Traditional software follows instructions written by programmers: if this, then that. Machine learning is different. Instead of receiving the rules, a model is shown thousands of examples and finds the patterns itself.
In breast imaging, that means showing the model tens of thousands of mammograms — each labelled normal or cancerous, ideally with biopsy confirmation — until it learns the subtle visual features that separate the two. What emerges is not a rule but a statistical understanding of what cancer looks like.
The core idea
- Learning from examples, not hand-coded rules
- Finding statistical patterns in large datasets
- Generalising: performing well on unseen images
- Improving with better data and methods