Bridging the gap in cardiovascular magnetic resonance imaging artificial intelligence implementations: From ambitious goals to real-world progress using foundation models.

Many improvements did not reach statistical significance, such as mapping segmentation (gains of only 0.1–0.2 percentage points), LGE segmentation (gains of 0.1–1.8 percentage points), both LGE hyperenhancement detection tasks (despite 3.7–6.6 percentage points gains), and landmark localization (where one landmark improved by 0.1 mm while the other worsened by 0.1 mm).

also in 111,027 other papers