Image classification, the process of categorizing images into predefined classes, plays a vital role in everyday life and various computer applications. For individuals, it enhances daily experiences through features like facial recognition in security systems and smartphone photo organization, where algorithms automatically categorize images based on content, making it easier to search and manage large libraries. Convolutional Neural Networks (CNNs) fall into the category of deep neural networks which are especially effective for medical imaging and image classification problems due to their architecture, which is particularly suited to processing pixel data. This paper explores the critical influence of CNN hyperparameters on training performance. By systematically adjusting key hyperparameters namely the count of convolutional layers, filter sizes, and learning rates, we demonstrate how these variables affect the model's ability to detect and classify pathological features in three general public datasets like MNIST, Fashion MNIST and CIFAR-10 and one medical imaging dataset named Chest X-Ray.