Machine learning describes a family of algorithms capable of identifying complex patterns without explicit human instruction. Classical methods work well with modest datasets, while deep neural networks excel with larger volumes and can process images, text, or clinical data. Applications in medicine now include image-based diagnosis, automated documentation, and decision support, with research showing performance that can parallel or exceed clinician accuracy. Training relies on iterative error correction through back-propagation. Challenges persist, including fragmented and unstructured data, concerns about interpretability, and the need for transparent model explanations such as LIME and saliency maps. As datasets grow and computational tools become more accessible, prospective trials are beginning to test whether real-time AI assistance can improve clinical outcomes.