Examinations were performed in accordance with the standards of the institutional review board. Informed consent was not required by the review board. At first, we performed thin-section iodized-oil CT immediately after segmental chemoembolization of S2 and S3 in 27 consecutive patients with small (≤3 cm) solitary nodular HCC at the left lateral segment (LLS) from June 2003 to March 2005. There were 20 men and seven women ranging in age from 52-75 years, with a median age of 62-years-old. Multidetector row helical CT was performed on a Sensation 16 (Siemens Medical Solutions, Forchheim, Germany) or LightSpeed (GE Medical Systems, Milwaukee, WI) with 1-mm thick sections. The axial images were transferred to a PC equipped with dedicated three-dimensional (3D) reconstruction software (Rapidia^®; Infinitt, Seoul, Korea), and sagittal multiplanar reformation (MPR) images of the LLS were generated. We measured the angle ('+' was defined as the anterior direction and '-' as the posterior direction) of the plane dividing S2 and S3 on the basis of a vertical plane at three different parts and calculated the average (Fig. 2). These three parts were the medial and lateral ends of the plane (as divided by lipiodol uptake) and the center of the plane.

In this retrospective study, a computerized search of the TACE registry database of our department during a five-year period from March 1998 to June 2003 revealed triple-phase helical CT of 106 patients with small (≤ 3 cm) solitary nodular HCC specifically within the LLS. Segmental location was confirmed by selective angiography of segmental arteries and segmental TACE followed by iodized-oil CT. Homogeneous iodized-oil accumulation in the entire tumor nodule without a defect was the inclusion criteria.

Four patients were excluded from the analysis for any of the following reasons: (a) variations in S2 and S3 portal vein comprised of common trunk (n = 2), (b) nonvisualization of the left portal vein or its branches on CT scan (n = 1), and (c) an undersized S3 portal vein branch (n = 1).

Our study group was comprised of the remaining 102 patients. The mean age of the patients was 65 years (age range, 38-81 years). Seventy-eight patients were men, and 24 patients were women. The mean diameter of the solitary nodules was 2.1 cm (size range, 1.0-3.0 cm). The average time interval between CT and TACE was 3.1 weeks. An institutional review board approved this retrospective review. Informed consent was not required by the review board.

All patients underwent triple-phase CT imaging consisting of pre-contrast, arterial dominant, and portal dominant phases before embolization. A single detector helical CT (Somatom plus: Siemens, Erlangen, Germany) was used for six patients and a multidetector helical CT (LightSpeed: GE Medical Systems, Milwaukee, WI; MX8000: Marconi Medical Systems, Cleveland, OH) was used for 96 patients. For pre-contrast images, 5 mm section thickness was acquired for 95 patients; for the remaining seven patients, section thickness was 8 mm. After the administration of 120 ml of Ultravist 370 iopromide, a nonionic contrast material (Schering AG, Berlin, Germany) at a rate of 3 mL/sec using a power injector, arterial and venous phase helical CT scans were obtained. The scanning parameters for the single detector helical CT scanners were 5 mm slice thickness, pitch of 1.5, 3 mm reconstruction interval for the arterial phase and 5 mm interval for the portal venous phase, 150 mAs, 120 kVp, and a 512×512 matrix. Scanning parameters for multidetector helical CT scanners included gantry rotation time of 0.5-0.8 sec with 4×1.25 mm or 8×1.25 mm detector configuration, 2.5 mm slice thickness, pitch of 1-1.5, 3 mm reconstruction interval for both phases, 150 mAs, 120 kVp, and a 512×512 matrix. Arterial phase scans were initiated 11 seconds (for single detector helical CT) or 13 (for multi-detector helical CT) seconds after the aortic attenuation reached 100 HU using bolus tracking. After completion of arterial phase scanning, a 15-second delay (for single detector helical CT) or 30-second delay (for multidetector helical CT) was used for the portal venous phase. Unenhanced follow-up CT examination was performed during the second or third weeks after embolization.

The horizontal method divides S2 and S3 by a horizontal plane at the level of the umbilical portion of the left portal vein. The nodule above the umbilical portion of the left portal vein is considered to be S2 and the nodule below the umbilical portion is considered to be S3.

We established a hypothesis that a plane, which is vertical and of equal distance from the S2 and S3 portal veins, can divide S2 and S3 more accurately than the transverse plane. In the cases of liver cirrhosis, the vertical plane becomes the curved plane when the left lateral segment is hypertrophied. According to the vertical method, all the nodules anterior to the plane are considered to be in S3, whereas the nodules posterior to the plane are considered to be S2. How to create the vertical plane in each case is illustrated in Figure 3. The procedure was performed on a PACS viewer using the cine mode.

We applied two different segmentation methods (vertical method proposed by the authors versus horizontal method) in each patient. Two board certified radiologists independently interpreted triple-phase helical CT examination of the 102 patients and predicted the segmental location of the tumor using the vertical method independently. To prevent bias, the reviewers were uninformed as to the results of angiography, other reviewer's opinions, and other segmental methods.

After six months or more, the same reviewers applied the horizontal method to predict segmental location of tumors without knowing the results of angiography and the previous segmentation method.

When a tumor was intersected by a dividing plane during the process of segmental localization, it was classified as being in a "borderline location," abbreviated as S23, and the possibility of blood supply from both segmental hepatic arteries was suggested.

Using the S2 portal vein and S3 portal vein as a standard, the LLS of the liver can be divided into three parts, both in the anterior-posterior direction (anterior, middle, and posterior) and superior-inferior directions (superior, middle, and inferior). The LLS of the liver is thus divided into nine zones: superior-anterior (SA), superior-middle (SM), superior-posterior (SP), middle-anterior (MA), middle-middle (MM), middle-posterior (MP), inferior-anterior (IA), inferior-middle (IM), and inferior-posterior (IP).

According to this concept, the distribution of 102 nodules among the nine zones was accordingly assessed (Fig. 4).

A 5-Fr RH catheter (Cook, Frankfurt, Germany) was used to perform celiac arteriography via the right femoral artery. After selective placement of a microcatheter (Microferret-18, Cook, Canada) in the S2 or S3 hepatic arteries, the segmental arterial blood supply to the individual tumor nodule was evaluated with digital subtraction angiography. After identifying tumor-feeding segmental hepatic arteries, segmental TACE with an emulsion of doxorubicin hydrochloride and iodized-oil (Lipiodol Ultrafluid; Laboratoire Andre Guerbet) was performed. Two weeks later, unenhanced iodized-oil CT was performed and homogeneous iodized-oil accumulation in the entire tumor nodule without a defect was confirmed. The angiographic results of tumor feeding arteries were used as a standard reference for the comparison (Fig. 3).

One radiologist validated and compared the results of segmentation by two different methods based on the results of angiography and segmental TACE.

We used the weighted kappa coefficient of agreement (weighted Kappa) of GraphPad Prism, version 8.00 for Windows (GraphPad Software, San Diego, CA) to evaluate the agreement between the angiographic tumor location and CT interpretation and to evaluate the interobsever's agreement. A kappa value of 1 indicates complete agreement; on the other hand, a kappa value of 0 indicates only random agreement. We defined the segments as identical for values of > 0.75 (very good agreement), as coinciding acceptably by values between 0.45 and 0.75 (good agreement), and as coinciding poorly for values below 0.45 (poor agreement) (13).

Chi-square test (SPSS for Windows, release 10.0.7; SPSS, Chicago, IL) and the weighted kappa coefficient of agreement were used to compare two methods with respect to the segmental location of the tumor for each reviewer.

In the investigation with sagittal MPR images of the iodized-oil CT, the average angle of the plane dividing S2 and S3 was slanted 15 degrees anteriorly. The average angle from the vertical plane in the right-left direction was +12 degrees at the medial end of the plane, +17 degrees at the center of the plane, and +16 degrees at the lateral end of the plane (Fig. 2).

Forty-one nodules were located in S2, 56 nodules were found in S3, and five nodules were in the borderline location with blood supply from both segmental hepatic arteries (which will be abbreviated as S23).

The left lateral segment was divided into nine zones. The distribution of 102 nodules among these nine zones, as well as the tumor-feeding arteries in each zone, are illustrated in Figure 4.

All of the 28 nodules in front of the S3 portal vein were supplied by the S3 hepatic artery, and all of the 15 nodules posterior to S2 portal vein were supplied by the S2 hepatic artery. There was one exceptional case with partial S3 hepatic arterial supply to the IP zone. Among the 59 nodules between the S2 and S3 portal veins (in the SM, MM, IM zones), they were almost equally supplied by the S2 and S3 hepatic arteries: 28 nodules were solely supplied by the S3 hepatic artery, 27 nodules by the S2 hepatic artery, and the remaining four by both hepatic arteries. In the SM zone, nodules supplied by the S2 hepatic artery were dominant. On the other hand, IM zone nodules were more frequently supplied by the S3 hepatic artery.

The agreement rates between CT interpretation by vertical method and angiographic results were 88.2% (90 of 102) for Reviewer 1 and 84.3% (86 of 102) for Reviewer 2 (Table 1). Weighted Kappa values were 0.838 for Reviewer 1 and 0.756 for Reviewer 2, which indicates 'very good' agreement between CT interpretation by vertical method and angiographic results.

Approximately 1/2 of all mismatches occurred in tumor nodules interpreted as S23 (that is, "borderline location") on CT interpretation (seven of 12 mismatches by Reviewer 1 and seven of 16 mismatches by Reviewer 2). In 11 cases with the tumors in the "borderline location" on CT interpretation either by Reviewer 1 or 2, five cases were angiographically confirmed as S2 lesions while the other two cases were S3 lesions. The remaining four cases were supplied by both the S2 and S3 hepatic arteries.

When the LLS of the liver was divided into nine zones by the standard of the S2 portal vein and S3 portal vein, mismatches by both Reviewer 1 and 2 occurred in SM, MM, and IP zones (two cases in SM, two cases in MM, and one case in IP zone).

Inter-observer agreement rate was 96.1% (98 of 102 cases) and weighted Kappa value was 0.918 with 'very good' degree in the strength of agreement (Table 2). Mismatches between reviewers were observed only in four cases.

The agreement rate between CT interpretation by the horizontal method and angiographic results was 78.4% (80 cases of 102 cases) from Reviewer 1 and 75.5% (77 cases of 102 cases) from Reviewer 2 with the weighted Kappa value of 0.611 for Reviewer 1 and 0.510 for Reviewer 2 (Table 3). Inter-observer agreement rate was 91.2% (93 of 102 cases) and weighted Kappa value was 0.813 with 'very good' degree of strength of agreement (Table 2).

The agreement rates in the vertical method were higher than those in the horizontal method from both reviewers (88.2% vs. 78.4% from Reviewer 1 and 84.3% vs. 75.5% from Reviewer 2). The difference was statistically significant (p-value > 0.0001 for Reviewer 1 and for Reviewer 2 by Chi-square test). Weighted Kappa value between two methods was 0.654 with Reviewer 1 and 0.543 with Reviewer 2.

The model we propose has its limitations, however. This technique cannot be applied when the portal vein is not well delineated. This occasion may happen with liver cirrhosis, tumor invasion to the portal vein, or with inappropriate portal venous phase scans. Severe anatomical variation or tortuosities of S2 or S3 portal veins can cause incorrect segmental localization of tumors or give rise to considerable individual variations in interpretation. Fortunately, the LLS showed relatively constant portal anatomy (18, 19).

We compared the vertical plane for dividing the S2 and S3 segments, or vertical method, to the horizontal method. In our clinical experience of segmental angiography, the practical plane may be more oblique than horizontal, but we defined the actual conventional method as a purely horizontal plane simply because of individual variation and obscure definitions of the relationship between S2 and S3. According to the results of this study, the orientation of the intersegmental plane is slightly slanted anteriorly from the vertical plane. This was not only seen in the preliminary study using iodized-oil CT immediately after segmental chemoembolization, but was also observed in analysis of tumor-feeding arteries using small HCC in the LLS.

In conclusion, our proposed vertical method is simple and effective in accurately distinguishing the division between section 2 (S2) and section 3 (S3) of the liver.