Deep learning accurately and reliably segments pelvic vascular structure in CT scans of gynecologic cancer patients

Description

Hyeonseung Kim and ten colleagues teach a neural network to outline pelvic blood vessels on CT scans for cancer surgery planning, in a 2026 paper. Gynecologic operations thread through a thicket of iliac arteries and veins whose branches vary from person to person. Misreading that anatomy risks serious bleeding.

Using contrast scans from forty-seven patients, the team trained a leading medical segmentation framework to label nine vessel categories. Big vessels came out sharp: overlap scores above 0.83 for major arteries and 0.75 for veins. Smaller branches of the internal iliac system scored lower but stayed anatomically coherent and recognizable, notably the obturator and gluteal arteries. The authors call it the first systematic evaluation of branch-level pelvic vessel segmentation in this patient group. They point to safer planning, teaching, and surgery as the payoff.

Catalog record

Source: https://europepmc.org/articles/PMC13128585