Microelectrode arrays (MEAs) permit recordings with high electrode counts, thus generating complex datasets that would benefit from precise neuronal spike sorting for meaningful data extraction. Nevertheless, conventional spike sorting methods face limitations in recognizing diverse spike shapes. Here, we introduce PseudoSorter, which uses self-supervised learning techniques, a density-based pseudolabeling strategy, and an iterative fine-tuning process to enhance spike sorting accuracy. Through benchmarking, we demonstrate the superior performance of PseudoSorter compared to other spike sorting algorithms before applying PseudoSorter on MEA recordings from hippocampal neurons exposed to subneuronal concentrations of monomeric Tau as a model for Alzheimer's disease. Our results unveil that Tau diminishes the firing rate of a subset of neurons, which complement our findings observed using conventional electrophysiology analysis, and demonstrate that PseudoSorter's high accuracy and throughput make it a valuable tool for studying neurodegenerative diseases, enhancing our understanding of their underlying mechanisms, as well as for therapeutic drug screening.