Patient-specific three-dimensional lung models and digital twin technologies in thoracic surgery: current applications, technical challenges, and future perspectives—a narrative review

Introduction

Patient-specific three-dimensional (3D) lung models have been employed in thoracic surgery for reconstructing, on an individual-patient basis, anatomical information derived from preoperative computed tomography (CT), including pulmonary vessels, bronchi, tumors, and lung segments, and for visualizing spatial relationships that are difficult to appreciate on two-dimensional (2D) images. Traditionally, they have primarily been used to support preoperative anatomical understanding, assessment of the planned resection area, and surgical simulation. However, more recently, attempts have been made to extend the use of such 3D information beyond preoperative planning by combining it with intraoperative display, virtual, augmented, mixed, or extended reality (VR/AR/MR/XR), and image registration, to support intraoperative anatomical recognition and localization.

Against this background, sublobar resection, particularly segmentectomy, has become increasingly relevant for early-stage lung cancer. The widespread use of low-dose CT screening and high-resolution CT has increased the detection of early-stage disease, and large randomized trials have established the clinical value of low-dose CT screening (1,2). Accordingly, sublobar resection has gained importance as a means of preserving pulmonary function while maintaining oncological radicality. In the JCOG0802/WJOG4607L trial, segmentectomy demonstrated favorable survival outcomes compared with lobectomy for small peripheral non-small cell lung cancer (3).

Accurate assessment of tumor location, surgical margins, segmental vessels and bronchi, and the intersegmental plane is crucial in sublobar resection, including segmentectomy. However, 2D CT images and conventional multiplanar reconstruction alone may be insufficient for an intuitive understanding of 3D anatomy, particularly for lesions in anatomically complex locations. This anatomical complexity directly influences preoperative determination of the resection area, intraoperative vascular management, bronchial division, intersegmental plane formation, and surgical margin assessment (4,5).

To address these challenges, patient-specific 3D reconstruction, 3D CT bronchovascular angiography (3D-CTBA), and 3D-printed models have been employed in thoracic surgery. These technologies transform patient-specific anatomical information into a practical information platform that supports preoperative simulation, surgical planning, margin assessment, and communication within the surgical team (4-6). Accordingly, 3D models are no longer considered merely as educational or explanatory tools, but are increasingly positioned as practical planning tools in thoracic surgery, particularly for pulmonary segmentectomy (4).

However, the significance of patient-specific 3D lung models is not limited to preoperative planning for segmentectomy. In video-assisted thoracoscopic surgery and robot-assisted thoracic surgery, surgeons must proceed while identifying the pulmonary hilum, segmental vessels, bronchi, and tumor location through a restricted endoscopic view. Therefore, connecting 3D information generated from preoperative CT with the intraoperative view, surgical manipulation, and the surgeon’s decision-making process is an important challenge. Reported approaches include intraoperative display of 3D models, model overlay onto the operative field, intersegmental plane identification using indocyanine green fluorescence, and integration with robotic surgical platforms (7-10). Thus, 3D models play a role in expanding from static image-based tools for preoperative anatomical understanding to intraoperative decision-support systems.

However, static 3D models according to preoperative CT have inherent limitations. During surgery, the lung undergoes substantial deformation due to collapse, positional changes, traction, and compression. Therefore, accurately mapping the positions of vessels, bronchi, and tumors on preoperative images to the intraoperative lung needs image registration, compensation for lung deformation, intraoperative image acquisition, and real-time updating (9,10). This issue represents a central barrier to advancing patient-specific 3D lung models from static preoperative images to deformation-aware intraoperative navigation. In the future, 3D models may evolve beyond fixed anatomical models developed from preoperative images into dynamic models that integrate intraoperative imaging, surgical manipulation, robotic vision, sensor information, artificial intelligence (AI)-based structure recognition, and lung deformation models to update the current lung shape and deep anatomical structure positions. This direction may provide the foundation for lung digital twin technology in thoracic surgery.

Recent reviews have focused on AI-driven 3D reconstruction for preoperative assessment and planning (11), whereas intraoperative XR, deformation-aware registration, and digital-twin technologies have received less attention. This review examines the continuum from static preoperative models to dynamic intraoperative decision support, summarizes current evidence and technical limitations, and identifies priorities for clinical implementation. We present this article in accordance with the Narrative Review reporting checklist (available at https://shc.amegroups.com/article/view/10.21037/shc-2026-0024/rc).

Methods

This narrative review employed a structured literature search to identify studies on patient-specific 3D lung reconstruction, XR-based surgical guidance, deformation-aware registration, and digital-twin technologies in thoracic surgery.

PubMed was searched for English-language articles published from January 1, 2023 through August 19, 2026. The search strategy comprised four branches, each combining a main search block covering the surgical setting and patient-specific modeling with a subdomain-specific search block using the Boolean operator AND. The subdomains were: (I) clinical utility, comparative evaluation, or validation; (II) VR, AR, MR, or XR; (III) lung deformation, nonrigid registration, or intraoperative registration; and (IV) the definition or implementation of digital twins. Because evidence on clinically validated digital twins in thoracic surgery remains limited, the fourth branch also included studies from other surgical specialties that addressed the definition or implementation of dynamic patient-specific models.

The searches and study selection were conducted by one reviewer (S.M.) according to predefined eligibility criteria. The four branches yielded 10, 6, 3, and 2 records, respectively, for a total of 21 records before deduplication. After one duplicate PubMed identifier (PMID) was removed, 20 unique articles meeting the eligibility criteria were included in the core evidence set. Eligible publications included adult clinical, comparative, and validation studies, systematic reviews and meta-analyses, and, for the digital-twin branch, conceptual or perspective articles addressing surgical digital-twin definitions or implementation; case reports and study protocols were excluded. Landmark and contextual references were added where needed to describe foundational concepts or earlier technological developments but were not included in the 20-article core evidence set. Table 1 summarizes the literature search methods. The complete Boolean search strings, the number of records retrieved from each branch, and the corresponding PMIDs are provided in Table S1.

Overview of the core evidence set

Table 2 summarizes the characteristics and principal findings of the 20 studies included in the core evidence set. The evidence base encompassed randomized and nonrandomized clinical studies, prospective and retrospective validation studies, technical feasibility studies, and a systematic review and meta-analysis. Clinical evidence was relatively well developed for preoperative 3D reconstruction and selected XR-based localization applications. By contrast, deformation-aware navigation was supported only by small technical validation studies, while the digital-twin literature remained largely conceptual and cross-specialty in scope.

Clinical evidence for patient-specific 3D reconstruction in lung resection

The clinical significance of patient-specific 3D models in pulmonary resection lies not only in displaying anatomical structures in 3D but in transforming information derived from preoperative imaging into actionable data for defining the extent of resection and supporting intraoperative orientation. Tumor localization, margin assessment, and recognition of patient-specific bronchovascular variation are particularly important. Accordingly, 3D reconstruction should be evaluated as a surgical planning and decision-support tool rather than solely as a visualization technique.

Contemporary evidence suggests that the clinical value of 3D reconstruction in pulmonary resection is not uniform and may depend on procedural complexity. A systematic review and meta-analysis of 10 comparative studies involving 1,424 patients found that patient-specific 3D reconstruction and virtual simulation were associated with lower risks of inadequate surgical margins [odds ratio (OR), 0.09; 95% confidence interval (CI): 0.02–0.50] and postoperative complications (OR, 0.53; 95% CI: 0.38–0.74). In complex segmentectomy, operative time was reduced by 10.85 min and blood loss by 5.41 mL, whereas the number of lymph nodes resected, chest drainage, chest tube duration, and postoperative hospital stay were similar overall (13). In contrast, a three-center randomized controlled trial of 191 patients found no difference in median operative time, postoperative complications, or other perioperative outcomes between preoperative 3D reconstruction and standard CT (19). This trial excluded nodules reaching segment borders; the resulting concentration of relatively straightforward cases may have reduced the ability to detect an incremental benefit from 3D reconstruction.

Observational studies also provide supportive, although less definitive, evidence. In a retrospective cohort of 265 thoracoscopic segmentectomies, preoperative 3D reconstruction was associated with less blood loss, a shorter postoperative hospital stay, greater lymph-node sampling, and fewer 30-day complications than 2D CT. However, the nonrandomized groups differed in baseline and procedural characteristics (17). In 49 patients with multiple pulmonary nodules, a prospective self-controlled study found that 3D reconstruction improved planning success for both novice and experienced surgeons (16). Because conventional CT assessment always preceded 3D assessment, bias related to assessment order cannot be excluded.

Anatomical recognition appears to be a particularly consistent use case. A CT-based 3D bronchovascular mapping study of 2,007 patients identified clinically important right middle-lobe venous variants and confirmed these classifications intraoperatively (14). In a prospective pilot using 3D Slicer in 13 patients, previously unrecognized anatomical variants and changes in the procedure or transection planes were each reported in 62%; however, one A6 branch was omitted from the reconstruction (18). Furthermore, a prospective diagnostic-accuracy study in 447 patients showed only moderate agreement between preoperative 3D fissure grading and intraoperative findings (exact concordance, 55.42%; kappa =0.48) (12). These findings support the clinical utility of 3D mapping while emphasizing that branch-level completeness and anatomical fidelity cannot be assumed.

Specific computational methods can reduce the time and operator dependence involved in model generation. A fully automated convolutional neural network-based reconstruction system trained on 500 cases achieved a Dice coefficient of 89.2%. In prospective validation, accuracy against intraoperative anatomy reached 97.0% for bronchi, 99.1% for arteries, and 98.8% for veins, and reconstruction was 14.2 min faster than manual Mimics reconstruction (21). In a multicenter multireader crossover study, AI-driven 3D reconstruction improved anatomical-variant and procedure-selection accuracy by 8%, reduced the respective error counts by 41% and 35%, and shortened planning time by 25% (15). These studies illustrate two specific AI functions: convolutional neural network-based segmentation for model generation and algorithm-assisted identification of anatomical variants and procedural options. They should not be interpreted as evidence of autonomous intraoperative decision-making, and broader external and outcome-based validation remains necessary.

Physical 3D printing remains an alternative format for representing patient-specific anatomy. In an earlier quasi-randomized trial of 89 patients, adding a printed model to 3D CT reconstruction was associated with fewer changes in surgical approach, fewer conversions to lobectomy, shorter operative time, and less blood loss, but not fewer postoperative complications (6). Beyond preoperative planning, postoperative CT-based 3D registration has also been used to quantify resection margins. In a 12-patient feasibility study, target registration error was less than 2.5 mm in all cases, enabling calculation of postoperative margin distance, although external validation and correlation with recurrence were not assessed (20).

Overall, clinical evidence is strongest for improving anatomical understanding, identifying bronchovascular variants, and supporting procedural planning, particularly in complex segmentectomy. Evidence for universal improvement in perioperative outcomes is inconsistent, and many favorable results are derived from observational cohorts, planning studies, or reader studies. Future evaluations should define case complexity, verify branch-level anatomical accuracy, report reconstruction failures, and use clinically relevant outcomes. The next translational step is to connect these predominantly static preoperative models with the operative field through XR and image registration.

XR-assisted intraoperative guidance: visualization, localization, and surgical-field overlay

Patient-specific 3D reconstruction provides the anatomical substrate for XR-assisted guidance. VR presents patient-specific anatomy within an immersive virtual environment, whereas AR overlays virtual information onto the operative field or endoscopic image. MR allows more interactive handling of virtual content within the real-world environment. In thoracic surgery, the clinical value of these technologies depends less on the display modality itself than on whether they improve lesion localization, anatomical recognition, or intraoperative decision-making.

The simplest intraoperative application of preoperative 3D information is to display reconstructed anatomy in the operating room or on the robotic console and refer to it during dissection. A reported workflow for robot-assisted pulmonary segmentectomy incorporated patient-specific 3D models into the da Vinci console (10). More recently, Ujiie et al. developed a patient-specific CT-derived VR system for preoperative planning and TilePro display during robot-assisted thoracic surgery. Although qualitative correspondence between the reconstructed and operative anatomy was illustrated, the report did not provide a clear cohort denominator or cohort-level quantitative accuracy or outcome data (25). Accordingly, console display should currently be regarded as an orientation aid rather than validated navigation.

Current clinical evidence is strongest for XR-assisted pulmonary nodule localization rather than real-time overlay of the collapsed lung. In a five-center randomized noninferiority trial, 296 patients with suspected early-stage lung cancer were randomized and 270 were included in the modified intention-to-treat analysis. AR-guided localization achieved successful resection in 98.5% of patients, which was noninferior to 99.3% with conventional CT-guided localization. Radiation exposure, pain, puncture time, and the interval from localization to incision were also lower with AR guidance (22). These findings demonstrate clinical utility within a defined localization workflow but do not establish the accuracy of registration after lung collapse or surgical manipulation.

Supporting evidence is provided by prospective single-arm studies. In 106 patients undergoing VATS nodule resection, AR-guided localization succeeded in 95.28% (95% CI: 91.25–99.32%), with a median localization time of 3.1 min and no localization-related complications (23). In a smaller 10-patient pilot, HoloLens-based AR navigation with an individualized 3D plan and LungBrella marker achieved localization in seven patients, with no complications reported (26). The 30% failure rate in this pilot emphasizes that feasibility in a small uncontrolled series should not be equated with reproducible clinical performance.

Patient-specific holograms may also improve recognition of pulmonary vascular anatomy. In a prospective comparative study of 52 lobectomy cases, interreader agreement in pulmonary artery counts was 98% with AR holograms compared with 85% using standard CT (24). However, this was a single-center study involving only two readers, and the endpoints assessed anatomical accuracy rather than clinical outcomes.

Operative-field overlay represents a more technically demanding application. At our institution, Izawa et al. reported a noninvasive XR-assisted lung tumor marking method for minimally invasive sublobar resection. Patient-specific models of the pulmonary vessels, bronchi, lung segments, and tumors were displayed on HoloLens 2, manually aligned using the interlobar fissure, pulmonary artery, and bronchus as landmarks, and superimposed onto the thoracoscopic image. The composite image was returned to the da Vinci console (Figure 1). Overlay was feasible in all 20 cases, and accurate marking concordance was achieved in 18 (90%) (8). Because alignment remained manual, this study demonstrated workflow feasibility rather than automatic registration.

Figure 1 Representative intraoperative view of extended reality-assisted pulmonary tumor marking at Kobe University Hospital. A patient-specific 3D lung model from preoperative thin-slice CT was overlaid onto the thoracoscopic image and returned to the da Vinci console. 3D, three-dimensional; CT, computed tomography.

In a separate technical-feasibility study, Peek et al. used a stereo-camera point cloud, semiautomated 3D registration, and AR overlay in eight patients undergoing VATS or robot-assisted pulmonary resection. Overlay creation took approximately 5 min. Point-cloud construction showed a median absolute measurement error of 0.20 mm, whereas landmark-based registration had a mean root-mean-square deviation of 6.6 mm (range, 4.45–9.60 mm); adequate overlay and pulmonary artery identification were achieved in five of eight patients (9). The three inadequate overlays and the distinction between surface error and deep-target error underscore the need for clinically relevant validation metrics.

Overall, current evidence supports XR as an aid for pulmonary nodule localization, anatomical reference, and intraoperative orientation. Evidence for continuous and deformation-aware operative-field navigation remains preliminary. Clinical implementation still faces challenges in registration accuracy, calibration, latency, device ergonomics, and workflow integration. These challenges are amplified in thoracic surgery because one-lung ventilation, lung collapse, traction, compression, and parenchymal transection produce substantial changes from the lung configuration captured on preoperative CT. Reliable surgical-field overlay therefore requires registration and compensation for lung deformation, as discussed in the next section.

Lung deformation and deformation-aware registration

Organ deformation is the principal limitation in transferring 3D models derived from preoperative CT to the operative field. Deformation-aware registration is the process of adapting or updating a preoperative patient-specific model to the current intraoperative configuration. Unlike rigid registration, which assumes that the anatomy retains its shape, deformation-aware methods must account for local changes in surface geometry and the positions of internal structures. Such updating is required to move from static 3D visualization to reliable intraoperative navigation.

The lung is a challenging registration target because its geometry changes markedly between preoperative inflation and intraoperative collapse. Changes in position, deflation, and sliding along interlobar fissures and the chest wall produce substantial heterogeneous deformation (27,28).

Clinical imaging studies demonstrate both the magnitude and complexity of lung deformation. Alvarez et al. compared preoperative supine CT with intraoperative cone-beam CT acquired in the lateral decubitus position under general anesthesia before surgical incision in six patients undergoing thoracoscopic surgery. Their deformable registration incorporated sliding motion along the fissures and chest wall. The average target-registration error was less than 1 mm, although lung displacement exceeded 40 mm and interlobar or chest-wall sliding exceeded 30 mm (28). These findings show that rigid registration is insufficient even before surgical manipulation. However, the study included only six patients and did not prospectively evaluate navigation or clinical outcomes.

Clinical feasibility studies further illustrate the difference between geometric surface alignment and clinically reliable navigation. In the eight-patient study by Peek et al., the median absolute point-cloud measurement error was 0.20 mm, whereas the mean landmark-registration root-mean-square deviation (RMSD) was 6.6 mm (range, 4.45–9.60 mm); adequate overlay and pulmonary artery identification were achieved in only five patients (9). These findings indicate that accurate point-cloud measurement alone does not ensure reliable anatomical registration or localization of deep targets.

Mank et al. subsequently evaluated AI-based segmentation combined with a finite-element four-dimensional (4D) deformable lung model using retrospectively collected operative video from 30 pulmonary resections. The reported detection accuracies were 0.987 for arteries, 0.976 for veins, and 1.000 for bronchi; processing latency was 0.23±0.06 s, and 4D reconstruction required 8.47 s (27). These results support the computational feasibility of dynamic model generation. However, the reported detection metrics were not spatial registration errors, and the system was not evaluated prospectively for surgical decision-making or patient outcomes.

The accuracy required depends on the intended intraoperative task. An error acceptable for lobar or segmental orientation may be inadequate for localizing a small deep nodule or identifying a small bronchovascular branch. Registration methods should therefore be assessed in relation to the intended clinical task, using not only mean target-registration error but also surface and deep-target errors, failure rates, processing time, and performance after surgical manipulation.

Overall, current evidence demonstrates that intraoperative lung deformation is substantial and that semiautomated registration and deformable 4D modeling are technically feasible. It does not yet establish continuous, prospectively validated navigation of the collapsed and manipulated lung. Future studies should use standardized anatomical targets, report unsuccessful registrations, quantify uncertainty, and test clinical utility in prospective multicenter settings. Deformation-aware model updating is the essential bridge between static 3D or XR guidance and a dynamic lung digital twin.

AI and lung digital twins: toward dynamic and predictive thoracic navigation

Deformation caused by lung collapse, surgical manipulation, and parenchymal transection represents a central limitation when connecting static 3D models based on preoperative CT to the intraoperative lung. Therefore, the next stage in the development of patient-specific 3D models is not simply to achieve more detailed 3D visualization, but to efficiently generate patient-specific anatomical information, associate it with intraoperative data, and update it according to the deformed state of the lung. The endpoint of this trajectory is a lung digital twin that incorporates intraoperative data, updates the current organ state, and potentially predicts the consequences of surgical manipulation.

In this trajectory, AI should not be regarded as a single, undifferentiated technology. Its potential contributions can be divided into four specific functions. First, AI can support automated segmentation, 3D reconstruction, and anatomical labeling of pulmonary vessels, bronchi, lobes, segments, and tumors from preoperative CT. Second, it can support intraoperative scene understanding by recognizing organs, vessels, instruments, and surgical phases in endoscopic or robotic images. Third, AI can assist registration between the preoperative model and intraoperative observations, accelerate lung deformation estimation, and update the model state. Fourth, it can help predict and display deep critical structures, transection lines, resection margins, or the effects of surgical actions in a form suitable for decision support. These functions are enabling components of a digital twin; an individual AI process does not itself constitute a digital twin.

These enabling components have already been evaluated individually in clinical studies. AI-driven 3D reconstruction has improved the accuracy and efficiency of preoperative planning (15), whereas AI-based segmentation combined with a finite-element 4D deformable lung model has demonstrated the technical feasibility of dynamically estimating the intraoperative lung state (27). However, these studies validate individual components rather than an integrated lung digital twin that is updated continuously throughout surgery.

During surgery, diverse and incomplete information is continuously generated, including endoscopic images, robotic vision, instrument positions, organ deformation, traction, bleeding, sensor data, and operative records. Surgical data science has been proposed as an interdisciplinary framework that collects, organizes, analyzes, and models data related to patients, surgeons, surgical teams, and devices, and applies these data to surgical decision-making and assistance (31,32). In thoracic surgery, Sadeghi et al. integrated deep learning-based segmentation of the pulmonary vasculature, lung lobes, and airways with a deformable 3D model based on finite-element simulation and AR display on the robotic console in two lower lobectomy cases (33). This study demonstrated the technical feasibility of an integrated workflow; however, real-time alignment required manual adjustment, and automatic registration and dynamic tissue tracking had not been achieved.

In this review, a digital twin is defined as a dynamic patient-specific digital model built on an individualized anatomical representation, continuously updated with real-world data from imaging, sensors, robotics, clinical records, or physiological measurements, and used for state estimation, prediction, and decision support. In this respect, a lung digital twin is distinct from a static 3D model or a single AR overlay without updating capability. Batool et al. characterized surgical digital twins as patient-specific models updated across the preoperative, intraoperative, and postoperative phases and identified the integration of imaging, sensors, robotics, and physiological data, together with interoperability, latency, cost, regulation, ethics, and equity, as major implementation considerations (29). Asciak et al. proposed a progression from conventional computer-assisted surgery toward digital twin-assisted surgery with real-time patient-specific updating and predictive capability (30). Consistent with these concepts, a scoping review emphasized a viewable, multidimensional, patient-specific digital replica that informs decisions, while noting substantial heterogeneity in the extent of dynamic updating and prediction across reported systems (34).

Cross-specialty studies do not provide direct evidence of clinical effectiveness for lung digital twins, but they are important for defining the required system architecture. Ding et al. positioned digital twins as a unifying framework for surgical data science and emphasized the role of geometric scene understanding, including segmentation, object detection, depth estimation, 3D reconstruction, and pose estimation, in constructing and updating intraoperative models (35). From this perspective, a digital twin is not simply a 3D display but an information platform that connects low-level scene recognition with model updating, prediction, and higher-level surgical decision-making. Nevertheless, the cross-specialty literature remains predominantly conceptual and does not validate accuracy, safety, or clinical outcomes in pulmonary resection.

Precedents for lung digital twins can also be found in respiratory disease research. Digital twins have been discussed as a framework for disease-progression prediction, personalized treatment, monitoring, and treatment-response prediction in chronic lung disease (36). Furthermore, a human lung digital twin integrating imaging, respiratory and circulatory physiology, molecular analyses, transcriptomics, metabolomics, and proteomics has been reported in the setting of ex vivo lung perfusion (37). These studies do not directly represent thoracic surgical navigation; however, they are important precedents showing that multimodal patient-specific data can be integrated to model the state of the lung and predict treatment response.

By contrast, the digital twin required for pulmonary resection differs from those envisioned for chronic lung disease or ex vivo lung perfusion. In thoracic surgery, a patient-specific anatomical model derived from preoperative CT must integrate intraoperative deformation caused by lung collapse and surgical manipulation, endoscopic images, robotic instrument positions, and, in the future, force or haptic information to dynamically update the current lung shape and the positions of deep structures. If such a model could predict the effects of intersegmental plane formation, bronchovascular division, resection margins, and residual pulmonary function, the lung digital twin could serve as a predictive surgical planning platform rather than solely as a navigation tool.

AI-assisted surgical autonomy should be considered a longer-term application beyond lung digital twins. A systematic review of Food and Drug Administration-cleared surgical robots reported that most approved systems remain at Level 1 robot assistance (38). In soft-tissue surgery, Saeidi et al. integrated imaging, tissue tracking, plan updating, and robotic control to achieve limited autonomous execution of small-bowel anastomosis in a porcine model (39). This study does not constitute direct evidence for autonomy in thoracic surgery, but it illustrates the components required for limited autonomy in deformable organs. Pulmonary resection is more challenging because nonrigid deformation of the collapsed lung, invisibility of deep structures, the risk of vascular injury, tumor-margin requirements, and interindividual anatomical variation coexist.

Accordingly, development in thoracic surgery is likely to proceed stepwise. The first stage is AI-assisted recognition, in which tumors, vessels, bronchi, intersegmental planes, and other critical structures are identified and presented to the surgeon. The next stage is AI-assisted navigation, including registration between the preoperative model and intraoperative view, AR overlay, uncertainty visualization, and warnings regarding critical areas. Task-level automation, including camera control, instrument tracking, or margin simulation, may follow, and conditional autonomy may eventually permit a robot to perform limited surgeon-approved tasks. In this pathway, a lung digital twin may function as a world model that represents the intraoperative environment, predicts the consequences of subsequent actions, and assesses safety margins. However, current evidence consists mainly of retrospective validation of individual technologies, small feasibility studies, and cross-specialty conceptual frameworks; the clinical validity of a continuously updated lung digital twin has not yet been established.

Quality assurance, standardization, and implementation barriers

Clinical implementation requires standardized quality assurance across image acquisition, segmentation, model generation, registration, and intraoperative updating. Validation should report reconstruction and registration failures, reproducibility, processing time, uncertainty, and task-specific accuracy after lung collapse or surgical manipulation (9,15,21,27,28). Implementation also depends on workflow integration, system interoperability, usability and ergonomics, cost-effectiveness, data governance, cybersecurity, and regulatory responsibility. These requirements are particularly important for digital-twin systems that continuously update patient-specific information for intraoperative decision-making (29,30,34,35).

Strengths and limitations of this review

A strength of this review is that it considers patient-specific 3D lung models in thoracic surgery not merely as imaging technologies, but as part of a broader continuum from preoperative planning to intraoperative decision-making, navigation, AI assistance, and future digital twin applications. This review integrates evidence on 3D reconstruction, XR-assisted intraoperative guidance, deformation-aware registration, AI-assisted modeling, and digital twin frameworks, and positions these technologies within a common clinical challenge: how to translate patient-specific anatomical information into actionable intraoperative support.

This review has several limitations. Much of the available evidence is derived from single-center studies, retrospective analyses, small cohorts, or technical feasibility studies, and the effects of these technologies on long-term clinical outcomes and cost-effectiveness remain insufficiently validated. Terminology, validation methods, and assessment metrics also vary considerably across studies, limiting direct comparison. In addition, the structured search was restricted to PubMed, English-language articles, and studies published from 2023 onward. Although earlier landmark studies and relevant background literature were added where necessary, this was a narrative rather than a systematic review, and some relevant studies may not have been identified.

Conclusions

Patient-specific 3D lung models have been used as practical tools to support preoperative anatomical assessment and surgical planning for pulmonary resection, and their applications are expanding toward intraoperative orientation and XR-assisted surgical guidance. Current evidence is strongest for preoperative 3D reconstruction and selected XR-based lesion-localization applications, whereas continuous deformation-aware intraoperative navigation and clinically validated lung digital twins have not yet been established. Because the intraoperative lung undergoes substantial nonrigid deformation due to collapse and surgical manipulation, further advances will require reliable registration, intraoperative model updating, and integration of AI with intraoperative data. Before these technologies are introduced into routine clinical practice, their task-specific accuracy, safety, compatibility with surgical workflows, and effects on clinical outcomes must be evaluated.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://shc.amegroups.com/article/view/10.21037/shc-2026-0024/rc

Peer Review File: Available at https://shc.amegroups.com/article/view/10.21037/shc-2026-0024/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://shc.amegroups.com/article/view/10.21037/shc-2026-0024/coif). The 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 representative intraoperative image shown in Figure 1 was obtained from a study approved by the Kobe University Hospital Institutional Review Board (No. B-230126), and the requirement for written informed consent for publication was waived by the Institutional Review Board.

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. Aberle DR, Adams AM, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011;365:395-409. [Crossref] [PubMed]
2. de Koning HJ, van der Aalst CM, de Jong PA, et al. Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial. N Engl J Med 2020;382:503-13. [Crossref] [PubMed]
3. 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]
4. Xiang Z, Wu B, Zhang X, et al. Preoperative Three-Dimensional Lung Simulation Before Thoracoscopic Anatomical Segmentectomy for Lung Cancer: A Systematic Review and Meta-Analysis. Front Surg 2022;9:856293. [Crossref] [PubMed]
5. Wang M, Lv H, Wu T, et al. Application of three-dimensional computed tomography bronchography and angiography in thoracoscopic anatomical segmentectomy of the right upper lobe: A cohort study. Front Surg 2022;9:975552. [Crossref] [PubMed]
6. Chen Y, Zhang J, Chen Q, et al. Three-dimensional printing technology for localised thoracoscopic segmental resection for lung cancer: a quasi-randomised clinical trial. World J Surg Oncol 2020;18:223. [Crossref] [PubMed]
7. Huang R, Du J, Xu G, et al. Application of three-dimensional printed models with near-infrared fluorescence technology in video-assisted thoracoscopic surgery segmentectomy: a single-center propensity-score matching analysis. J Thorac Dis 2024;16:4474-86. [Crossref] [PubMed]
8. Izawa R, Tane S, Doi T, et al. Novel, noninvasive, extended reality-aided pulmonary tumor marking method for sublobar resection in minimally invasive surgery. JTCVS Tech 2025;32:157-62. [Crossref] [PubMed]
9. Peek JJ, Zhang X, Hildebrandt K, et al. A novel 3D image registration technique for augmented reality vision in minimally invasive thoracoscopic pulmonary segmentectomy. Int J Comput Assist Radiol Surg 2025;20:787-95. [Crossref] [PubMed]
10. Baste JM, Soldea V, Lachkar S, et al. Development of a precision multimodal surgical navigation system for lung robotic segmentectomy. J Thorac Dis 2018;10:S1195-204. [Crossref] [PubMed]
11. Miao Y, Yu Q, Zhang Z, et al. Artificial Intelligence-Driven Three-Dimensional Reconstruction in Lung Cancer Surgery: Current Status and Future Perspectives. ANZ J Surg 2026;96:1101-8. [Crossref] [PubMed]
12. Mo X, Zhang H, Chen C, et al. Prediction and intraoperative validation of pulmonary interlobar fissure development based on high-resolution thin-slice computed tomography and multiplanar reconstruction three-dimensional reconstruction technology: a prospective study. J Thorac Dis 2026;18:459. [Crossref] [PubMed]
13. He H, Yu C, Yang Y, et al. Three-Dimensional Reconstruction and Virtual Simulation of Patient-Specific Anatomy for Procedural Planning in Thoracoscopic Segmentectomy: A Systematic Review and Meta-Analysis. Eur J Cardiothorac Surg 2025;67:ezaf283. [Crossref] [PubMed]
14. Hao T, Xu L, Jiang SM, et al. Anatomical Classification and Clinical Significance of Right Middle Lobe Vein Confluence Variations in Right Upper Lobectomy: A Three-Dimensional Reconstruction Study. Clin Exp Pharmacol Physiol 2025;52:e70058. [Crossref] [PubMed]
15. Chen X, Dai C, Peng M, et al. Artificial intelligence driven 3D reconstruction for enhanced lung surgery planning. Nat Commun 2025;16:4086. [Crossref] [PubMed]
16. 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]
17. He H, Wang P, Zhou H, et al. The advantages of preoperative 3D reconstruction over 2D-CT in thoracoscopic segmentectomy. Updates Surg 2024;76:2875-83. [Crossref] [PubMed]
18. Laven IEWG, Oosterhoff VPS, Franssen AJPM, et al. Evaluating three-dimensional lung reconstructions for thoracoscopic lung resections using open-source software: a pilot study. Transl Lung Cancer Res 2024;13:1595-608. [Crossref] [PubMed]
19. Chen K, Niu Z, Jin R, et al. Three-dimensional reconstruction computed tomography in thoracoscopic segmentectomy: a randomized controlled trial. Eur J Cardiothorac Surg 2024;66:ezae250. [Crossref] [PubMed]
20. Lin YH, Chen LW, Wang HJ, et al. Quantification of Resection Margin following Sublobar Resection in Lung Cancer Patients through Pre- and Post-Operative CT Image Comparison: Utilizing a CT-Based 3D Reconstruction Algorithm. Cancers (Basel) 2024;16:2181. [Crossref] [PubMed]
21. 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]
22. Song Z, Wang Z, Yao H, et al. Single-Encounter Augmented Reality-Guided Localization for Resection of Suspected Early-Stage Lung Cancer: A Randomized Clinical Trial. JAMA Surg 2026;161:871-9. [Crossref] [PubMed]
23. Mei J, Liu Z, Wu Q, et al. A Prospective, Multicenter, Single-Arm Study to Evaluate the Efficacy and Safety of Augmented Reality Navigation-Guided Pulmonary Nodule Localization System. Eur J Cardiothorac Surg 2026;68:ezag049. [Crossref] [PubMed]
24. Petrella F, Rizzo SMR, Rampinelli C, et al. Assessment of pulmonary vascular anatomy: comparing augmented reality by holograms versus standard CT images/reconstructions using surgical findings as reference standard. Eur Radiol Exp 2024;8:57. [Crossref] [PubMed]
25. 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]
26. Li C, Ji A, Jian Z, et al. Augmented reality navigation-guided intraoperative pulmonary nodule localization: a pilot study. Transl Lung Cancer Res 2023;12:1728-37. [Crossref] [PubMed]
27. Mank QJ, Kieft T, Siregar S, et al. Deformable lung models for anatomical lung resections: The introduction of simulated reality for imaging guidance. JTCVS Tech 2026;35:102147. [Crossref] [PubMed]
28. Alvarez P, Chabanas M, Sikora S, et al. Measurement and Analysis of Lobar Lung Deformation After a Change of Patient Position During Video-Assisted Thoracoscopic Surgery. IEEE Trans Biomed Eng 2023;70:931-40. [Crossref] [PubMed]
29. Batool Z, Wu Q, Li J, et al. Digital twin-assisted surgery: technological architecture, integration across surgical phases, implementation challenges and future directions. Int J Surg 2026; [Crossref] [PubMed]
30. Asciak L, Kyeremeh J, Luo X, et al. Digital twin assisted surgery, concept, opportunities, and challenges. NPJ Digit Med 2025;8:32. [Crossref] [PubMed]
31. Maier-Hein L, Vedula SS, Speidel S, et al. Surgical data science for next-generation interventions. Nat Biomed Eng 2017;1:691-6. [Crossref] [PubMed]
32. Maier-Hein L, Eisenmann M, Sarikaya D, et al. Surgical data science - from concepts toward clinical translation. Med Image Anal 2022;76:102306. [Crossref] [PubMed]
33. Sadeghi AH, Mank Q, Tuzcu AS, et al. Artificial intelligence-assisted augmented reality robotic lung surgery: Navigating the future of thoracic surgery. JTCVS Tech 2024;26:121-5. [Crossref] [PubMed]
34. Drummond D, Gonsard A. Definitions and Characteristics of Patient Digital Twins Being Developed for Clinical Use: Scoping Review. J Med Internet Res 2024;26:e58504. [Crossref] [PubMed]
35. Ding H, Seenivasan L, Killeen BD, et al. Digital twins as a unifying framework for surgical data science: the enabling role of geometric scene understanding. Art Int Surg 2024;4:109-38.
36. Gonsard A, Genet M, Drummond D. Digital twins for chronic lung diseases. Eur Respir Rev 2024;33:240159. [Crossref] [PubMed]
37. Zhou X, Wang B, Wei Y, et al. Digital twins of ex vivo human lungs enable accurate and personalized evaluation of therapeutic efficacy. Nat Biotechnol 2026; [Crossref] [PubMed]
38. Lee A, Baker TS, Bederson JB, et al. Levels of autonomy in FDA-cleared surgical robots: a systematic review. NPJ Digit Med 2024;7:103. [Crossref] [PubMed]
39. Saeidi H, Opfermann JD, Kam M, et al. Autonomous robotic laparoscopic surgery for intestinal anastomosis. Sci Robot 2022;7:eabj2908. [Crossref] [PubMed]