Anatomical studies in the plant model system Arabidopsis thaliana are vital for elucidating gene functions, such as those involved in solute transport, hormone signaling, and development in plant roots. In this research, we introduce SAMPLS, a novel approach that is built on a foundation model called the Segment Anything Model (SAM). We evaluate its zero-shot learning capability and prompt engineering to determine whether it can reduce the effort and time consumed in dataset annotation, facilitating a semi-automated fine-tuning process for segmentation of Arabidopsis root cells. Our proposed method improved the detection rate of cells and reduced the error rate as compared to PlantSeg, a state-of-the-art method for confocal image analysis in plants. We compared the IoU scores between our method and PlantSeg, highlighting the trade-off between segmentation accuracy and detection rate across different image qualities and data qualities. Our findings showed the efficiency of SAM in confocal image segmentation, demonstrated the potential of foundation models in specialized domains, and underscored the importance of tailored approaches for achieving accurate semantic segmentation in confocal imaging.