Clinical Report: Comparing Three Modeling Approaches to Predict Resection Zones

Overview

This study describes a retrospective comparison of distance-based, perfusion-based, and deep-learning models for predicting parenchyma-sparing liver resection zones. The dataset included 104 patients and 104 retrospectively defined resection zones; the supplied material describes the methods but does not report comparative model-performance results.

Background

Liver resection and transplantation are principal curative treatments for primary hepatic tumors, and resection planning must preserve sufficient future liver remnant (FLR) to reduce the risk of postoperative liver failure. The source gives FLR thresholds of at least 20–30% for healthy livers and at least 40% when hepatic reserve is impaired. Parenchyma-sparing surgery seeks to limit the volume removed while maintaining oncological safety and preserving liver function for potential subsequent resections. Planning must account for tumor location, hepatic perfusion, biliary drainage, and patient characteristics.

Data Highlights

Key Findings

• The analysis used 104 retrospectively defined resection zones from a single-center dataset.
• The original cohort comprised 312 patients who underwent 338 non-laparoscopic liver tumor resections, predominantly for hepatocellular carcinoma or intrahepatic cholangiocarcinoma.
• Seven surgeons performed the procedures; the analyzed segmentation masks were semiautomatically generated and supplied as STL files.
• STL masks were rasterized onto a shared voxel grid at 0.4 mm isotropic resolution, with reported boundary discretization errors under 1 mm.
• The study compared distance-based, perfusion-based, and deep-learning approaches, evaluating each independently.
• Spatial agreement was to be assessed using two overlap measures (DSC and IoU) and three distance measures (HD₉₅, HD_max, and ASD); comparative results are not included in the supplied material.

Clinical Implications

The source frames resection-zone prediction as a planning problem in which tumor location, vascular perfusion, biliary drainage, and preservation of adequate FLR are relevant. The supplied material does not provide comparative performance results or establish that any of the three modeling approaches should be preferred in clinical practice.

Conclusion

This retrospective study establishes a three-approach framework for evaluating parenchyma-sparing resection-zone prediction in 104 patients. The provided text reports the dataset and evaluation methods but not the models’ comparative results.

Related Resources & Content

1. EASL, EASL Clinical Practice Guidelines, 2025 -- EASL Clinical Practice Guidelines on the management of hepatocellular carcinoma
2. Rakshit et al., source citation [5] -- Geometric deep learning framework for liver resection-zone prediction (full bibliographic details not provided)
3. Messaoudi et al., source citation [6] -- Automated future liver remnant prediction (full bibliographic details not provided)
4. Updates in Surgery, 2022 -- Utilization of Hyper Accurate 3D Technology for Complex Liver Resection Planning: Initial Findings from a Single-Center Study
5. Surgical Endoscopy — Comparative Analysis of Robotic and Open Hepatectomy for Expert-Level Liver Resection: A Risk-Adjusted Study
6. Langenbecks Archives of Surgery — Innovative 3D Surgical Planning for Liver Transplantation: Advancements in Preoperative Evaluation and Management of Vascular and Biliary Issues
7. Comparative Evaluation of Local Semi-Automated 3D Liver Reconstruction Versus External Services: Performance and Time Analysis
8. EASL Clinical Practice Guidelines on the management of hepatocellular carcinoma
9. Utilization of Hyper Accurate 3D Technology for Complex Liver Resection Planning: Initial Findings from a Single-Center Study
10. Effect of anatomical liver resection for hepatocellular carcinoma: a systematic review and meta-analysis - PMC