The knowledge of the total emitted radiation is essential for the understanding and control of tokamak plasmas and its relevance is going to be even higher in the next generation of metallic devices. The total radiation is measured with specific detectors called bolometers along defined collection volumes. The local emission from these integrated measurements is obtained with sophisticated tomographic algorithms, which are required to solve very ill-posed inversion problems. The maximum likelihood (ML) tomography is one of the most advanced techniques applied to the bolometric tomography in tokamaks. In this work the latest developments of the ML algorithm are overviewed. Firstly, a matrix formulation of the algorithm allows reducing the computational times of orders of magnitude, making the approach suitable to real time applications. Two distinct filtering techniques have been developed to regularize the solution and obtain physically meaningful results: one is optimised for real-time control, the other for offline analysis. Both versions are combined with an adaptive procedure, autonomously adjusting the filtering to the radiation patterns; this improvement reduces the requirements in terms of human intervention and contributes to the standardisation of the results. Furthermore, the uncertainty estimation provided by the ML for each voxel is improved and validated with systematic Monte Carlo simulations. The performances of the new algorithms are tested with synthetic data and compared with the main methods reported in the literature. The potential of the new algorithms is demonstrated with its application to the measurements of JET bolometry in discharges with a metallic wall, reconstructing the evolution of the emitted radiation in phenomena such as MARFE, temperature hollowness and disruptions.