Radiology AI's evolution organized into five computational epochs, from CAD to reasoning

A conceptual review maps radiology AI into five epochs: handcrafted CAD, deep learning, global-context modeling, multimodal foundations, and inference-stage reasoning. The last remains preclinical.

• The five epochs are handcrafted CAD, deep and volumetric learning, global-context modeling and learned reconstruction, multimodal foundation models, and inference-stage reasoning.
• Epochs I–IV focus on representation or image formation; Epoch V shifts to adaptive case-specific inference, with radiology-specific evidence still largely preclinical.
• The framework proposes matching verification to participation level and characteristic failure modes, though later-stage verification requirements are only framework-derived.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.