This study was conducted at The General Hospital of Western Theater Command PLA from January 1, 2021, to June 1, 2024, and was approved by the institutional ethics committee (license number: 2023EC5-ky005). Adult patients aged 18 to 65 years were considered eligible if they (1) showed no evidence of space-occupying lesions, biliary dilatation, or cirrhosis; (2) had no previous surgical history or anatomical variations affecting hepatic pedicle morphology; (3) had clear and complete contrast-enhanced CT scans demonstrating the main hepatic pedicle structures, including primary and secondary branches; and (4) consented to participate after review by the ethics board.

The validation cohort consisted of 30 adult patients undergoing left hemihepatectomy, deliberately selected to standardize intraoperative measurements and focus on the left hepatic pedicle. This surgical procedure was chosen because it provides consistent anatomical exposure of the left portal branch, facilitating precise intraoperative measurements of its length, external diameter, and angle relative to the main hepatic trunk. These measurements were critical for validating the accuracy of the Laennec's capsule-based 3D reconstruction method by enabling direct comparisons with preoperative imaging data. The focus on left hemihepatectomy ensured methodological consistency and minimized variability in surgical access to the hepatic pedicle structures.

All participants underwent abdominal contrast-enhanced CT scanning using a 64-row multi-slice spiral CT scanner (GE LightSpeed VCT, GE, USA). The DICOM images were then imported into the Hisense Computer Assisted Surgery (CAS) system for 3D processing. Laennec's capsule was identified on contrast-enhanced CT as a thin, hypodense fibrous layer surrounding the hepatic pedicle, distinct from the hyperdense Glisson's sheath and adjacent parenchyma. Two senior radiologists independently annotated the capsule slice by slice, with inter-observer agreement assessed via Cohen's kappa (κ = 0.85). Measurements were performed twice by each radiologist, and averaged values were used for analysis. The grayscale values in each DICOM slice to Hounsfield units (HU) using the following formula:HU=pixel_value*RescaleSlope+RescaleInterceptHere, pixel_value denotes the gray value of each pixel in the original DICOM image, while RescaleSlope and RescaleIntercept are obtained from the DICOM image header file. Afterward, the window level was adjusted to −450 to 600 HU (window width 1,500–2,000 HU) for optimal liver structure visualization. Each labeled CT dataset was then cropped to a minimal cuboid region enclosing the liver parenchyma, and all data (CT images and manual labels) were resampled using third-order spline interpolation to standardize voxel size (1.0 mm × 0.75 mm × 0.75 mm recommended).

A 3D U-net deep neural network was applied to segment the hepatic pedicle. Data augmentation steps—including angle rotation, grayscale adjustments, and flipping—were performed to enhance the network's generalizability. The training process used a combined cross-entropy and Dice coefficient loss function, defined as:LossFunction=LCross−Entropy+LDice=−∑n=1Nyilog(yi^)−1|N|∑n∈N2∑yiyi^∑(yi+yi^)Where y_i and ŷ represent the standard and predicted values for pixel i, respectively. The Ranger optimizer, which merges RAdam and LookAhead, was used, and the initial learning rate of 1 × 10⁻⁴ employed a warmup strategy. After training, the network automatically segmented the primary, secondary, and tertiary branches of the hepatic pedicle, from which the system extracted measurements of length, external diameter, and branch angles.

Between January and June 2024, a separate cohort of 30 adult patients (16 females and 14 males; age range 34–86 years) underwent left hemihepatectomy for various liver conditions (13 Intrahepatic Bile Duct Stone, 12 hepatocellular carcinomas, 4 hepatic hemangiomas and 1 Cirrhotic Nodular Hyperplasia). Preoperatively, contrast-enhanced CT scans were acquired and processed for 3D reconstruction using the same protocol described above.

Intraoperatively, the first hepatic hilum was fully exposed by dissecting along the plane between Laennec's capsule and the Glisson pedicle. The “first hepatic hilum” refers to the initial bifurcation point of the hepatic pedicle into left and right main branches. The length and external diameter of the left branch (from its origin to the third hepatic portal) and the angle between the left branch and the main hepatic trunk were measured with a probe. The “third hepatic portal” denotes the tertiary branching point within the intrahepatic pedicle. These intraoperative measurements were then compared with the corresponding preoperative 3D reconstruction data to evaluate the accuracy of the imaging-based method.

All statistical analyses were conducted using SPSS 25.0 software (IBM Corp., Armonk, NY, USA). Continuous variables are presented as mean ± standard deviation (x¯ ± s), whereas categorical variables are expressed as percentages (%). A paired t-test was used to compare the imaging-based measurements with those obtained intraoperatively, and values of P < 0.05 were considered statistically significant.

A total of 30 patients (17 females, 13 males; age range 56–75 years) who underwent left hemihepatectomy were used to validate the accuracy of the three-dimensional (3D) reconstruction based on Laennec's capsule. The demographic and clinical characteristics of this cohort are summarized in Table 1, including gender, age, diagnosis, pathological type, and comorbid diseases, providing a comprehensive overview of the study population. Intraoperatively, the hepatic pedicle and its left branch were dissected (Figures 1a,b), and the length and external diameter of the left main branch, as well as the angle between the left branch and the main trunk, were measured. Comparison of these operative findings with the corresponding 3D reconstruction data (Figures 1c,d) showed no significant differences (Figures 1e–g). Specifically, the mean left branch length measured 20.05 ± 6.81 mm via imaging vs. 20.51 ± 6.56 mm intraoperatively (P = 0.067), while the external diameter was 17.89 ± 4.34 mm on imaging vs. 18.41 ± 4.35 mm in surgery (P = 0.774). The angle between the left main branch and the hepatic trunk showed a similar lack of difference between imaging (114.67 ± 24.60°) and operative (115.01 ± 24.48°) measurements (P = 0.999). These findings confirm that the 3D reconstruction method based on Laennec's capsule accurately reflects clinical anatomy.

The 3D U-Net was validated on the dataset to assess its performance in segmenting liver and tumor structures. The algorithm achieved a mean Dice Similarity Coefficient (DSC) of 0.984 for liver segmentation and 0.840 for tumor segmentation, outperforming the U-Net baseline method. These results highlight the robustness of the 3D U-Net in accurately delineating hepatic structures, particularly the liver parenchyma, though tumor segmentation presented greater challenges due to its morphological variability. The annotation process for the hepatic pedicle involved initial labeling by personnel with professional imaging knowledge, followed by confirmation from a specialist doctor, ensuring a rigorous review process to enhance the reliability of the segmented data used for subsequent measurements.

Among 100 adult livers reconstructed by the same 3D method, four distinct branching patterns were identified based on the main trunk, left branch, and right branch configuration (Figure 2). •

Type I (Figures 2a,b): Classic bifurcation into left and right main branches, observed in 88% (88/100) of cases.

Table 2 summarizes the length and external diameter of the main trunk, left main branch, right main branch, and their secondary divisions in the 100 reconstructed specimens. At the first hepatic hilum, the main hepatic pedicle's mean outer diameter was 24.10 ± 6.16 mm. The left main branch measured 20.59 ± 6.38 mm in length and 18.04 ± 4.48 mm in outer diameter, whereas the right main branch was 21.99 ± 7.98 mm long and 21.18 ± 4.23 mm in outer diameter. For the two major divisions of the right main branch (the right anterior and posterior lobar branches), both length and diameter also exhibited inter-individual variability.

The angular relationships are presented in Table 3. In Type I livers (the majority subtype), the mean angle between the hepatic pedicle and the left main branch (ALM) was 114.35 ± 22.91°, and between the pedicle and the right main branch (ARM) was 140.81 ± 16.72°. The angle formed by the left and right main branches (ALR) was 142.63 ± 15.66°. These measurements highlight the considerable anatomical variability within the hepatic hilum, reinforcing the need for careful preoperative evaluation.

Based on the statistical analysis of Type I livers, which constituted nearly 90% of cases, a representative 3D liver model was designed (Figure 3a). This model was then printed using a transparent resin to illustrate the main vascular structures within the Laennec's capsule (Figure 3b). The resulting physical model proved valuable for preoperative planning, intraoperative orientation, and educational demonstrations, offering a highly detailed depiction of the hepatic pedicle and surrounding anatomy.