Liver segments on CT

Liver segments on CT are the eight Couinaud units — each with its own portal inflow, hepatic-vein outflow, and bile duct. This page is that map on a contrast CT. It is not an organ-mask how-to, and it is not ultrasound.

If you meant the liver organ mask (automated contouring) → liver segmentation on CT. If you meant training a U-Net on Task03 → automatic liver segmentation — Part 1. If you meant the same map on ultrasound → ultrasound liver segments. Generic methods → medical image segmentation. nnU-Net → nnU-Net for medical image segmentation.

The Couinaud map

Claude Couinaud’s 1957 system splits the liver into eight functionally independent segments (I–VIII). Radiologists and surgeons talk in that language so a lesion in segment VI is the same place on the scan and in the OR. The middle hepatic vein splits left from right. Portal-vein branches then cut each hemiliver into the numbered units. The gallbladder fossa sits between IVb and V.

Segment 4 is often split into IVa (superior) and IVb (inferior, next to the gallbladder). That split matters more on ultrasound than on a routine axial CT — see the IVa/IVb table on ultrasound liver segments.

Why assignment is hard

The map is clean. The scan is not. Accessory fissures, odd portal branching, and a cirrhotic contour move the landmarks. On axial CT, published work puts mis-assignment at about 17.3% of the area in the marginal zones and about 51.6% in the central zones — the places where two segments meet and a thin slice can put a lesion on the wrong side of a vein. Read the veins in more than one plane. Multiplanar reformats are the cheap fix; a 3D render is a second look, not a substitute for the source series.

Four-phase protocol

You read Couinaud on the phase that shows the veins. Stolen here from the older protocol note, then we stop — this is still anatomy, not a masking article.

Bolus tracking beats a fixed clock. Motion (breath, a late arterial) smears the middle hepatic vein and you lose the left/right split. That is an acquisition problem, not a Couinaud problem.

CRL-R and LSVR

Once the segments exist as volumes, two ratios are the ones people actually quote. CRL-R is caudate (segment 1) over the right lobe (5–8). LSVR is (left lobe 2–4 + caudate) over the right lobe. In cirrhosis the right lobe shrinks and both ratios climb.

Measure on the source slices (or a tool that uses them), not on a cinematic still. 3D Slicer is one place people do this. The mask that feeds the volume is liver segmentation on CT — a different job.

Surgical use, one page

A right hepatectomy is segments 5–8. A left is 2–4, sometimes 1. A segmentectomy is one number. The future liver remnant is those leftover segment volumes — if the remnant is too small, the resection is not safe. Living-donor work is the same arithmetic on both sides. Vascular variants (accessory hepatic vein, early portal split) move the cut. That is why the CT is read with the Couinaud table in the room, not after.

AI on this page means a segment-boundary helper — a U-Net that proposes I–VIII so a radiologist can accept or move a line. It is not “draw the whole liver as one blob.” That organ-mask job, and how to train it on Task03, live on liver segmentation on CT and Part 1 of the MONAI series.

PYCAD builds viewers that have to treat Couinaud labels, an organ mask, and a volume render as three different tools. Case studies.