Artificial intelligence (AI) applications and their impact on thoracic surgery: a narrative review

Introduction

Artificial intelligence (AI) is transforming numerous fields, including medicine, by applying advanced computational techniques to analyze complex datasets and support clinical decision-making. In thoracic surgery, AI has demonstrated significant potential in enhancing diagnostic accuracy, surgical precision, intraoperative guidance, and postoperative management (1,2).

AI encompasses a range of interrelated subfields: (I) machine learning (ML), which enables systems to learn from data and make predictions without explicit programming, is used in thoracic surgery to predict surgical outcomes, complications, and survival based on patient-specific clinical features; (II) deep learning (DL), a subset of ML using layered neural networks to handle complex tasks like image and speech recognition, is particularly effective in thoracic imaging tasks such as pulmonary nodule detection, tumor segmentation, and classification from computed tomography (CT) or positron emission tomography (PET) scans; (III) computer vision (CV), which allows machines to interpret and process visual information such as radiological scans, enhancing real-time navigation and robotic precision during procedures; and (IV) natural language processing (NLP), which enables the extraction and interpretation of information from clinical texts, such as electronic health records (2). Different AI models exist such as recurrent neural networks (RNNs), convolutional neural networks (CNNs) and decision trees. A comparative overview of key AI model types is provided in (Table S1).

AI encompasses a variety of subfields, including ML, DL, CV, and NLP, which are increasingly being integrated across the continuum of thoracic surgical care. Among these applications, robotic-assisted thoracic surgery (RATS) serves as a notable example where AI technologies enhance real-time intraoperative decision-making. Rather than functioning independently, AI tools within RATS augment the surgeon’s capabilities by providing visual overlays, anatomical guidance, and precision targeting through data-driven modeling. These AI systems are capable of processing large clinical and imaging datasets to identify patterns, support intraoperative navigation, and predict surgical risks. For instance, ML algorithms applied to imaging data can accurately detect pulmonary nodules, classify malignancies, and assess lymph node involvement. AI-driven radiomics has further improved the diagnostic process by enabling early and personalized identification of thoracic pathologies (1).

In the preoperative phase, AI can assist in risk assessment and surgical planning by predicting patient outcomes and identifying potential complications. This enables thoracic surgeons to tailor their approach to individual patients, optimizing surgical strategies and improving overall outcomes. RATS is another area where AI has made significant strides (2,3). RATS enhances surgical precision, reduces operative times, and improves recovery rates by providing surgeons with advanced tools and real-time feedback during procedures (1,2).

During surgery, AI-powered image-guided navigation and augmented reality (AR) systems offer real-time decision support, helping surgeons navigate complex anatomical structures and make informed decisions. These technologies can improve surgical safety and efficacy, reducing the risk of complications and enhancing patient outcomes. Postoperatively, AI-driven predictive models and wearable monitoring devices enable early detection of complications and facilitate continuous patient follow-up, further improving recovery and long-term outcomes (1,4,5).

Despite the promising applications of AI in thoracic surgery, several challenges remain. Data integration across imaging, genomic, and clinical systems remains limited, hindering model training and deployment. AI models often exhibit algorithmic bias due to training on homogenous datasets, reducing their generalizability across diverse populations. Moreover, the black-box nature of DL limits interpretability, raising concerns over transparency, trust, and clinical accountability. Ethical and legal issues such as unclear data ownership, liability in AI-guided decisions, and regulatory gaps further complicate real-world adoption (1,6). Different AI tools and platforms in medicine and thoracic surgery with their advantages and disadvantages are presented in Table S2.

Previous reviews have primarily provided broad overviews of AI applications in thoracic surgery, focusing on diagnostic accuracy, robotic-assisted surgery, intraoperative guidance, and postoperative management (1). While these works have summarized key developments, they have not offered a fully integrated, phase-based analysis linking each AI tool to its specific role in the surgical pathway, nor have they incorporated the most recent literature from 2024–2025. The present review expands on prior work by synthesizing updated evidence across all surgical phases, integrating recent validation studies, and aligning AI applications with clinical workflow to highlight both current capabilities and translational gaps.

To systematically explore the role of AI in thoracic surgery, this review adopts a conceptual framework organized across three core domains: diagnostic, predictive, and therapeutic applications. In the diagnostic domain, AI enhances early detection and characterization of thoracic diseases through imaging and pattern recognition, such as AI-powered radiomics and nodule classification (7). In the predictive domain, AI-based models estimate patient-specific risks, surgical outcomes, and disease trajectories, often integrating imaging, clinical, and genomic data (7). Finally, in the therapeutic domain, AI supports personalized surgical planning, robotic assistance, and treatment optimization—including perioperative planning and intraoperative guidance—reflecting a cohesive influence on clinical decision-making in thoracic surgery (7).

This document will provide a comprehensive overview of the current applications of AI in thoracic surgery, focusing on diagnostic enhancements, preoperative planning, intraoperative guidance, and postoperative management. This review is highlighting the transformative potential of AI technologies in improving patient outcomes and surgical precision. We present this article in accordance with the Narrative Review reporting checklist (available at https://ccts.amegroups.com/article/view/10.21037/ccts-25-21/rc).

Methods

This narrative review was conducted to synthesize the current state of AI applications in thoracic surgery. A structured literature search was performed using PubMed, Scopus, Web of Science, and Google Scholar up to April 10, 2025. The search strategy combined Medical Subject Headings (MeSH) and free-text terms: “artificial intelligence”, “machine learning”, “deep learning”, “radiomics”, and “thoracic surgery”.

Inclusion criteria comprised English-language peer-reviewed articles involving human subjects and published original research studies or narrative reviews focused on AI applications relevant to thoracic surgical practice, including diagnostics, preoperative planning, intraoperative guidance, postoperative monitoring, and surgical education.

Exclusion criteria included editorials, letters to the editor, conference abstracts, animal studies, non-English publications, and articles not directly addressing thoracic surgery or AI applications.

Given the narrative nature of this review, a formal risk of bias assessment [e.g., Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) or Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2)] was not performed. However, key studies were evaluated for clinical relevance, methodological transparency, and outcome reporting, and a structured summary of key studies involving implemented or validated AI models in thoracic surgery—including model types, application domains, outcome metrics, and surgical phases—is provided in Table S3.

Two reviewers (Mohamed Rahouma and H.M.) independently screened titles and abstracts, and full texts were reviewed when necessary. Disagreements were resolved through discussion. A summary of the search strategy is presented in Table 1.

Diagnostic applications of AI in thoracic surgery

AI is offering significant advancements in diagnostic accuracy, surgical precision, and patient management. This review explores the current applications of AI in thoracic surgery, supported by recent medical literature. This section presents a unified overview of how AI technologies are applied across each phase of thoracic surgery — diagnosis, preoperative planning, intraoperative guidance, postoperative care, and personalized oncology—highlighting specific tools, methods, and clinical outcomes. An overview of AI tools in thoracic surgery is presented in Figure 1, Figure S1 and Table S4.

Figure 1 Overview of artificial intelligence applications across the thoracic surgical pathway. AI, artificial intelligence; RATS, robotic-assisted thoracic surgery.

Diagnostic enhancements, radiomics, and pulmonary nodule detection and classification

AI has advanced thoracic diagnostics by pairing DL with radiomics to improve pulmonary nodule detection, malignancy classification, and nodal assessment. Beyond narrative overviews (1-3), several primary studies report clinically relevant performance. For example, a PET/CT cross-modal DL model for occult nodal metastasis in non-small cell lung cancer (NSCLC) achieved area under the curves (AUCs) of 0.87 (training) and 0.84 (validation), supporting preoperative risk stratification (8). AI-driven radiomics has improved the diagnostic process, allowing for earlier and more accurate detection of thoracic diseases (1,2).

A recent meta-analysis of CT-based radiomics for nodule malignancy reported pooled discrimination in the moderate-to-good range, underscoring promise but also heterogeneity in methods and validation cohorts (9). In addition to AI-powered radiomics, open-source frameworks such as MONAI (Medical Open Network for AI) have facilitated the development of DL models for thoracic imaging. MONAI supports segmentation, classification, and detection tasks, enabling more robust and reproducible diagnostic pipelines in thoracic surgery research and clinical applications (10).

AI applications in imaging and radiomics have shown remarkable efficacy in thoracic surgery. AI algorithms can process large volumes of imaging data, identifying patterns that may be missed by human eyes. This capability enhances the detection and classification of pulmonary nodules, aiding in the early diagnosis of lung cancer. AI-driven radiomics can extract quantitative features from medical images, providing valuable insights into tumor characteristics and aiding in personalized treatment planning (1-3).

AI has significantly improved the detection and classification of pulmonary nodules. ML algorithms can analyze CT scans to identify nodules with high sensitivity and specificity. These algorithms can also differentiate between benign and malignant nodules, reducing the need for invasive procedures and enabling early intervention. AI’s ability to analyze large datasets and learn from them enhances its diagnostic accuracy, leading to better patient outcomes (1,2).

Failure modes and current limitations

Despite encouraging AUCs, real-world deployment must account for false-positives (e.g., inflammatory or post-infectious nodules flagged as malignant) and false-negatives (e.g., subsolid or very small lesions missed), dataset and scanner domain shift, and class imbalance that can inflate apparent accuracy while depressing positive predictive value (PPV) in screening-prevalence settings. Radiomics pipelines remain sensitive to segmentation variability, feature harmonization, and leakage between training/validation splits; many studies are single-center and retrospective with limited calibration reporting (1-3,9,11). These gaps explain variability in reported sensitivity/specificity and the need for prospective, external validation before claims of reduced biopsies or earlier intervention can be generalized.

Diagnostic performance metrics of various AI models applied to thoracic imaging modalities are detailed in Table S5. For example, radiomics-based CT analysis achieved a sensitivity of 86% and specificity of 84% for pulmonary nodule detection (9), deep residual network CT models reached 91.07% and 88.64% for lung cancer classification (12), and PET/CT-based models predicted lymph node metastasis with 82% sensitivity and 86% specificity (11).

Predictive applications in thoracic surgery

Risk assessment and surgical planning

AI-driven risk assessment tools can analyze patient data to predict surgical risks and outcomes. These tools consider various factors, including patient demographics, comorbidities, and imaging data, to provide personalized risk assessments. This information helps surgeons make informed decisions about the best surgical approach, ultimately improving patient safety and surgical success rates (2,3).

Examples include the C2-Ai (Clinical AI) platform, which uses large-scale, globally sourced clinical datasets to generate personalized preoperative risk scores. By identifying patients at high risk for complications, C2-Ai supports targeted ‘prehabilitation’ and individualized surgical planning (13).

As AI becomes more integrated into thoracic surgical workflows, the need for transparency and interpretability in clinical decision-making is paramount. Explainable AI (XAI) techniques such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) help elucidate how complex models make predictions, enhancing trust and clinical adoption in high-stakes scenarios like cancer staging or resection planning. In thoracic surgery, where decisions are often life-altering, XAI supports safer, more accountable practices by revealing which variables drive AI outputs, allowing surgeons to validate or contest those decisions (14).

Models such as CXR-CTSurgery (Chest X ray-Computed tomography Surgery model), a DL model trained on 9,283 patients undergoing thoracic surgery for various conditions—including lung cancer and other thoracic neoplasms—used preoperative chest radiographs to predict postoperative mortality. The model achieved an AUC of 0.82, demonstrating predictive accuracy comparable to traditional risk tools such as the Society of Thoracic Surgeons Predicted Risk of Mortality (STS-PROM) score (1). Esteva et al. highlight the role of artificial neural networks (ANNs) in delivering personalized, data-driven risk evaluations, while cautioning that heterogeneous retrospective datasets may limit model Validity, but highlight AI’s potential as a decision-support tool in thoracic surgery (15). Zhong et al. developed a cross-modal DL model that integrates PET/CT imaging to non-invasively predict occult nodal metastasis in stage N0 NSCLC patients, achieving high predictive accuracy with AUCs of 0.87 and 0.84 in training and validation cohorts, respectively (8).

Fluorescence endoscopy represents a promising tool for early neoplasia detection in the esophagus. The integration of AI into fluorescence imaging has been shown to enhance lesion recognition and improve biopsy targeting by reducing observer variability. As highlighted in the review by Lazăr et al., AI-assisted analysis can augment the diagnostic accuracy of fluorescence-based techniques, particularly in the assessment of Barrett’s esophagus and early esophageal carcinoma, thus supporting clinical decision-making even among non-expert endoscopists (16).

Penso et al. developed a ML model using the XGBoost algorithm to predict mitral valve (MV) repair failure and recurrent mitral regurgitation in patients with MV prolapse, based on preoperative clinical and echocardiographic data. The model demonstrated strong predictive performance, with an AUC of 0.75 at 1 month and 0.92 at 3 years post-surgery. This study highlights the potential of ML to guide surgical planning by identifying high-risk patients, particularly those with complex MV anatomy, and supports its integration into clinical workflows to enhance long-term outcomes in valvular surgery (17).

Therapeutic applications in thoracic surgery

Preoperative planning

AI plays a crucial role in preoperative planning by assisting in risk assessment and surgical planning. Specifically, AI algorithms analyze large datasets from electronic health records, imaging, and clinical parameters to assist in risk stratification, outcome prediction, and surgical strategy optimization. For example, ML models have been developed to predict postoperative complications such as the following: prolonged air leaks, atrial fibrillation, and respiratory failure in lung resection patients. This enables surgeons to anticipate and mitigate these risks preemptively. One study demonstrated that AI-based risk assessment models predicted major complications with an accuracy exceeding 85%, outperforming traditional scoring systems (2).

In terms of outcome prediction, AI tools utilize radiomic features extracted from CT and PET scans to forecast tumor behavior and lymph node involvement. This directly impacts surgical planning, such as the extent of resection and lymphadenectomy required (3). DL models have also been applied to stratify patients by their likelihood of 30-day mortality or prolonged hospital stay, aiding in personalized care planning. AI optimizes surgical strategies by integrating multi-modal data to simulate operative scenarios, thereby assisting surgeons in selecting minimally invasive approaches versus open surgery when appropriate. For example, AR platforms enhanced by AI enable real-time three-dimensional (3D) visualization of anatomical structures, improving intraoperative navigation and reducing operative times (2).

Quantitatively, the implementation of AI-driven preoperative tools has been associated with a 10–15% reduction in postoperative complication rates and improved overall survival in select patient cohorts, reflecting meaningful clinical benefits.

Intraoperative guidance

AI technologies have significantly enhanced intraoperative guidance, providing real-time decision support and improving surgical precision. RATS and AI-powered image-guided navigation systems are key examples of how AI is transforming the surgical landscape (1,2).

The da Vinci Surgical System is a widely adopted robotic-assisted platform in thoracic surgery, integrating high-definition 3D visualization with AI-enabled motion scaling and tremor filtration to enhance surgical precision (18).

RATS

RATS has significantly enhanced surgical precision and enabled minimally invasive approaches that may improve operative outcomes. AI algorithms integrated into robotic systems provide real-time feedback and assist surgeons in navigating complex anatomical structures. This technology allows for minimally invasive procedures, reducing patient morbidity and improving postoperative outcomes (1,2).

In a large single-center series of 504 pulmonary resections using the da Vinci Xi system, complication rates decreased from 35–39% to 15% (P<0.001), with reduced hospital stay and readmissions (19).

AI is increasingly influencing RATS by supporting critical phases of the surgical workflow: preoperative planning, intraoperative navigation, and the conceptual groundwork for future postoperative analytics (6,20).

Preoperative planning

PulmoVR is a virtual reality platform integrated with AI-powered segmentation that enables the reconstruction of detailed 3D models of the lung anatomy, including bronchi, pulmonary vessels, and tumor margins (6). In a prospective pilot study of ten patients, PulmoVR-based planning led to changes in the planned surgical approach in 40% of cases (6). These models are reviewed by the surgical team prior to robotic segmentectomy and used to formulate or refine resection strategies. In a pilot study involving five patients undergoing segmentectomy using the da Vinci Xi robotic system, PulmoVR was used to segment patient-specific anatomy and simulate the procedure preoperatively. This immersive 3D visualization supported surgical planning and provided enhanced anatomical understanding (6).

Intraoperative guidance and surgeon-AI interaction

During the procedure, the PulmoVR-generated 3D model was displayed on the da Vinci Xi console using the TilePro interface. The surgical team performed manual alignment of the model to match the intraoperative anatomy, accounting for lung deflation and deformation. This process enabled AR guidance during hilar dissection and identification of the intersegmental plane. Updates to the AR model were feasible within 15 seconds, and no complications or workflow interruptions occurred. Surgeons reported enhanced spatial orientation and increased confidence during critical phases of the surgery (6).

In a separate validation study involving eight thoracic procedures (five RATS), a stereo-endoscopic AR overlay method achieved a median registration error of 0.20 mm and reprojection error of 0.22 pixels with a setup time of ~5 minutes (20).

Comparative platform integration

Peek et al. introduced a stereo-endoscopic AR overlay method using point cloud registration that works on both video-assisted thoracoscopic surgery (VATS) and RATS platforms. In their study involving eight thoracic procedures (five RATS), this AI-assisted registration system achieved a median registration error of 0.20 mm, a reprojection error of 0.22 pixels, and a setup time of approximately 5 minutes. It enabled accurate identification of segmental arterial branches with high visual fidelity (20).

Similarly, the HoloSurgical ARAI™ (Augmented reality and artificial intelligence-assisted) system offers AR-assisted navigation, providing real-time, high-fidelity 3D anatomical overlays that facilitate complex dissections and minimize intraoperative errors (21).

AI-powered image-guided navigation and AR

AI-powered image-guided navigation systems and AR technologies offer real-time decision support during surgery. These systems use AI algorithms to process imaging data and provide surgeons with detailed, real-time visualizations of the surgical field. This enhances surgical accuracy, reduces the risk of complications, and improves overall surgical outcomes (1,4,5).

Peek et al.’s AI-enhanced AR system for minimally invasive thoracoscopic pulmonary segmentectomy uses intraoperative stereoscopic images to generate 3D point clouds aligned with preoperative CT models, achieving sub-millimeter accuracy (20). This platform builds on their earlier stereo-endoscopic overlay approach, confirming its reproducibility and adaptability across platforms.

Sadeghi et al. demonstrated the feasibility of real-time AI-assisted AR during RATS using PulmoVR, an AI-driven system that generates dynamic 3D lung models from CT scans. These deformable models were manually aligned intraoperatively to reflect anatomical changes during surgery, enabling surgeons to visualize patient-specific anatomy with depth and accuracy. This first-in-human study showed promising results in enhancing anatomical orientation and safety during lung resections (6).

Postoperative management

AI-driven predictive models and wearable monitoring devices play a crucial role in postoperative management. These technologies enable early detection of complications, continuous patient monitoring, and personalized follow-up care, ultimately improving patient outcomes and reducing hospital readmissions (1,4,5).

Predictive models and wearable monitoring devices

AI-driven predictive models can analyze postoperative data to identify patients at risk of complications. These models consider various factors, including patient demographics, surgical details, and postoperative recovery data, to provide personalized risk assessments. Wearable monitoring devices equipped with AI algorithms can continuously monitor patients’ vital signs and detect early signs of complications, allowing for timely interventions and improved patient outcomes (1,4).

The Advance Alert Monitor (AAM) exemplifies such tools, continuously analyzing postoperative vital signs and clinical parameters to detect early signs of deterioration, thereby enabling prompt interventions and reducing readmissions (22).

A phase-wise breakdown of AI applications across the thoracic surgical pathway is summarized in Table 2.

AI in personalized thoracic oncology

AI plays an increasingly important role in personalized thoracic oncology by integrating genomic data, imaging, and clinical risk factors to support tailored treatment decisions. AI algorithms are capable of identifying molecular subtypes, predicting therapeutic responses, and stratifying patients based on individualized risk profiles. These advances facilitate precision medicine strategies that improve treatment efficacy and reduce unnecessary interventions (23).

Zhong et al. developed a cross-modal DL model that integrates PET/CT imaging to non-invasively predict occult nodal metastasis in stage N0 NSCLC patients. The model demonstrated strong predictive performance, with an AUC of 0.87 in the training cohort and 0.84 in the external validation cohort, highlighting the potential of multimodal AI tools in treatment planning and risk stratification in early-stage lung cancer patients (8).

Beyond imaging, AI supports the convergence of radiomics and genomics to improve tumor characterization and therapeutic guidance. Zhu et al. summarized the application of AI across the lung cancer pathway, emphasizing its role in integrating radiogenomics for tumor subtyping, predicting treatment response, and personalizing care strategies in clinical settings (24).

Furthermore, Huang et al. reviewed how AI algorithms contribute to genomic biomarker identification and facilitate individualized immunotherapy planning. Their analysis highlights how AI models trained on both genomic and imaging data are now capable of predicting immune checkpoint inhibitor response, thereby assisting oncologists in selecting optimal therapies for NSCLC patients (25).

Radiogenomics, the fusion of radiological and genomic data using AI algorithms, is another frontier. As described by Li and Zhou, AI-enhanced radiogenomics improves the prognostic stratification of lung tumors by extracting quantitative imaging features and associating them with specific genetic mutations or expression profiles. These methods help identify tumor aggressiveness and predict recurrence, paving the way for precision interventions (26).

Precision oncology: beyond surgical techniques

AI increasingly supports multimodal, personalized lung cancer management—not limited to surgical procedures—but also in immunotherapy response prediction and genomics-guided therapy.

Predicting immunotherapy efficacy

A real-world prospective study of 480 advanced NSCLC patients used explainable AI [CatBoost, logistic regression, neural networks, support vector machines (SVMs)] integrated with SHAP values. The model achieved 83% accuracy for 6-month overall survival and highlighted key predictors [e.g., neutrophil-to-lymphocyte ratio, programmed death-ligand 1 (PD-L1) expression] (23).

Multimodal decision platforms (I3LUNG)

The EU-funded I3LUNG platform integrates clinical, radiologic, molecular, and patient-reported data across ~2,000 NSCLC cases to develop AI/ML tools that predict immunotherapy response and support treatment decisions in real-world settings (27).

Radiogenomic applications

Zhu et al. in 2025 emphasizes CNNs and transformers enabling multimodal data integration—combining imaging with genomic markers—to refine personalized screening and therapeutic strategies based on tumor biology (24).

Potential impact on training and education

AI holds transformative potential in surgical training and education. AI-powered simulation platforms, such as Touch Surgery, provide interactive, anatomically accurate modules for various thoracic procedures. These simulations are validated by clinical studies and offer an immersive, risk-free environment for learning operative steps and decision-making (28). AI-assisted skill assessment tools, like SurgTrainer and motion-analysis platforms, utilize video and sensor data to evaluate technical performance based on objective metrics (e.g., economy of motion, precision, and time efficiency). Such systems can benchmark surgical performance against expert standards, enabling data-driven training feedback (29).

Simulation platforms such as Touch Surgery, eoSim, and other simulation tools have demonstrated educational benefits in thoracic surgical training as seen in Table 3.

Surgical simulation and training are increasingly being augmented by AI to enable objective, data-driven evaluation and proficiency tracking. Recent virtual reality simulators for robot-assisted thoracic surgery now report metrics such as time-to-completion, economy of motion, and workspace efficiency—allowing comparisons of surgical performance with minimal and controlled movements (33). Simulation systems can also generate objective performance indicators (OPIs), derived from kinematic data (e.g., instrument path length, velocity, active time), which enable standardized performance assessment across trainees and experts (34). More advanced frameworks like SATRDL (Surgical Skill Assessment and Task Recognition with Deep Learning) employ DL to analyze raw motion sequences and assess technical skill and task recognition accurately (with reported accuracy up to 96%) (35). Adaptive training systems utilizing haptic feedback have also shown statistically significant improvements in task time, path straightness, and targeting error when compared to control groups (P<0.05) (36). These systems often incorporate validated evaluation tools such as OSATS (Objective Structured Assessment of Technical Skills) or GEARS (Global Evaluative Assessment of Robotic Skills), alongside automated analytics, to form a robust evaluation framework for simulation efficacy (34).

Technological innovations in AI

ML and DL techniques

ML and DL techniques are at the forefront of AI innovations in thoracic surgery. ML algorithms can analyze vast amounts of data to identify patterns and make predictions that improve patient care. In thoracic surgery, ML has been used to enhance diagnostic processes, optimize preoperative planning, and predict postoperative outcomes (1,2).

DL, a subset of ML, involves neural networks with multiple layers that can learn from large datasets as seen in Figure 2. DL techniques have shown remarkable efficacy in analyzing medical images, such as CT scans, to detect and classify pulmonary nodules. These techniques can differentiate between benign and malignant nodules with high accuracy, reducing the need for invasive procedures and enabling early intervention (1,2).

Figure 2 AI-powered machine learning and deep learning in thoracic surgery. AI, artificial intelligence; CT, computed tomography.

The integration of ML and DL in thoracic surgery has led to significant advancements in personalized medicine. AI algorithms can analyze patient data, including imaging, demographics, and clinical history, to provide tailored risk assessments and surgical plans. This personalized approach improves surgical outcomes and reduces the risk of complications (2,37,38).

CV and multi-modal models

CV is another critical AI technology in thoracic surgery. It involves the use of algorithms to interpret and process visual data from medical images. CV techniques can enhance the accuracy of image-guided navigation systems and AR applications during surgery (39).

Multi-modal models combine data from various sources, such as imaging, clinical records, and genetic information, to provide comprehensive insights into patient health. These models can improve diagnostic accuracy, optimize treatment plans, and predict patient outcomes. In thoracic surgery, multi-modal models have been used to enhance the detection and classification of lung cancer, predict lymph node metastasis, and assess surgical risks (39).

The integration of CV and multi-modal models in thoracic surgery has led to significant improvements in intraoperative guidance. AI-powered image-guided navigation systems use CV algorithms to provide real-time visualizations of the surgical field, helping surgeons navigate complex anatomical structures and make informed decisions. AR applications further enhance surgical precision by overlaying digital information onto the physical environment, providing surgeons with detailed, real-time visualizations (1,5).

Clinical outcomes and benefits

Improved diagnostic accuracy and surgical precision

AI has significantly enhanced diagnostic accuracy and surgical precision in thoracic surgery. ML algorithms and DL techniques analyze vast amounts of imaging data to identify patterns and make predictions that improve patient care. AI-driven radiomics can extract quantitative features from medical images, providing valuable insights into tumor characteristics and aiding in personalized treatment planning (1,2).

AI applications in imaging have shown remarkable efficacy in detecting and classifying pulmonary nodules. ML algorithms can analyze CT scans to identify nodules with high sensitivity and specificity. These algorithms can differentiate between benign and malignant nodules, reducing the need for invasive procedures and enabling early intervention. AI’s ability to analyze large datasets and learn from them enhances its diagnostic accuracy, leading to better patient outcomes (1,2).

Intraoperatively, AI-powered image-guided navigation systems and AR technologies offer real-time decision support, helping surgeons navigate complex anatomical structures and make informed decisions. These systems use AI algorithms to process imaging data and provide surgeons with detailed, real-time visualizations of the surgical field. This enhances surgical accuracy, reduces the risk of complications, and improves overall surgical outcomes (1).

Enhanced recovery rates and reduced operative times

AI technologies have led to enhanced recovery rates and reduced operative times in thoracic surgery. RATS is a key example of how AI is transforming the surgical landscape. RATS enhances surgical precision, reduces operative times, and improves recovery rates by providing surgeons with advanced tools and real-time feedback during procedures (1,2).

AI-driven risk assessment tools can analyze patient data to predict surgical risks and outcomes. These tools consider various factors, including patient demographics, comorbidities, and imaging data, to provide personalized risk assessments (2,38).

The integration of AI in preoperative planning has led to optimized surgical strategies and improved overall outcomes. AI algorithms can predict patient outcomes, identify potential complications, and tailor surgical plans to individual patients. This personalized approach reduces the risk of adverse events and enhances recovery rates (2,38).

Early detection of complications

AI-driven predictive models and wearable monitoring devices play a crucial role in the early detection of complications. These technologies enable continuous patient monitoring, personalized follow-up care, and timely interventions, ultimately improving patient outcomes and reducing hospital readmissions (1,5). A summary of these clinical benefits is visually presented in Figure S2.

AI-driven predictive models can analyze postoperative data to identify patients at risk of complications. These models consider various factors, including patient demographics, surgical details, and postoperative recovery data, to provide personalized risk assessments. Wearable monitoring devices equipped with AI algorithms can continuously monitor patients’ vital signs and detect early signs of complications, allowing for timely interventions and improved patient outcomes (1,5).

AI applications in postoperative management have shown significant potential in improving patient follow-up and reducing hospital re-admissions. Continuous monitoring and early detection of complications enable timely interventions, reducing the risk of adverse events and improving long-term patient outcomes (1,2,5).

Practical AI tools currently used in thoracic surgery are summarized in Table S4, highlighting their clinical applications and operational considerations.

Challenges and limitations

Data integration and algorithmic biases

One of the primary challenges in implementing AI in thoracic surgery is data integration. AI algorithms require large, high-quality datasets to train and validate models effectively. However, the integration of diverse data sources, such as imaging, clinical records, and genetic information, poses significant technical and logistical challenges. Ensuring data compatibility, standardization, and interoperability across different systems is crucial for the successful deployment of AI technologies (1,5). A structured summary of these challenges and their proposed solutions is provided in Table 4.

Algorithmic biases are another critical concern. AI models are trained on existing data, which may contain inherent biases. These biases can lead to skewed predictions and reinforce disparities in healthcare. For instance, if the training data predominantly represents a specific demographic, the AI model may not perform well for other populations. Addressing algorithmic biases requires careful consideration of data diversity and the development of techniques to mitigate bias during model training and validation (1,5).

Ethical and regulatory concerns

The integration of AI into thoracic surgery raises several ethical and regulatory concerns. Data privacy and security are paramount, as AI systems often require access to sensitive patient information. Ensuring robust data protection measures and compliance with regulations, such as the Health Insurance Portability and Accountability Act (HIPAA), is essential to maintain patient trust and safeguard their privacy (1,2).

Patient autonomy must also be respected by ensuring informed consent explicitly addresses the use of AI tools in diagnosis and surgical planning (40).

Ethical concerns also extend to the transparency and accountability of AI systems. The “black box” nature of some AI models can make it challenging to understand how decisions are made, potentially undermining clinical accountability. Developing XAI techniques that provide insights into the decision-making process is crucial for maintaining transparency and trust in AI-driven clinical decisions (3,41).

Moreover, AI algorithms may reflect or amplify existing biases in healthcare data, potentially exacerbating disparities in care across racial, gender, or socioeconomic groups. Addressing algorithmic bias is essential to ensure equitable outcomes for all patient populations (42).

Regulatory frameworks must evolve to address the unique challenges posed by AI in healthcare. Establishing guidelines for the development, validation, and deployment of AI technologies is essential to ensure their safe and effective use. Collaboration between regulatory bodies, healthcare providers, and AI developers is necessary to create comprehensive and adaptive regulations that keep pace with technological advancements (3,41).

Additionally, legal accountability remains unclear in AI-assisted care: when errors occur, it is not always obvious whether liability lies with the clinician, developer, or institution. Clarifying this responsibility is critical to fostering ethical deployment of AI in clinical practice (40).

High implementation costs

The implementation of AI technologies in thoracic surgery involves significant financial investments. Developing, validating, and deploying AI systems require substantial resources, including high-performance computing infrastructure, specialized personnel, and ongoing maintenance. These costs can be prohibitive, particularly for smaller healthcare institutions with limited budgets (1,38).

Additionally, the integration of AI into existing clinical workflows necessitates training and education for healthcare providers. Ensuring that surgeons and other medical staff are proficient in using AI tools and interpreting their outputs is crucial for maximizing the benefits of AI technologies. This training process can be time-consuming and costly, further adding to the overall implementation expenses (38,43).

Despite these challenges, the potential benefits of AI in thoracic surgery, such as improved diagnostic accuracy, surgical precision, and patient outcomes, make it a worthwhile investment. Addressing the financial barriers requires strategic planning, resource allocation, and collaboration between stakeholders to ensure the sustainable integration of AI into clinical practice (1,38,43).

Reproducibility, evaluation metrics, and model transparency

Reproducibility and generalizability

A significant limitation in current AI literature is the lack of reproducibility and external validation. Many models are trained and tested on retrospective, single-center datasets, which limits their applicability to diverse clinical settings. This reduces confidence in their generalizability and poses challenges for real-world integration (44).

Overfitting and data volume limitations

AI models developed on small or homogenous datasets are at risk of overfitting—performing well on internal validation but poorly on unseen data. This issue remains a critical barrier to clinical deployment, emphasizing the need for large, multi-institutional, and prospectively collected datasets (16).

Evaluation metric inconsistencies

There is a lack of standardized performance metrics across AI studies in thoracic surgery. Studies variably report AUC, accuracy, precision-recall, or F1 scores, making comparisons between models difficult. Adoption of consistent and clinically relevant benchmarks is necessary to evaluate model effectiveness fairly (1).

XAI

Although AI models can make highly accurate predictions, their black-box nature hinders clinical acceptance. In high-stakes situations like cancer staging or resection planning, XAI techniques like SHAP and LIME improve clinical adoption and trust by clarifying how complex models make predictions. For example, SHAP has been used to show how certain characteristics—like tumor size, lymph node status, and PET uptake values—help predict nodal metastases in NSCLC. Likewise, LIME has been utilized to decipher AI-generated risk scores for postoperative complications, allowing physicians to comprehend which clinical factors had the greatest impact on the model’s output and instantly verify its suggestions (45).

Challenges to clinical translation

While AI holds transformative promise for thoracic surgery, several real-world challenges hinder its seamless clinical integration. These barriers span infrastructural, educational, ethical, and legal domains, and must be addressed to ensure effective and equitable implementation.

Infrastructure and cost barriers

The adoption of AI technologies in thoracic surgery requires substantial infrastructural investment. Robotic systems such as the da Vinci platform can cost between $1 million and $2.5 million, with additional maintenance and per-procedure costs exceeding $1,500 per case. In addition, AI-enabled image-guidance, 3D reconstruction software, and simulation platforms impose significant financial burdens on institutions. These expenses are particularly challenging for low-resource settings, potentially widening disparities in access to advanced care (46).

Workflow disruption and surgeon training

Integrating AI into existing surgical workflows may initially increase procedural complexity. The learning curve for RATS is steep, with estimates suggesting 150–250 cases are required to achieve proficiency. The addition of AI-guided navigation, AR, and real-time analytics may further increase intraoperative cognitive load. Moreover, survey data suggest that institutional support for surgeon training in AI technologies is inconsistent, leading to hesitancy or resistance among practitioners (47).

Ethical, legal, and regulatory challenges

AI introduces complex questions surrounding data ownership, informed consent, and accountability. The use of patient data for model training raises concerns about privacy and HIPAA compliance. The lack of transparency in DL systems—often referred to as “black-box” models—complicates the clinician’s ability to justify decisions derived from AI outputs. In the event of an error, legal liability remains ambiguous: responsibility may lie with the clinician, software developer, or institution. These uncertainties underscore the need for robust regulatory frameworks tailored to AI-enabled clinical tools (48).

Future directions

Digital twin technology and federated learning

Digital twin technology and federated learning represent promising future directions for AI in thoracic surgery. Digital twins are virtual replicas of physical entities, such as patients, that can simulate various clinical scenarios and predict outcomes. This technology allows for personalized treatment planning and real-time monitoring, enhancing patient care. Federated learning, on the other hand, enables the training of AI models across multiple decentralized datasets while preserving data privacy. This collaborative approach can improve the robustness and generalizability of AI models, addressing the limitations of data scarcity and enhancing predictive accuracy (1).

XAI for improved interpretability

XAI is crucial for improving the interpretability and transparency of AI models in thoracic surgery. XAI techniques aim to make AI decision-making processes more understandable to clinicians, thereby increasing trust and facilitating clinical adoption. By providing insights into how AI models arrive at their predictions, XAI can help identify and mitigate biases, ensure accountability, and enhance the overall reliability of AI-driven clinical decisions. This is particularly important in high-stakes surgical environments where understanding the rationale behind AI recommendations is essential for patient safety (1).

Multicenter validation and standardized AI frameworks

Multicenter validation and standardized AI frameworks are essential for the widespread adoption of AI in thoracic surgery. Multicenter studies can provide robust evidence of AI model performance across diverse patient populations and clinical settings, ensuring generalizability and reliability. Standardized frameworks for AI development, validation, and deployment can streamline the integration of AI technologies into clinical practice, ensuring consistency and compliance with regulatory requirements. These frameworks can also facilitate collaboration between researchers, clinicians, and regulatory bodies, promoting the responsible and ethical use of AI in healthcare (1,3).

Strengths and limitations of this review

This narrative review offers a comprehensive and up-to-date synthesis of AI applications in thoracic surgery across the full clinical workflow—from diagnosis to surgical training—while integrating information on underlying technologies, clinical outcomes, and interpretability tools. A key strength of this paper lies in its structured, workflow-based framework, which enables readers to clearly understand how specific AI technologies are deployed in real-world thoracic surgical practice. Additionally, the inclusion of recent, validated studies with performance metrics and tools enhances the review’s clinical relevance. However, this review has several limitations. First, as a narrative review, it lacks a formal meta-analytic approach and may be subject to selection bias despite the structured search. Second, many of the included studies are retrospective, single-center, or lack external validation, which limits the generalizability of their findings. Furthermore, some reported AI models rely on homogenous datasets, raising concerns about algorithmic bias and real-world applicability. Lastly, rapid technological evolution may render some findings time sensitive. Despite these limitations, the review provides a valuable foundation for future research and clinical translation in this emerging field.

Conclusions

AI has demonstrated significant potential to advance thoracic surgery by enhancing diagnostic accuracy, surgical precision, intraoperative guidance, and postoperative management. AI-driven technologies—such as ML, DL, CV, and robotic-assisted surgery—have shown remarkable efficacy in improving patient outcomes and optimizing clinical workflows. Despite challenges related to data integration, algorithmic biases, and ethical concerns, AI continues to evolve, offering new opportunities for personalized and precision medicine in thoracic surgery. To fully realize this potential, further research and ethical AI governance are imperative. Continued advancements in digital twin technology, federated learning, and explainable AI can improve the interpretability, reliability, and accessibility of AI models. Multicenter validation studies and standardized AI frameworks are essential for ensuring safe and effective implementation. Moreover, addressing ethical concerns related to data privacy, transparency, and accountability is crucial for maintaining patient trust and ensuring equitable access to AI-driven healthcare. With responsible integration and ongoing research, AI is poised to play a pivotal role in shaping the future of thoracic surgery and improving patient care and outcomes.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://ccts.amegroups.com/article/view/10.21037/ccts-25-21/rc

Peer Review File: Available at https://ccts.amegroups.com/article/view/10.21037/ccts-25-21/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://ccts.amegroups.com/article/view/10.21037/ccts-25-21/coif). Mohamed Rahouma serves as an unpaid editorial board member of Current Challenges in Thoracic Surgery from March 2025 to February 2027. 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.

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. Leivaditis V, Maniatopoulos AA, Lausberg H, et al. Artificial Intelligence in Thoracic Surgery: A Review Bridging Innovation and Clinical Practice for the Next Generation of Surgical Care. J Clin Med 2025;14:2729. [Crossref] [PubMed]
2. Bellini V, Valente M, Del Rio P, et al. Artificial intelligence in thoracic surgery: a narrative review. J Thorac Dis 2021;13:6963-75. [Crossref] [PubMed]
3. Seastedt KP, Moukheiber D, Mahindre SA, et al. A scoping review of artificial intelligence applications in thoracic surgery. Eur J Cardiothorac Surg 2022;61:239-48. [Crossref] [PubMed]
4. Abbaker N, Minervini F, Guttadauro A, et al. The future of artificial intelligence in thoracic surgery for non-small cell lung cancer treatment a narrative review. Front Oncol 2024;14:1347464. [Crossref] [PubMed]
5. Cusumano G, D'Arrigo S, Terminella A, et al. Artificial Intelligence Applications for Thoracic Surgeons: "The Phenomenal Cosmic Powers of the Magic Lamp". J Clin Med 2024;13:3750. [Crossref] [PubMed]
6. 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]
7. Aleem MU, Khan JA, Younes A, et al. Enhancing Thoracic Surgery with AI: A Review of Current Practices and Emerging Trends. Curr Oncol 2024;31:6232-44. [Crossref] [PubMed]
8. Zhong Y, Cai C, Chen T, et al. PET/CT based cross-modal deep learning signature to predict occult nodal metastasis in lung cancer. Nat Commun 2023;14:7513. [Crossref] [PubMed]
9. Shi L, Sheng M, Wei Z, et al. CT-Based Radiomics Predicts the Malignancy of Pulmonary Nodules: A Systematic Review and Meta-Analysis. Acad Radiol 2023;30:3064-75. [Crossref] [PubMed]
10. Cardoso MJ, Li W, Brown R, et al. MONAI: An open-source framework for deep learning in healthcare [Internet]. arXiv; 2022 [cited 2025 Aug 9]. Available online: https://arxiv.org/abs/2211.02701
11. Li Y, Deng J, Ma X, et al. Diagnostic accuracy of CT and PET/CT radiomics in predicting lymph node metastasis in non-small cell lung cancer. Eur Radiol 2025;35:1966-79. [Crossref] [PubMed]
12. Nibali A, He Z, Wollersheim D. Pulmonary nodule classification with deep residual networks. Int J Comput Assist Radiol Surg 2017;12:1799-808. [Crossref] [PubMed]
13. Sharma V, Pritchard-Jones R, Scott S, et al. Accuracy of a tool to prioritise patients awaiting elective surgery: an implementation report. BMJ Health Care Inform 2023;30:e100687. [Crossref] [PubMed]
14. Tjoa E, Guan C. A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI. IEEE Trans Neural Netw Learn Syst 2021;32:4793-813. [Crossref] [PubMed]
15. Esteva H, Núñez TG, Rodríguez RO. Neural networks and artificial intelligence in thoracic surgery. Thorac Surg Clin 2007;17:359-67. [Crossref] [PubMed]
16. Lazăr DC, Avram MF, Faur AC, et al. The Impact of Artificial Intelligence in the Endoscopic Assessment of Premalignant and Malignant Esophageal Lesions: Present and Future. Medicina 2020;56:364. [Crossref] [PubMed]
17. Penso M, Pepi M, Mantegazza V, et al. Machine Learning Prediction Models for Mitral Valve Repairability and Mitral Regurgitation Recurrence in Patients Undergoing Surgical Mitral Valve Repair. Bioengineering (Basel) 2021;8:117. [Crossref] [PubMed]
18. Kent MS, Hartwig MG, Vallières E, et al. Pulmonary Open, Robotic, and Thoracoscopic Lobectomy (PORTaL) Study: An Analysis of 5721 Cases. Ann Surg 2023;277:528-33. [Crossref] [PubMed]
19. Soliman BG, Nguyen DT, Chan EY, et al. Impact of da Vinci Xi robot in pulmonary resection. J Thorac Dis 2020;12:3561-72. [Crossref] [PubMed]
20. 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]
21. Siemionow KB, Katchko KM, Lewicki P, et al. Augmented reality and artificial intelligence-assisted surgical navigation: Technique and cadaveric feasibility study. J Craniovertebr Junction Spine 2020;11:81-5. [Crossref] [PubMed]
22. Martinez VA, Betts RK, Scruth EA, et al. The Kaiser Permanente Northern California Advance Alert Monitor Program: An Automated Early Warning System for Adults at Risk for In-Hospital Clinical Deterioration. Jt Comm J Qual Patient Saf 2022;48:370-5. [Crossref] [PubMed]
23. Prelaj A, Galli EG, Miskovic V, et al. Real-world data to build explainable trustworthy artificial intelligence models for prediction of immunotherapy efficacy in NSCLC patients. Front Oncol 2022;12:1078822. [Crossref] [PubMed]
24. Zhu E, Muneer A, Zhang J, et al. Progress and challenges of artificial intelligence in lung cancer clinical translation. NPJ Precis Oncol 2025;9:210. [Crossref] [PubMed]
25. Huang D, Li Z, Jiang T, et al. Artificial intelligence in lung cancer: current applications, future perspectives, and challenges. Front Oncol 2024;14:1486310. [Crossref] [PubMed]
26. Li S, Zhou B. A review of radiomics and genomics applications in cancers: the way towards precision medicine. Radiat Oncol 2022;17:217. [Crossref] [PubMed]
27. Prelaj A. I3LUNG: Integrative Science, Intelligent Data Platform for Individualized LUNG Cancer Care With Immunotherapy [Internet]. clinicaltrials.gov; 2023 Jan [cited 2025 July 11]. Report No.: NCT05537922. Available online: https://clinicaltrials.gov/study/NCT05537922
28. Varas J, Coronel BV, Villagrán I, et al. Innovations in surgical training: exploring the role of artificial intelligence and large language models (LLM). Rev Col Bras Cir 2023;50:e20233605. [Crossref] [PubMed]
29. Jun SK, Narayanan MS, Agarwal P, et al. Robotic Minimally Invasive Surgical skill assessment based on automated video-analysis motion studies. In: 2012 4th IEEE RAS & EMBS International Conference on Biomedical Robotics and Biomechatronics (BioRob) 2012:25-31. Available online: https://ieeexplore.ieee.org/document/6290869
30. Burkhart HM, Riley JB, Hendrickson SE, et al. The successful application of simulation-based training in thoracic surgery residency. J Thorac Cardiovasc Surg 2010;139:707-12. [Crossref] [PubMed]
31. Grossi S, Cattoni M, Filipponi L, et al. ResearchGate [Internet]. [cited 2025 May 16]. Training simulator efficacy in developing thoracic and general surgical skills in a residency program: a pilot study. Available online: https://www.researchgate.net/publication/378101844_Training_simulator_efficacy_in_developing_thoracic_and_general_surgical_skills_in_a_residency_program_a_pilot_study
32. Koliakos E, Abdelnour E, Hasenauer A, et al. Objective improvement in dexterity for trainees undergoing a video-assisted thoracoscopic surgery simulation program, a prospective single center study. J Thorac Dis 2023;15:6674-86. [Crossref] [PubMed]
33. Spinelli A, La Raja C. Simulation in robotically assisted surgery and the surgical training landscape. Br J Surg. 2024;111:znae208.
34. Choksi S, Ballo M, Profant C, et al. Standardizing surgical training with objective performance indicators: a prospective cohort study. Surg Endosc 2025;39:2316-23. [Crossref] [PubMed]
35. Wang Z, Fey AM. SATR-DL: Improving Surgical Skill Assessment and Task Recognition in Robot-assisted Surgery with Deep Neural Networks. 2018 [cited 2025 Aug 10]; Available online: https://arxiv.org/abs/1806.05798
36. Ershad M, Rege R, Majewicz Fey A. Adaptive Surgical Robotic Training Using Real-Time Stylistic Behavior Feedback Through Haptic Cues. IEEE Trans Med Robot Bionics 2021;3:959-69. [Crossref] [PubMed]
37. Baudo M, Shmushkevich S, Rahouma M. Commentary: Machine Learning Prediction of Acute Kidney Injury With Cardiac Surgery. Seminars in Thoracic and Cardiovascular Surgery 2021;746-7. [Crossref] [PubMed]
38. Etienne H, Hamdi S, Le Roux M, et al. Artificial intelligence in thoracic surgery: past, present, perspective and limits. Eur Respir Rev 2020;29:200010. [Crossref] [PubMed]
39. Abràmoff MD, Tarver ME, Loyo-Berrios N, et al. Considerations for addressing bias in artificial intelligence for health equity. NPJ Digit Med 2023;6:170. [Crossref] [PubMed]
40. Gerke S, Minssen T, Cohen IG. Ethical and Legal Challenges of Artificial Intelligence-Driven Healthcare. Rochester, NY: Social Science Research Network; 2020. Available online: https://papers.ssrn.com/abstract=3570129
41. Arjomandi Rad A, Vardanyan R, Athanasiou T, et al. The ethical considerations of integrating artificial intelligence into surgery: a review. Interdiscip Cardiovasc Thorac Surg 2025;40:ivae192. [Crossref] [PubMed]
42. Chen Z, Zhang Y, Yan Z, et al. Artificial intelligence assisted display in thoracic surgery: development and possibilities. J Thorac Dis 2021;13:6994-7005. [Crossref] [PubMed]
43. Morris MX, Song EY, Rajesh A, et al. Ethical, Legal, and Financial Considerations of Artificial Intelligence in Surgery. Am Surg 2023;89:55-60. [Crossref] [PubMed]
44. Song Z, Izhar A, Wei B. Artificial intelligence-assisted three-dimensional reconstruction in thoracic surgery: a narrative review. Curr Chall Thorac Surg 2025;7:6.
45. Ladbury C, Zarinshenas R, Semwal H, et al. Utilization of model-agnostic explainable artificial intelligence frameworks in oncology: a narrative review. Transl Cancer Res 2022;11:3853-68. [Crossref] [PubMed]
46. Novellis P, Alloisio M, Vanni E, et al. Robotic lung cancer surgery: review of experience and costs. J Vis Surg 2017;3:39. [Crossref] [PubMed]
47. Barbash GI, Glied SA. New technology and health care costs--the case of robot-assisted surgery. N Engl J Med 2010;363:701-4. [Crossref] [PubMed]
48. Davenport T, Kalakota R. The potential for artificial intelligence in healthcare. Future Healthc J 2019;6:94-8. [Crossref] [PubMed]