<p>Understanding neuronal morphology is crucial for advancing neuroscience and developing treatments for neurodegenerative diseases. However, accurate analysis is challenged by low-resolution imaging and uneven illumination, which significantly affect segmentation performance. In this study, a novel dataset of pyramidal neurons obtained from the rat prefrontal cortex (PFC-PN) is introduced. The data were collected at low resolution during patch-clamp experiments using inverted microscopy. The segmentation performance of convolutional neural networks (CNNs), specifically the U-Net, Attention U-Net, and Residual U-Net architectures, was evaluated. These models were assessed both with and without mask pre-processing to determine their impact on performance. The results indicate that improved segmentation in low-resolution images from electrophysiological experiments can be achieved using Convolutional Neural Networks (CNNs) that do not require large datasets. These findings address limitations identified in previous studies on live cell segmentation. A comparison of the Dice-Sørensen coefficient (DSC) measures revealed scores of 0.73 for U-Net, 0.74 for Attention U-Net, and 0.75 for Residual U-Net. Additionally, qualitative analysis showed that the Residual U-Net outperformed the others in preserving thin neuronal structures, such as dendrites. This study shows that using advanced CNN architectures alongside low-cost inverted microscopy is an effective method for accurately segmenting live neurons. This approach also addresses important issues such as phototoxicity, even when working with low-resolution images. Furthermore, the annotated dataset of PFC-PN images serves as a valuable resource for future research in neuronal morphology and the image analysis of live neurons.</p>