Through supervised machine learning, the AI system is essentially trained the way a kid would be taught with flash cards. The neural network, which mimics how our brains work, is shown an image of a confirmed malignant mass, then a benign one, then another malignant one, and so on (and on, and on). After millions of iterations—far more images than any human radiologist could see in a lifetime—these neural networks begin to recognize subtle patterns that might escape even the most experienced human eye. They detect tiny calcification clusters, specific tissue patterns, and subtle density changes. The training data is carefully curated and sourced from real-world reports and imagery that have been verified by pathology results. The AI system is continuously fed new data to improve accuracy. When a neural net is then shown a fresh ultrasound or X-ray image, it is able to make a quick evaluation based on all those millions of images it has seen before.

Joanna Stern, I Am Not a Robot- My Year U…, loc. 213