Automated fetal MRI lung segmentation enables rapid total lung volume estimation across diverse pulmonary hypoplasia phenotypes

Automated fetal MRI lung segmentation achieved excellent agreement with manual volumetry for total lung volume (ICC 0.955, Dice 0.81, median error 11.1%) and took only 4 seconds per exam, potentially improving workflow for pulmonary hypoplasia assessment.

• Automated total lung volume (TLV) segmentation showed excellent agreement (ICC 0.955, mean Dice 0.81, median absolute percentage error 11.1%).
• Percent predicted lung volume (PPLV) segmentation had lower performance (ICC 0.727, mean Dice 0.76, median error 14.5%).
• Human-in-the-loop correction of the 25 worst TLV outputs averaged 3.1 minutes per case, enabling oversight for challenging cases.

Automated summary

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