AI diagnoses spine problems from MRI scans with 90% accuracy and explains its reasoning

Researchers built an AI system that reads MRI scans to diagnose lumbar spinal stenosis (a common spine problem) and generates written explanations of what it found, mimicking how radiologists actually work. The system achieves 90% diagnostic accuracy while staying interpretable to doctors, meaning clinicians can see the AI's reasoning instead of treating it as a black box.

Medical imaging AI usually chooses between two bad options: work well but show no reasoning, or show reasoning but fail on difficult cases. This system handles both simultaneously by using control-theory math to force the AI to learn from rare, hard cases instead of just memorizing common ones. The real shift is that it generates actual radiology reports instead of just marking images, which means a doctor could theoretically use this as a second reader rather than a screening tool.