₎

In this example, multiple binary classifiers were developed for long and short TE of SV MRS data from INTERPRET [5]. In [20, 21] the PCA covered the 75% of the variance of the dataset. For SC, a variance of 80% has been covered. As performance measure, the AUC and its standard error (SE) were used. The number between brackets refers to the number of principal components used. The classes are: 1) glioblastomas, 2) meningiomas, 3) metastases and 4) astrocytomas grade II.

ISSN: 1471-2105