The role of neural networks in improving the accuracy of MR spectroscopy for the diagnosis of head and neck squamous cell carcinoma

摘要: BACKGROUND AND PURPOSE: MR Spectroscopy (MRS) has the unique ability to analyze tissue at molecular level noninvasively. The purpose of this study was determine if peak heights revealed by proton MRS ( 1 H-MRS) signals showed that neural networks (NN) provided better accuracy than linear discriminant analysis (LDA) in differentiating head and neck squamous cell carcinoma (SCCA) from muscle METHODS: In vitro 11-T H-MR spectra were obtained on SCCA samples (n = 16) 12). seven metabolite resonances measured: olefinic acids 5.3 ppm, inositol 3.5 taurine 3.4 choline (Cho) 3.2 creatine (Cr) 3.0 sialic acid 2.2 methyl 0.9 ppm. Using leave-one-out experimental design receiver operating characteristic curve analysis, NN LDA classifiers distinguish compared (given equal weighting false-negative false-positive errors). These also with an existing method forms a diagnosis using Cho/Cr area ratio. RESULTS: classifiers, which identified height data, achieved sensitivity specificity rates distinguishing SCAA did or data. Sensitivity/specificity for 87.5% 83.3%, respectively, one-hidden-node network 81.2% 91.7%, two-hidden-node network. Additional nodes not improve accuracy. 50%, heights, 68% 83%, CONCLUSION: data superior ratio normal muscle. results show can diagnostic H-MRS malignant tissue. Further studies are necessary confirm our initial findings.