Artificial intelligence-assisted surgical simulation system based on non-enhanced computed tomography images in thoracoscopic pulmonary segmentectomies

Introduction

Thoracoscopic anatomical lung segmentectomy has emerged as a noninferior treatment of small (≤2 cm) early-stage non-small cell lung cancer (NSCLC) comparable to lobectomy, as evidenced by findings from notable clinical trials [JCOG0802/WJOG4607 (1) and CALGB140503 (2)]. The success of this surgical approach hinges on precise preoperative localization and a well-conceived surgical strategy, particularly for surgeons with limited experience (3). In this context, three-dimensional (3D) reconstruction technology plays a pivotal role in the planning of lung resections, offering surgeons clear and intuitive 3D anatomic models (4-7). These models facilitate a comprehensive observation of lung nodules from any perspective, accurately show the relative positions of pulmonary nodules to their adjacent tissues, and provide personalized guidance for surgeons, which has been validated in several studies (8-10). The traditional 3D-computed tomography (CT) bronchography and angiography are valuable for surgical guidance, but these methods rely on multi-step operations and manual anatomical matching, require professional expertise and result in lengthy reconstruction time. Moreover, they necessitate the use of enhanced CT scanning, which is invasive and contraindicated for individuals with allergies to contrast agents, further restricting their applicability (11-13).

The artificial intelligence (AI) techniques are presumed to be a good choice to address these challenges (14,15). In this work, we investigated an innovative software (LungDimensionGo V1.0 SunKun) program that significantly simplifies the operation process. This software is capable of precise preoperative screening and diagnosis of lung nodules and fully automatically reconstructing the anatomic models for intraoperative navigation within 2–3 minutes, enhancing intraoperative navigation accuracy and helping to reduce lung damage and accelerate patient recovery.

The primary aim of our study is to evaluate the performance and efficiency of this AI-assisted surgical simulation system in video-assisted thoracoscopic segmentectomies. We conducted a comprehensive comparison with traditional method, examining preoperative, intraoperative, and postoperative aspects to assess the potential advantages and limitations of this innovative approach. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-375/rc).

Methods

Patient population

We conducted a retrospective comparison study using the picture archiving and communication system (PACS) and electronic health record (EHR) to identify patients with surgically resected lung nodules from January 2021 to October 2023. All CT examinations and pulmonary nodule resection surgeries were performed at The Affiliated Panyu Central Hospital of Guangzhou Medical University. All scans were conducted by The Affiliated Panyu Central Hospital of Guangzhou Medical University’s radiographers using identical imaging equipment and acquisition parameters, and all surgeries were performed by thoracic surgeons from The Affiliated Panyu Central Hospital of Guangzhou Medical University. This search identified 1,551 patients. Inclusion criteria for the study were preoperative chest CT scans within 2 weeks, thoracoscopic lung segmentectomies, and preoperative 3D reconstruction using either AI-assisted or traditional methods. Indications for anatomical segmentectomy in early-stage lung cancer included preoperatively biopsied lung tumor nodules or non-biopsied nodules highly suspected to be malignant and smaller than 2 cm, as well as nodules with a 50% ground-glass appearance on CT or those confirmed by radiologic surveillance to have a long doubling time (greater than 400 days). Patients who underwent lung wedge resections (n=673), pulmonary lobectomies (n=533), or combined surgical resections (n=201) were excluded, leaving 144 patients. Further exclusions were made for those with a history of lung surgery or trauma (n=9), incomplete surgical records or immunohistochemical assessments (n=8), or surgeries performed by external experts (n=2), resulting in 125 patients split into the AI-assisted method group (n=64) and the traditional method group (n=61). Out of these, 52 patients had surgical videos recorded. Six patients were excluded due to the absence of preoperative chest enhanced CT scans, and three were excluded for incomplete surgical videos. Ultimately, 43 patients with complete surgical videos were selected for assessing the accuracy of intraoperative anatomical structures using 3D images from both reconstruction methods. The patient selection process is summarized in Figure 1.

Figure 1 Flowchart shows patient selection process. 3D, three-dimensional; AI, artificial intelligence; CT, computed tomography; EHR, electronic health record; PACS, picture archiving and communication system.

Building on this, we prospectively recruited 38 consecutive patients using the same research methods from November 2023 to July 2024, recording surgical data and videos. Cases were randomly assigned to either the traditional or AI group by independent researchers (who were not involved in outcome assessment), and the grouping information was concealed from the operators. The results from the AI group were presented through a standardized interface without revealing the algorithmic source. Operators were only aware that the two reconstruction software systems differed in operational methodology but were not informed of the specific details. Therefore, this setup did not influence reconstruction time, surgical outcomes, or complication rates. Outcomes were assessed by an independent researcher who was not involved in the procedural operations.

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The research protocol was approved by the Institutional Ethics Committee of The Affiliated Panyu Central Hospital of Guangzhou Medical University (IRB: PYRC-2023-392; date of approval: December 15, 2023). Informed consent was obtained from all participants in the prospective cohort; yet, it was waived in the retrospective cohort.

AI-assisted or conventional surgical simulation

Chest CT scans (non-enhanced or enhanced, slice thickness <1.25 mm) of the patients were uploaded from the PACS to the AI-assisted reconstruction software in the form of digital imaging and communication in medicine (DICOM). The reconstruction process consists of three main steps. Firstly, the 3D-UNet is utilized to initially segment the lung structures. Next, the deep distance transform was incorporated to refine the boundaries of the tubular structures such as vessels and bronchi. Finally, the marching cube algorithm is adopted to generate a mesh representation. During the training, four types of loss function are calculated to optimize the model, including focal loss, dice loss, cross-entropy loss and distance loss. The 3D images showed nodules, margin spheres (2–5 cm), lung segment planes, arteries, veins, and bronchi (Figure 2). Methods details available in Appendix 1.

Figure 2 A 55-year-old woman with three high-risk nodules in right lung. (A-D) AI-assisted reconstruction results. (A) Spatial location of the nodule in lung segments. (B) Pulmonary veins. (C) Arteries. (D) Bronchi. AI, artificial intelligence.

Chest contrast-enhanced CT images were imported into the CT post-processing workstation for 3D reconstruction. The operator utilitied the Lung VCAR software to manually locate the suspicious nodule. Then the software automatically analyzed the nodule’s characteristics (morphology, volume, size, CT value, etc.). The specific structures, such as pulmonary vessels, bronchi, and nodules, are first semi-automatically reconstructed respectively. Then, these reconstructed structures, each represented by unique color, were fused together to create a 3D image of the lung containing multiple structures. During the fusion process, the images were optimized by adjusting the window width, window level and the transparency of each structure. Multi-planar reconstruction (MRP), maximal intensity projection (MIP), and minimum intensity projection (MinIP) were used to analyze the spatial positions between pulmonary nodules, vessels, and bronchi, as well as to measure the size of the nodule and evaluate the morphology of nodule. Lung segments were reviewed by using Thoracic VCAR and Lobe Segmentation software packages.

Anatomical accuracy verification

Anatomical accuracy verification of 3D model by both AI-assist and traditional reconstruction method were compared with actual surgical video findings for patients who has complete surgical video (Figure 3). This process was evaluated postoperatively by two thoracic surgeons, one of whom holds a senior or associate senior professional title. When the results of the 2 were inconsistent, another senior thoracic surgeon arbitrated the determination. The evaluation process involved three key steps: (I) verifying the presence and accuracy of arteries, veins, and bronchi in the 3D images compared to intraoperative observations; (II) assessing whether the anatomical classifications of vessels and bronchi from both methods were consistent with the actual surgical findings; (III) determining the alignment of preoperative plans for lung segment resections and the sequencing of bronchial and vascular disconnections with actual surgical outcomes. Any bronchi or blood vessels observed in the surgical videos but not displayed in the 3D images were recorded as unreconstructed.

Figure 3 The 3D reconstruction was consistent with the intraoperative exploration. (A-C) A 36-year-old man with a high-risk nodule in LS10. (D-F) A 31-year-old woman with a high-risk nodule in LS(4+5). (A) The CT image of LS10. (B) The AI-assisted reconstruction image of LS10. (C) The intraoperative image of LS10. (D) The CT image of LS(4+5). (E) The traditional reconstruction image of LS(4+5). (F) The intraoperative image of LS(4+5). 3D, three-dimensional; AI, artificial intelligence; CT, computed tomography; L, left; S, segment.

Surgical technique

General anesthesia was intravenous inhalation and double-lumen endotracheal intubation with unilateral lung ventilation. Then the surgical field was prepared with routine disinfection and draping. Single-port or multi-port thoracoscopic surgery was performed and the incision length was approximately 3–5 cm. The patient’s position was left or right lateral decubitus and the surgical access approach including via anterior mediastinum, posterior mediastinum, or interlobar fissure. Nodule location was verified against thoracic landmarks, pleural inspection, and manual palpation. Anatomical segmentectomy entailed excising the segmental pedicle and dissecting the designated bronchial and vascular segments. Segmental boundaries were identified using a lung expansion-collapse method, followed by resection with a linear stapler. Electrocautery helped create an intersegmental plane for pedicle separation. A minimum 2 cm surgical margin was required for malignant nodules. Systematic mediastinal and hilar nodal sampling and an intraoperative frozen-section analysis was then performed. The operative procedure is shown in Figure 4.

Figure 4 A 52-year-old woman with a high-risk nodule in junction of S8 and S9 segments of right lung. (A-E) The patient’s entire treatment process. (A) A ground glass nodule was seen at the junction of S8 and S9 segments. (B) 3D image made in AI-assisted software. (C) Pathology (HE, ×40): minimally invasive adenocarcinoma. (D,E) S8+S9 segmentectomies were performed under 3D image navigation. 3D, three-dimensional; AI, artificial intelligence; HE, hematoxylin-eosin; S, segment.

Study parameters and statistical analysis

Study parameters included preoperative indicators (e.g., patient’s gender, age, smoking history, comorbidity, family history, nodule location, maximum nodule diameter, nodule type, nodule characteristics, reconstruction time), intraoperative indicators (e.g., accuracy of anatomical structures, consistency of preoperative planning, lung segments resected, operative time, intraoperative blood loss, pleural adhesions), postoperative indicators (e.g., postoperative chest tube duration, 3-day postoperative chest drainage, postoperative hospital stays, postoperative complications, postoperative pathology, total hospital stay). Data were analyzed using R 4.3.0 (https://www.r-project.org/). Continuous variables with normal distributions were presented as mean ± standard deviation, and continuous variables with non-normal distributions were presented as median (interquartile range). Categorical variables were presented as frequency and percentage. A Mann-Whitney U analysis (non-normal distribution) or an independent sample t-test (normal distribution) was used to compare the continuous data of the two groups. Fisher’s exact test (n<40 or T <1) or χ² (n≥40 and T ≥1) were used to compare the two groups’ categorical data. A two-sided P value that below 0.05 was considered statistically significant.

Results

Preoperative patient characteristics and nodules analysis

Patient characteristics and nodule features for both the retrospective and prospective studies are summarized in Tables 1,2, respectively. In the retrospective study, 125 patients were enrolled, with 64 in the AI-assisted group and 61 in the traditional group. The mean age of the patients in the AI-assisted group was 54±13.2 years, compared to 52.3±11.8 years in the traditional group. There were no significant differences between the two groups in terms of age, gender, smoking history, comorbidities, family history, or nodule pathology. Traditional 3D reconstruction took an average of 15.3±6.9 minutes, whereas AI-assisted 3D reconstruction significantly reduced this time to 2.5±1.2 minutes. A total of 133 pulmonary nodules were resected, with 70 in the AI-assisted group and 63 in the traditional group. No significant differences were found between the groups in nodule CT features [lobulation sign: irregular and lobulated contour of the nodule, reflecting uneven tumor growth. Spiculation sign: spiculated margins with radial strands, suggesting local invasion. Vacuole sign: small intranodular air-filled spaces (<5 mm), frequently observed in adenocarcinoma. Pleural indentation sign: linear pleural retraction caused by tumor traction. These features were independently assessed by two thoracic radiologists, with discrepancies resolved by consensus], nodule diameter, nodule volume, maximum CT value, mean CT value, 3D features (vascularization, bronchial threading), or nodule type.

In the prospective study, 38 patients were enrolled, with an equal distribution of 19 patients in both the AI-assisted and traditional groups. The mean age of patients in the AI-assisted group was 50.5±14.3 years, while the traditional group had a mean age of 56.2±11.6 years. Again, there were no significant differences in age, gender, smoking history, comorbidities, family history, or nodule pathology between the groups. The average time for traditional 3D reconstruction was 10.4±5.3 minutes, whereas AI-assisted 3D reconstruction reduced this to 2.1±1.1 minutes. A total of 38 pulmonary nodules were resected, with an equal number of nodules in both the AI-assisted and traditional groups. Similar to the retrospective study, there were no significant differences in nodule CT features, nodule diameter, nodule volume, maximum CT value, mean CT value, 3D features, or nodule type between the two groups.

Intraoperative accuracy and efficiency assessment

The results of the accuracy verification are presented in Table 3 and Figure 5. In the retrospective study, 43 surgical videos were reviewed to analyze the accuracy of AI-assisted versus traditional 3D reconstruction methods. The AI-assisted approach reconstructed arteries with an accuracy of 94.6%, compared to 86.1% with the traditional method (P=0.02). For veins, the AI-assisted method achieved an accuracy of 93.6%, while the traditional method had an accuracy of 81.6% (P=0.007). Overall, the AI-assisted method reconstructed anatomical structures with 95.0% accuracy, compared to 86.9% for the traditional approach (P<0.001). When considering both reconstruction and classification of all anatomical structures, the AI-assisted method reached 83.7% accuracy, significantly higher than the 58.1% accuracy of the traditional method (P=0.02). However, there were no significant differences in bronchi accuracy (97.8% vs. 95.6%, P=0.68), classification accuracy (90.7% vs. 81.4%, P=0.35), or the consistency (97.7% vs. 86.0%, P=0.11) between preoperative 3D reconstruction planning and intraoperative operation.

Figure 5 (A-H) Accuracy verification of 3D model by AI-assist and traditional reconstruction method was compared with actual surgical video findings. (A,E) CT images. (B,F) Intraoperative exploration in surgical video. (C,G) The traditional reconstruction image. (D,H) The AI-assisted reconstruction image. (A-D) A 36-year-old man with a high-risk nodule in LS10. A branch of the V10 was identified in both the CT images and intraoperative exploration. It could be seen that the AI-assisted reconstructed 3D model included this vessel, whereas the traditional reconstruction method did not reconstruct this vessel. The red circles in (A,B) demonstrate that this branch was observed in both the CT images and the surgical video. The red circle in (C) shows that the traditional 3D reconstruction failed to reconstruct this blood vessel, while the red circle in (D) indicates that the AI group’s 3D reconstruction successfully reconstructed it. (E-H) A 31-year-old woman with a high-risk nodule in LS(4+5). In the original CT images and surgical video, two branches of V5 were observed. The AI-assisted 3D model reconstructed these two branches, but the traditional reconstruction method did not. The red circles in (E,F) illustrate that the vascular bifurcation was observed in both the CT images and the surgical video. The red circle in (G) reveals that the traditional 3D reconstruction did not reconstruct the vascular bifurcation, whereas the yellow circle in (H) shows that the AI group’s 3D reconstruction successfully reconstructed the vascular branch. 3D, three-dimensional; AI, artificial intelligence; CT, computed tomography; L, left; S, segment.

In the prospective study, 38 surgical videos were similarly reviewed. The AI-assisted method reconstructed arteries with an accuracy of 96.1%, compared to 88.9% for the traditional method (P=0.02). Vein accuracy was 96.6% for the AI-assisted approach, versus 89.7% for the traditional method (P=0.04). Overall anatomical structure reconstruction accuracy was 96.8% with the AI-assisted method, compared to 91.1% with the traditional approach (P=0.001). For both reconstruction and classification of all anatomical structures, the AI-assisted method achieved 89.5% accuracy, significantly outperforming the 65.8% accuracy of the traditional method (P=0.01). Again, no significant differences were found in bronchi detection accuracy, classification accuracy, or the consistency between preoperative 3D reconstruction planning and intraoperative operation.

Intraoperative and postoperative data analysis

Both patient groups successfully underwent surgeries without converting to open thoracotomy, and no perioperative mortality was reported. Detailed surgical data are presented in Table 4, while Table 5 illustrates the segmentectomy locations. In the retrospective study, the AI-assisted group demonstrated several advantages over the traditional group: a shorter operative time (128.7±49.1 vs. 161.8±54.8 minutes, P<0.001), reduced intraoperative blood loss [10.0 (5.0, 20.0) vs. 20.0 (10.0, 30.0) mL, P=0.005], shorter postoperative chest tube duration [2.0 (2.0, 3.0) vs. 3.0 (2.0, 3.0) days, P=0.005], shorter postoperative hospital stay (3.8±1.8 vs. 5.0±2.1 days, P=0.001), and a reduced total hospital stay (7.3±3.9 vs. 9.4±3.2 days, P=0.001). Additionally, the AI-assisted group experienced fewer postoperative complications, including air leaks (P=0.005), pleural effusion (P<0.001), pleurisy (P=0.004), and infections (P=0.01). There were no significant differences in the incidence of pleural adhesions or 3-day postoperative chest drainage between the two groups (P>0.05).

In the prospective study, similar benefits were observed for the AI-assisted group: a shorter operative time (105.2±21.7 vs. 169.6±39.7 minutes, P<0.001), reduced intraoperative blood loss [10.0 (10.0, 20.0) vs. 20.0 (10.0, 25.0) mL, P=0.046], shorter postoperative chest tube duration [2.0 (1.0, 2.0) vs. 2.0 (2.0, 3.0) days, P=0.002], shorter postoperative hospital stay (2.9±0.8 vs. 5.4±3.1 days, P=0.001), and a reduced total hospital stay (5.0±1.9 vs. 9.8±4.3 days, P<0.001). No significant differences were noted in the incidence of pleural adhesions or 3-day postoperative chest drainage between the groups (P>0.05).

Discussion

In this study, AI-assisted surgical simulation process was proven to be more concise and efficient, which was superior to traditional method of CT post-processing in preoperative preparation, intraoperative guidance, and postoperative patient recovery. In preoperative preparation, AI-assisted method had significant reduction in 3D reconstruction time because it enables automated segmentation of the pulmonary nodules, parenchyma, vessels, and bronchi. Furthermore, either enhanced or non-enhanced CT images can be used for reconstruction, so it also can be used for those with allergies to contrast agents. In intraoperative guidance, AI-assisted reconstruction model had higher anatomical structures accuracy than traditional reconstruction, the former group had shorter operative time and less intraoperative blood loss, which might mean the AI-assisted reconstruction model could navigate more complex surgeries and improve efficiency and safety of surgery. In postoperative patient recovery, AI-assisted reconstruction group had fewer postoperative complications, postoperative chest tube duration, shorter postoperative hospital stays and total hospital stays. The discrepancy between the prospective study and retrospective study results is primarily attributed to the limited sample size in postoperative complications. Although the AI-assisted group showed numerically equal or higher complication rates compared to the traditional method, no statistically significant difference was observed between the two groups. 3D reconstruction provides a favorable and intuitive surgical perspective that is close to virtual reality navigation. This technology has been widely used in surgical procedures such as liver, pancreas (17), and bone. Preoperative planning of the surgical procedure by 3D reconstruction software helps improve the surgical efficiency, reduce lung injury in patients, and accelerate postoperative recovery of lung function (18-20). However, the early 3D reconstruction software used in lung surgery required professional skills to identify and reconstruct each anatomical structure respectively, and relied on contrast-enhanced CT images for vascular segmentation, which made the reconstruction process cumbersome (15).

With the rapid development of AI technology, 3D reconstruction software based on AI algorithm has gradually emerged (21). Chen et al. (16) validated the auxiliary role of a fully automated reconstruction algorithm based on non-contrast CT in segmentectomy surgery planning, and their team further rendered a semi-automated approach by developing an AI-based chest CT semantic segmentation algorithm (22). Takamoto et al. (23) confirmed that AI-assisted 3D modeling software was a suitable replacement for surgeon in the 3D liver reconstruction process compared with the conventional reconstruction method. Li et al. (14) also demonstrated that AI-based 3D reconstruction had high accuracy and efficiency in predicting pulmonary anatomy compared with the semi-automatic reconstruction software Mimics, which can shorten the operation time. The emergence of AI-based 3D reconstruction technology has reduced the reconstruction steps and decreased the user’s working load to promote the development of the 3D reconstruction in lungs (24,25).

We found AI-assisted method reduced the operator’s working load and made the process more efficient and concise due to the fully automated nature, whereas the CT post-processing reconstruction was more cumbersome because it required the operator to have a certain level of expertise in lung anatomy and individual skill in 3D simulation software. Additionally, The AI-assisted reconstruction group enabled the more detailed visualization of vascular anatomy, by accurately displaying variations in small peripheral branches of lung segmental vessels (26-29). This made it easier to avoid damaging blood vessels during surgery, reducing intraoperative blood loss and lung tissue injury, improving surgical efficiency, and accelerating patient recovery. However, we acknowledge that routine hemoglobin measurement of pleural drainage fluid was not performed in this study, limiting our ability to quantitatively report hemothorax incidence. This parameter will be incorporated into future research for more comprehensive analysis. In contrast, the quality of CT post-processing 3D images is limited by the operator’s experience and patience such as manual annotation of accurate anatomy and adjustment of the window width and level. In addition, vascular image made by CT post-processing requires contrast-enhanced CT, factors such as contrast injection time and image acquisition time could affect the reconstructed difficulties and results. Therefore, the AI-assisted reconstruction method is superior to traditional method of CT post-processing in preoperative preparation, intraoperative guidance, and postoperative patient recovery in our study. With further advancement in AI technology, it may be possible to directly formulate the surgical plan for pulmonary nodules’ resection by AI, offering patients more precise and efficient medical services (30-32).

In terms of principles and workflow, traditional 3D reconstruction represents a semi-automated, human-assisted approach that requires manual intervention at certain stages, preserving a degree of human control and decision-making, with its fundamental principle involving software-based extraction and segmentation of two-dimensional (2D)-CT image data (including CT values) supplemented by manual bronchial/vascular identification to generate pulmonary 3D models, thus characterized by requiring specialized expertise, multi-step operations, anatomical matching, and contrast-enhanced imaging; whereas the novel AI-based 3D reconstruction constitutes a fully automated approach where advanced computational techniques (e.g., machine/deep learning) minimize or eliminate human intervention by autonomously processing the entire workflow from data input to final reconstruction, with contemporary algorithms capable of generating higher-precision pulmonary anatomical images within shorter timeframes—exemplified by AI-powered early lung cancer surgical decision systems employing 3D convolutional neural networks and other algorithms to automatically perform comprehensive image processing (including nodule diameter, density, volume, CT values, malignancy probability, and early adenocarcinoma subtyping analysis) alongside complete 3D model generation, although such automated methods may not consistently account for nuanced adjustments achievable by human operators and currently require extensive training data, model optimization, and clinical validation before achieving unsupervised clinical applicability, their segmentation accuracy can potentially match expert manual segmentation, making them invaluable for large-scale image analyses where manual processing is impractical.

In terms of cost and economic benefits, the novel AI-based surgical planning approach helps reduce deployment costs, decrease patient expenses, and generates socioeconomic benefits for both healthcare systems and society. First, with the increasing maturity of AI models, the upfront costs of AI-based surgical planning have been significantly reduced, whereas traditional CT post-processing workstations remain prohibitively expensive. Second, this approach eliminates the need for contrast-enhanced scans, reducing diagnostic costs for patients. Additionally, by shortening hospital stays and minimizing postoperative complications, it further decreases hospitalization expenses, thereby reducing the overall financial burden on both national healthcare systems and society.

There are several limitations in this study. Firstly, pleural adhesion and lymph node dissection significantly impact the duration and complexity of surgery. The current AI-assisted surgical simulation system does not yet assess the extent of pleural adhesions or perform automatic lymph node reconstruction, highlighting a need for optimization of large data models to address these deficiencies. Secondly, the study’s single-center design may limit the generalizability of its findings to broader populations. To enhance the external validity of the results, future studies should adopt a multi-center approach with a more diverse cohort of patients. Future research should focus on overcoming these challenges and further promoting the clinical application of AI-assisted surgical simulation in thoracoscopic lung nodule resection. With advancements in AI technology, it may become possible for AI to directly formulate surgical plans for pulmonary nodule resection, providing patients with more precise and efficient medical services.

Conclusions

The newly developed AI-assisted surgical simulation system could efficiently guide thoracoscopic pulmonary segmentectomies and has a potential clinical value. This novel AI-assisted method was superior to traditional method of CT bronchography and angiography in preoperative preparation, intraoperative guidance, and postoperative patient recovery.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-375/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-375/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-375/prf

Funding: This study was supported by the National Natural Science Foundation of China (grant No. 82171931).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-375/coif). The AI-assisted reconstruction software was invented by ShuKun Technology Co., Ltd. J.H. is employed as a clinical scientist in medical industry at ShuKun (Beijing) Technology Co., Ltd. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The research protocol was approved by the Institutional Ethics Committee of The Affiliated Panyu Central Hospital of Guangzhou Medical University (IRB: PYRC-2023-392; date of approval: December 15, 2023). Informed consent was obtained from all participants in the prospective cohort; yet, it was waived in the retrospective cohort.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Saji H, Okada M, Tsuboi M, et al. Segmentectomy versus lobectomy in small-sized peripheral non-small-cell lung cancer (JCOG0802/WJOG4607L): a multicentre, open-label, phase 3, randomised, controlled, non-inferiority trial. Lancet 2022;399:1607-17. [Crossref] [PubMed]
2. Altorki N, Wang X, Kozono D, et al. Lobar or Sublobar Resection for Peripheral Stage IA Non-Small-Cell Lung Cancer. N Engl J Med 2023;388:489-98. [Crossref] [PubMed]
3. Bian C, Fu C, Xue W, et al. Three-dimensional reconstruction for determining positional indications of pulmonary segmentectomy/subsegmentectomy for ground glass opacity-dominant clinical T1a-bN0 non-small cell lung cancer. Transl Lung Cancer Res 2024;13:2947-57. [Crossref] [PubMed]
4. Puca MA, Marella A, Marvulli M, et al. 3D Computed Tomography Reconstruction for Pre-Operative Diagnosis of Anomalous Venous Drainage of the Right Upper Lobe. Thorac Cancer 2025;16:e15525. [Crossref] [PubMed]
5. Ma X, Lu T, Qin D, et al. Analysis of pulmonary artery variation based on 3D reconstruction of CT angiography. Front Physiol 2023;14:1156513. [Crossref] [PubMed]
6. Wang LF, Zhao L, Lv CS, et al. Anatomical type analysis of right interlobar artery based on chest thin-slice CT scan and three-dimensional reconstruction. J Cardiothorac Surg 2022;17:328. [Crossref] [PubMed]
7. Zheng Z, Ren M, Li B, et al. Application Value of Artificial Intelligence-assisted Three-dimensional Reconstruction in Planning Thoracoscopic Segmentectomy. Zhongguo Fei Ai Za Zhi 2023;26:515-22. [Crossref] [PubMed]
8. Hashimoto K, Kato D, Ichinose J, et al. A prospective study of a training program for bronchial sleeve resection using operable 3-dimensional models. JTCVS Tech 2024;27:217-24. [Crossref] [PubMed]
9. Zhang X, Yang D, Li L, et al. Application of three-dimensional technology in video-assisted thoracoscopic surgery sublobectomy. Front Oncol 2024;14:1280075. [Crossref] [PubMed]
10. Ujiie H, Chiba R, Yamaguchi A, et al. Developing a Virtual Reality Simulation System for Preoperative Planning of Robotic-Assisted Thoracic Surgery. J Clin Med 2024;13:611. [Crossref] [PubMed]
11. Bagan P, Aissa K, Essid R, et al. Three-Dimensional Holographic-Guided Robotic Lung Segmentectomy for Deep Pulmonary Nodules: Technique and Initial Results. Surg Innov 2024;31:618-21. [Crossref] [PubMed]
12. Cannone G, Verzeletti V, Busetto A, et al. Three-Dimensional Imaging-Guided Lung Anatomic Segmentectomy: A Single-Center Preliminary Experiment. Medicina (Kaunas) 2023;59:2079. [Crossref] [PubMed]
13. Laven IEWG, Verkoulen GHJM, Verkoulen KCHA, et al. Three-Dimensional Lung Reconstructions for Preoperative Planning of Uniportal Video-Assisted Thoracoscopic Segmentectomies Using Visible Patient Software. Innovations (Phila) 2025;20:87-95. [Crossref] [PubMed]
14. Li X, Zhang S, Luo X, et al. Accuracy and efficiency of an artificial intelligence-based pulmonary broncho-vascular three-dimensional reconstruction system supporting thoracic surgery: retrospective and prospective validation study. EBioMedicine 2023;87:104422. [Crossref] [PubMed]
15. Kadry S. Three-dimensional reconstruction system in pulmonary broncho vascular surgery using AI. EBioMedicine 2023;89:104495. [Crossref] [PubMed]
16. Chen X, Wang Z, Qi Q, et al. A fully automated noncontrast CT 3-D reconstruction algorithm enabled accurate anatomical demonstration for lung segmentectomy. Thorac Cancer 2022;13:795-803. [Crossref] [PubMed]
17. Kiemen AL, Damanakis AI, Braxton AM, et al. Tissue clearing and 3D reconstruction of digitized, serially sectioned slides provide novel insights into pancreatic cancer. Med 2023;4:75-91. [Crossref] [PubMed]
18. Chen H, Lin K, Xiao G, et al. A Study on the Efficacy of Three-Dimensional CT Bronchial Angiography in Thoracoscopic Precision Resection of Lung Segments. Altern Ther Health Med 2023;29:335-9.
19. Wang X, Wang Q, Zhang X, et al. Application of three-dimensional (3D) reconstruction in the treatment of video-assisted thoracoscopic complex segmentectomy of the lower lung lobe: A retrospective study. Front Surg 2022;9:968199. [Crossref] [PubMed]
20. Liu B, Wang K, Ye S. Clinical observation of three-dimensional reconstruction in thoracoscopic segmental pneumonectomy. Technol Health Care 2025;33:195-203. [Crossref] [PubMed]
21. Najafi H, Savoji K, Mirzaeibonehkhater M, et al. A Novel Method for 3D Lung Tumor Reconstruction Using Generative Models. Diagnostics (Basel) 2024;14:2604. [Crossref] [PubMed]
22. Chen X, Xu H, Qi Q, et al. AI-based chest CT semantic segmentation algorithm enables semi-automated lung cancer surgery planning by recognizing anatomical variants of pulmonary vessels. Front Oncol 2022;12:1021084. [Crossref] [PubMed]
23. Takamoto T, Ban D, Nara S, et al. Automated Three-Dimensional Liver Reconstruction with Artificial Intelligence for Virtual Hepatectomy. J Gastrointest Surg 2022;26:2119-27. [Crossref] [PubMed]
24. Yang D, Miao Y, Liu C, et al. Advances in artificial intelligence applications in the field of lung cancer. Front Oncol 2024;14:1449068. [Crossref] [PubMed]
25. Vervoorn MT, Wulfse M, Mohamed Hoesein FAA, et al. Application of three-dimensional computed tomography imaging and reconstructive techniques in lung surgery: A mini-review. Front Surg 2022;9:1079857. [Crossref] [PubMed]
26. Nakazawa S, Nagashima T, Kawatani N, et al. Anatomy of the lung revisited by 3D-CT imaging. Video Assist Thorac Surg 2023;8:17. [Crossref] [PubMed]
27. Tsukamoto Y, Yabe M, Ishikawa A, et al. Anomalous bronchus and pulmonary artery in a patient who underwent subsegmentectomy of right S6a + b. Gen Thorac Cardiovasc Surg Cases 2023;2:35. [Crossref] [PubMed]
28. Fan K, Feng J, Li Y, et al. Application of three-dimensional reconstruction of left upper lung lobes in anatomical segmental resection. Thorac Cancer 2022;13:1176-83. [Crossref] [PubMed]
29. Zhu S, Xu W, Li Z, et al. Branching patterns and variations of the bronchus and blood vessels in the superior segment of the right lower lobe: a three-dimensional computed tomographic bronchography and angiography study. J Thorac Dis 2023;15:6879-88. [Crossref] [PubMed]
30. Li Z, Li R, Liu L, et al. The utility and feasibility of three-dimensional reconstruction in surgical planning for multiple pulmonary nodules: a prospective self-controlled study. Transl Lung Cancer Res 2025;14:194-208. [Crossref] [PubMed]
31. Shi J, Xing F, Liu Y, et al. Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules. J Vis Exp 2023;(200).
32. Rui W, Yuhang S, Yang L, et al. A new method for evaluating lung volume: AI-3D reconstruction. Front Physiol 2023;14:1217411. [Crossref] [PubMed]