Optimal classification of long echo timein vivo magnetic resonance spectra in the detection of recurrent brain tumors

Abstract

We describe the optimal high-level postprocessing of single-voxel (1)H magnetic resonance spectra and assess the benefits and limitations of automated methods as diagnostic aids in the detection of recurrent brain tumor. In a previous clinical study, 90 long-echo-time single-voxel spectra were obtained from 52 patients and classified during follow-up (30/28/32 normal/non-progressive tumor/tumor). Based on these data, a large number of evaluation strategies, including both standard resonance line quantification and algorithms from pattern recognition and machine learning, were compared in a quantitative evaluation. Results from linear and non-linear feature extraction, including ICA, PCA and wavelet transformations, and also the data from resonance line quantification were combined systematically with different classifiers such as LDA, chemometric methods (PLS, PCR), support vector machines and ensemble methods. Classification accuracy was assessed using a leave-one-out cross-validation scheme and the area under the curve (AUC) of the receiver operator characteristic (ROC). A regularized linear regression on spectra with binned channels reached 91% classification accuracy compared with 83% from quantification. Interpreting the loadings of these regressions, we find that lipid and lactate signals are too unreliable to be used in a simple machine rule. Choline and NAA are the main source of relevant information. Overall, we find that fully automated pattern recognition algorithms perform as well as, or slightly better than, a manually controlled and optimized resonance line quantification.

On Oblique Random Forests

“…A number of binary benchmark problems are used for the evaluation: binary classification problems of the UCI data repository (data sets 1-15, Table 1), synthetic data sets (16)(17)(18)(19)(20) [4], binary groupings of handwritten digits (21)(22)(23) from the MNIST data base, binary data from detection problems in archaeological remote sensing (24)(25) [26], from the analysis of magnetic resonance spectra for the detection of brain tumors or yeast infections (26)(27)(28)(32)(33)(34)(35)(36) [24,23], from the analysis of infrared spectra for BSE detection from blood serum and food analysis (29)(30)(31)(37)(38)(39)(40) [25,23].…”

Accurate classification of childhood brain tumours by in vivo 1H MRS – A multi-centre study

“…However, a multi-centre evaluation of these techniques is required. Although a large number of multi-centre studies on automatic classification of brain tumours has been reported in adults [8][9][10][11][12][13][14][15], these results cannot be extrapolated to children since the overall distribution of the tumour types, locations and etiology differs markedly from that of adults [16][17][18][19]. Establishing the optimal MRS acquisition protocol is important and MRS can potentially give accurate quantification of more metabolites by using a longer acquisition, which combines Short echo time (Short-TE) and Long echo time (Long-TE) MRS [20], but this has currently not been reported in pediatric brain tumours.…”

Pattern recognition approaches for biomedical data in computer-assisted cancer research

“…Furthermore, workflows including the compilation mechanism, quality control protocols, automatic pre-processing methods, high-level analysis, and model evaluation were also defined to manage the large volume of heterogeneous data. As a result, a CDSS for brain tumour classification based on SV MRS spectra was reported [74], in addition to a large improvement in PR methods for MRS analysis and brain tumour classification [75][76][77][78][79][80][81][82].…”