Poate AI reconstrui structuri osoase 3D din imagini radiografice standard ?

Imagistica medicală se bazează adesea pe scanările CT pentru reconstrucții 3D detaliate, însă acestea sunt costisitoare și expun pacienții la radiații mai mari. Radiografiile standard sunt mai accesibile, dar lipsesc informațiile de profunzime. Algoritmii AI ar putea, în teorie, deduce modele 3D ale oaselor din radiografii 2D, îmbunătățind precizia diagnosticului fără a necesita imagini suplimentare.

Background

Medical imaging often relies on CT scans for detailed 3D reconstructions, but these are costly and expose patients to higher radiation. Standard X-rays are more accessible but lack depth information. AI algorithms could potentially infer 3D bone models from 2D X-rays, improving diagnostic accuracy without additional imaging.

Current AI systems can reconstruct coarse 3D bone shapes from two or more standard X-ray images by using deep-learning models trained on large datasets of paired X-ray and CT volumes. Accuracy is highest for dense cortical bone and decreases for trabecular bone and small features, and the approach is primarily used for surgical planning and follow-up rather than definitive diagnostics. Research prototypes show promise for single-view methods under limited angles, yet these still lag behind multi-view accuracy and require specialized calibration.

— Enriched May 12, 2026 · Source: Radiological Society of North America (RSNA)