Multimodal imaging has significantly transformed biomedical and scientific research, providing profound insights into the study of biological samples. However, imaging modalities often encounter inherent noise and artifacts owing to the statistical nature of photon detection, which compromises image quality and impedes correct analysis. Traditional methods lack scalability and exhibit suboptimal performance in denoising complex structures. In recent years, deep learning (DL) has made substantial progress in image denoising. However, the training of DL models requires extensive datasets and may yield inaccurate results when the data are limited, posing challenges in areas such as medical diagnostics. To address the challenges of standard DL methods, we propose a physics-inspired DL model, specifically a modified incSRCNN with a physics-inspired loss function, designed to enhance denoising performance with limited datasets. Our method outperforms the conventional incSRCNN model, representing a significant advancement in denoising capabilities without the need for large datasets. The proposed methodology not only improves the quality of image denoising but also reduces the dependence on extensive datasets, making it particularly valuable in biomedical applications and diagnostics, where large data collection is challenging or constrained.