<p>Deep learning has advanced medical image classification by improving both diagnostic accuracy and efficiency. While most AI systems are designed for a single disease, this paper presents a multi-disease detection framework based on convolutional neural networks (CNNs) capable of analyzing diverse imaging modalities. The system adopts a two-model approach: the first targets radiology-based diseases such as breast cancer, brain tumours, polycystic ovary syndrome (PCOS), pneumonia, and tuberculosis using X-ray, MRI, CT, and ultrasound images; the second addresses dermatological conditions using skin images. Multiple CNN architectures were explored, with DenseNet-121 showing the best performance for radiology classification. For skin disease analysis, EfficientNet-B0 and ResNet-152 were evaluated, with EfficientNet-B0 providing the most reliable results during training. Data augmentation, normalization, and dataset balancing were applied to enhance model robustness. The proposed system demonstrates strong diagnostic potential and highlights the benefits of combining multiple CNN architectures for multi-disease detection. However, limitations remain: reliance on publicly available datasets, restricted diversity of clinical cases, and limited external validation, particularly for dermatological conditions. These factors raise concerns about overfitting and generalizability. Overall, this work shows promise for AI-assisted diagnostics but emphasizes the need for larger, multi-institutional validation before clinical deployment.</p>