Galaxy clusters are good tracers of the large scale distribution of matter in the Universe. Consequently, they are a privileged probe for cosmological models. They can be observed through several processes: galactic emissions, bremsstrahlung, gravitational lensing, Sunyaev-Zel'dovich effects. The thermal Sunyaev-Zel'dovich (tSZ) effect has proven to be a efficient way to detect high-z and high-mass galaxy clusters (see e.g., Planck, ACT, SPT galaxy cluster catalogs). First, I will present a method, MILCA, optimized for the extraction of the tSZ effect from multi-frequency observations of the microwave sky. Then, I will discuss the advantages and drawbacks of such reconstructed tSZ emission. Finally, I will present a validation method based on artificial neural network that have been used inside the Planck collaboration to build a high-purity galaxy cluster sample. The production of such high-purity galaxy cluster sample is a key step in order to set robust constraints on cosmological models.