The main aim of this work is to detect metabolites in the brain by means of nuclear magnetic resonance spectroscopy. We were mostly interested by glutathione (GSH) and more particularly by its cysteine part which forms a strongly coupled ABX spins system. As the concentration of this molecule is very low in the brain, and as its spectrum is a multiplet, it is impossible to detect it by standard spectroscopy sequences. Consequently, we have implemented and optimized a Double quantum Coherence Filtering sequence (DQCF) which eliminates all singlets and can select the multiplet of GSH. This sequence is a symmetric PRESS (Point-Resolved Spectroscopy) sequence with two non slice-selective π/2 pulses and two extra filtering gradient pulses. For PRESS and DQCF sequences, we have noticed important signal losses. We have determined that the main origins of these losses are related to the shape of the selective π pulses of the PRESS sequence and especially to the DQCF filtering gradients of the DQCF sequence. These gradients create Eddy current which decreases the effectiveness of the sequence. Furthermore they generate losses through convection mostly but diffusion also. We have estimated the total losses to be of about 60% . Those could be strongly decreased by applying very weak filtering gradients. In this case however, spurious signals can completely mask the spectral range of interest. We have determined the origin of these spurious signals which is different in vitro and in vivo. The in vitro signals are mainly double quantum signals created by the dipolar coupling between protons of water molecules. The in vivo signals are mainly five-pulse stimulated echoes created by two consecutive shots. We have developed methods to remove them. In spite of these losses, our DQCF sequence can detect around 1 mM of GSH in vitro. The GSH concentration in the brain is slightly higher and should be detected with this sequence. However the in vivo conditions are more difficult than the ones in vitro: the field's homogeneity is definitely worse and the molecules relaxation times are much shorter. Our first in vivo results show that the DQCF sequence is not appropriate for the GSH detection in vivo. The difference editing would be probably more effective. In a second part, this work deals with heating effects of a wire during RF excitation. We have shown that this effect depends on the wire's length and is maximum at resonant length.