Establishing an accurate radiological mapping is an essential step for the Dismantling \& Decommissioning (D\&D) processes since it allows the localization of possible hotspots. Nowadays, this work is usually done manually by operators who construct matrices and place each measured value in its respective position. However, this method presents many drawbacks, leading the literature to increasingly explore alternative solutions, such as combining radiological measurements with SLAM techniques (where SLAM stands for Simultaneous Localization and Mapping). SLAM technology allows the simultaneous mapping of the surroundings and indoor localization of a sensor without needing a GPS (usually inoperative in indoor nuclear facilities). Nonetheless, the current solutions based on SLAM algorithms are not suited to different nuclear measurement devices, are often bulky, and require post-processing. To address this matter, we started the development of a modular device for online 3D environment reconstruction and radioactivity measurement localization. This work presents the selection of the most suitable SLAM algorithm for the operation of nuclear installations under dismantling. As far as we know, this is the only study in the literature that evaluates the different SLAM algorithms regarding the nuclear facilities' characteristics. To address this issue, we established comprehensive state-of-the-art visual-sensor-based SLAM algorithms. Then, we critically assessed the algorithms from the literature regarding the characteristics present in dismantling environments. We selected and implemented five: Direct Sparse Odometry, Visual-Inertial Direct Sparse Odometry, Large Scale Direct Monocular SLAM, Semi-direct Visual Odometry, and Visual Inertial Semi-direct Visual Odometry. In order to compare them, we constructed a dataset not found in the literature representing the radiological mapping operations in nuclear facilities. Our dataset relies on stereo images from two cameras in a stereo configuration, inertial data from an Inertial Measurement Unit, and spherical information from a Lidar. This dataset allowed us to benchmark the algorithms considering algorithms' tracking and mapping accuracies. The Visual Inertial Semi-direct Visual Odometry presented the lowest average errors for the tracking and an equivalent performance as the other algorithms for the mapping, being the most suitable algorithm for the dismantling operation.