Convolutional Neural Networks (CNNs) have gained increasing importance in healthcare, particularly for image diagnosis. Ensuring high accuracy and precision in these models is paramount for their efficacy in healthcare applications, and hyperparameters’ determination, such as batch size, plays a pivotal role in shaping their network performance. However, defining the optimal batch size value represents a challenge, especially when dealing with imbalanced datasets, a common issue in healthcare studies. This study aimed to investigate the impact of batch size variation on the performance of ResNet-50 models in classifying imbalanced histopathological data. Employing batch sizes as a set of power-of-2 values, the research revealed that minimal batch sizes failed to effectively learn patterns, while higher values disproportionately affected imbalanced classes. Unexpectedly, a batch size of 4 demonstrated superior performance compared to commonly recommended values. In conclusion, optimizing hyperparameters like batch size is complex and emblematic of the amount and type of data, but is pivotal in achieving satisfactory performance in CNNs, particularly in healthcare applications. Future research will delve into strategies such as data augmentation for smaller classes and integrate novel models from the literature, focusing on ensemble methods. These efforts aim to enhance the robustness and effectiveness of CNNs in healthcare contexts.