Artificial Intelligence in Oncologic Thoracic Surgery: Clinical Decision Support and Emerging Applications

May combine two claims.

Teramoto et al. evaluated 298 images using DCNNs and achieved classification accuracies of 89% for adenocarcinoma, 60% for squamous cell carcinoma, and 70% for small-cell lung cancer, outperforming both cytotechnologists and pathologists.

“Teramoto et al. evaluated 298 images using DCNNs and achieved classification accuracies of 89% for adenocarcinoma, 60% for squamous cell carcinoma, and 70% for small-cell lung cancer—outperforming both cytotechnologists and pathologists [31].”

• cites [31] Teramoto, 2017, Automated classification of lung cancer types from cytological images using deep convolutional neural networks, BioMed Res. Int. sourceDisputed

  The Results (section 3) report 298 images and the stated classification accuracies: 89.0%, 60.0%, and 70.3%. But the Discussion (section 4) says the overall 71.1% accuracy was comparable to that of a cytotechnologist or pathologist, not that it outperformed them. (LLM, conf=0.90)

  Checked by a reasoning model — may be wrong.

  Experimental — accuracy not yet measured

May combine two claims.

AI can reduce inter-reader variability, improve consistency in nodule measurement, and assist in nodule characterization, potentially decreasing unnecessary work-up of benign nodules and expediting diagnosis of malignant ones.

“AI can reduce inter-reader variability, improve consistency in nodule measurement, and assist in nodule characterization, potentially decreasing unnecessary work-up of benign nodules and expediting diagnosis of malignant ones [15].”

• cites [15] Wu, 2024, Artificial intelligence in lung cancer screening: Detection, classification, prediction, and prognosis, Cancer Med. sourcePartly supported

  The NODULE CLASSIFICATION section says AI systems predict malignancy and characterize nodules, and the NODULE DETECTION section discusses reducing false positives. The cited copy does not establish reduced inter-reader variability or more consistent nodule measurements, nor does it link these capabilities to fewer work-ups of benign nodules or faster diagnosis of malignant ones. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

AI-powered imaging analysis and radiomics have improved detection and classification of pulmonary nodules, prediction of lymph node metastasis, and integration of molecular and radiomic features for non-invasive stratification.

“AI-powered imaging analysis and radiomics have improved detection and classification of pulmonary nodules, prediction of lymph node metastasis, and integration of molecular and radiomic features for non-invasive stratification [4,5,6].”

• cites [4] Mank, 2025, State-of-the-art artificial intelligence methods for pre-operative planning of cardiothoracic surgery and interventions: A narrative review, J. Thorac. Dis. sourcePartly supported

  The Introduction reports AI detection and classification of lung nodules, and Results > Lung, thoracic wall and mediastinal surgery reports CT-based classification of lung adenocarcinoma and improved diagnostic performance from combining conventional and radiomic features for mediastinal tumors (page numbers unavailable). Our copy does not report prediction of lymph node metastasis or integration of molecular and radiomic features for non-invasive stratification. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [5] Etienne, 2020, Artificial intelligence in thoracic surgery: Past, present, perspective and limits, Eur. Respir. Rev. sourcePartly supported

  The “Diagnostic for lung cancer” section reports AI-assisted detection and benign-versus-malignant classification of pulmonary nodules, including improved detection performance. It does not report prediction of lymph-node metastasis or integration of molecular and radiomic features for non-invasive stratification; searches for those terms found no supporting passage. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [6] Rizzo, 2005, CT images of abdomen and pelvis: Effect of nonlinear three-dimensional optimized reconstruction algorithm on image quality and lesion characteristics, Radiology sourceNot checked — no full text of the source

May combine two claims.

AI-driven systems enhance diagnostic accuracy, optimize preoperative planning, and support intraoperative decision-making in complex thoracic procedures such as lung cancer resection and mediastinal interventions.

“Recent advances in machine learning, deep learning, and computer vision have enabled AI-driven systems to enhance diagnostic accuracy, optimize preoperative planning, and support intraoperative decision-making, particularly in complex procedures such as lung cancer resection and mediastinal interventions [1,2,3].”

• cites [1] Leivaditis, V., Maniatopoulos, A.A., Lausberg, H., Mulita, F., Papatriantafyllou, A., Liolis, E., Beltsios, E., Adamou, A., Kontodimopoulos, N., and Dahm, M. (2025). Artificial intelligence in thoracic surgery: A review bridging innovation and clinical practice for the next generation of surgical care. J. Clin. Med., 14. sourcePartly supported

  Sections 3.1 and 3.8 support AI-enhanced diagnostic accuracy, thoracic surgical planning, and intraoperative decision support, including for lung cancer surgery. Section 3.1 also describes AI for mediastinal staging, but the discussion of mediastinal surgery in section 3.8 concerns robotic techniques rather than AI-driven planning or intraoperative decision-making for those procedures. Page numbers are not available in our copy. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [2] Bellini, 2021, Artificial intelligence in thoracic surgery: A narrative review, J. Thorac. Dis. sourcePartly supported

  The Thoracic lesions management section reports improved lung-nodule detection, and Preoperative evaluation and risk-assessment describes ML aiding risk assessment and therapeutic planning. But Surgical performance and planning says AI use in the operating room is limited and describes intraoperative decision support as a future prospect. It mentions mediastinal resection as a robotic procedure while explicitly distinguishing robotic-assisted surgery from AI, so it does not establish AI support for mediastinal interventions. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [3] Abbaker, N., Minervini, F., Guttadauro, A., Solli, P., Cioffi, U., and Scarci, M. (2024). The future of artificial intelligence in thoracic surgery for non-small cell lung cancer treatment: A narrative review. Front. Oncol., 14. sourcePartly supported

  The review’s “Imaging and staging” section describes AI-assisted improvements in lung-cancer diagnosis, while “Surgical candidacy” addresses preoperative assessment and “Surgical guidance and precision” describes intraoperative guidance for lung tumor resection (page numbers unavailable). Our copy does not establish the claimed application to mediastinal interventions, and some benefits are framed as potential rather than demonstrated outcomes. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

Most AI models lack robust external validation, and clinical effectiveness and cost-effectiveness data remain limited and uncertain.

“Most AI models lack robust external validation, and clinical effectiveness and cost-effectiveness data remain limited and uncertain [18].”

• cites [18] Geppert, 2025, Software with artificial intelligence-derived algorithms for detecting and analysing lung nodules in CT scans: Systematic review and economic evaluation, Health Technol. Assess. sourcePartly supported

  The Statement of principal findings says that no study provided direct evidence of clinical effectiveness or cost-effectiveness, and that the modelled findings were highly uncertain. It also reports that most included studies used retrospective datasets, while Applicability concerns notes small, enriched datasets and no prospective test-accuracy studies of consecutive clinical cohorts. These findings support concerns about validation, but the cited passages do not establish that most AI models lack robust external validation. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

Intraoperative AI applications such as augmented reality, real-time image-guided navigation, and robotic-assisted thoracic surgery have demonstrated increased surgical precision, reduced operative times, and enhanced safety.

“Intraoperatively, AI applications such as augmented reality (AR), real-time image-guided navigation, and robotic-assisted thoracic surgery (RATS) have demonstrated increased surgical precision, reduced operative times, and enhanced safety [7,8,9].”

• cites [7] Roche, 2025, Current and future applications of artificial intelligence in lung cancer and mesothelioma, Thorax sourcePartly supported

  The LUNG CANCER section (page unavailable) mentions the development of robot-assisted surgical resection. It does not describe augmented reality or real-time image-guided navigation as intraoperative AI applications, or report increased surgical precision, shorter operative times, or enhanced safety from these approaches. (LLM, conf=0.82)

  Experimental — accuracy not yet measured
• cites [8] Cusumano, G., D’Arrigo, S., Terminella, A., and Lococo, F. (2024). Artificial intelligence applications for thoracic surgeons: “The phenomenal cosmic powers of the magic lamp”. J. Clin. Med., 13. sourcePartly supported

  The Operating Room section says AI can enhance precision and safety in robotic-assisted surgery, but §6, “The Third Wish” describes these benefits as largely prospective. That section also says robotic-assisted surgery does not yet incorporate AI-based technology and is associated with prolonged—not reduced—procedural times. The cited passages do not establish the claimed benefits for augmented reality or real-time image-guided navigation. (LLM, conf=0.90)

  Experimental — accuracy not yet measured
• cites [9] Chen, 2021, Artificial intelligence-assisted display in thoracic surgery: Development and possibilities, J. Thorac. Dis. sourcePartly supported

  The “3D reconstruction/3D printing technology” section reports that intraoperative navigation with 3D-VATS reduces operational difficulty, blood loss, and length of operation. The “AR technology” section describes AR’s value for lesion navigation. The paper does not establish those outcomes for the full set of applications named in the claim, particularly robotic-assisted thoracic surgery, and describes video-based AI assistance as still in its infancy (Commentary and conclusions). (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

The proposed AI system could offer faster and more reliable intraoperative diagnostic support, potentially improving surgical outcomes.

“These findings indicate that the proposed system could offer faster and more reliable intraoperative diagnostic support, potentially improving surgical outcomes [52].”

• cites [52] Liu, H.C., Lin, M.H., Chang, W.C., Zeng, R.-C., Wang, Y.-M., and Sun, C.-W. (2023). Rapid on-site AI-assisted grading for lung surgery based on optical coherence tomography. Cancers, 15. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Kanavati et al. emphasize that deep-learning systems can provide critical support during surgery by offering real-time guidance.

“Kanavati et al. emphasize how deep-learning (DL) systems can provide critical support during surgery by offering real-time guidance.”

• cites [47] Kanavati, 2020, Weakly supervised learning for lung carcinoma classification using deep learning, Sci. Rep. sourcePartly supported

  The Introduction says deep learning can aid pathologists in obtaining diagnoses, and the Discussion describes potential uses in assessing lung tissue. Neither section describes real-time guidance during surgery; the paper concerns analysis of histopathology slides, including slides from surgical specimens. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

Chang et al. demonstrated that an AI-assisted predictive model enables integrated risk assessments and accommodates dynamic adjustments through clinician input, allowing it to adapt to a wide range of patient records.

“The AI-assisted predictive model not only enables integrated risk assessments but also accommodates dynamic adjustments through clinician input, allowing it to adapt to a wide range of patient records [46].”

• cites [46] Chang, Y., Hung, K., Wang, L., Yu, C.-H., Chen, C.-K., Tay, H.-T., Wang, J.-J., and Liu, C.-F. (2021). A real-time artificial intelligence-assisted system to predict weaning from ventilator immediately after lung resection surgery. Int. J. Environ. Res. Public Health, 18. sourceSupported

May combine two claims.

These studies addressed a five-category classification task-distinguishing lepidic, acinar, papillary, micropapillary, and solid growth patterns-and showed that AI can support accurate subtype identification and deeper characterization of tumor morphology.

“These studies addressed a complex five-category classification task—distinguishing lepidic, acinar, papillary, micropapillary, and solid growth patterns—and showed that AI can support both accurate subtype identification and a deeper characterization of tumor morphology.”

• cites [55] Sheikh, T.S., Kim, J., Shim, J., and Cho, M. (2022). Unsupervised learning based on multiple descriptors for WSIs diagnosis. Diagnostics, 12. sourceSupported
• cites [56] DiPalma, 2021, Resolution-based distillation for efficient histology image classification, Artif. Intell. Med. sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

Dynamic modeling suggests that long-term value is maximized when adaptive learning and workflow integration are considered, but indirect costs and equity issues are often underreported in current analyses.

“Dynamic modeling suggests that long-term value is maximized when adaptive learning and workflow integration are considered, but indirect costs and equity issues are often underreported in current analyses [59].”

• cites [59] 2025, Systematic review of cost effectiveness and budget impact of artificial intelligence in healthcare, NPJ Digit. Med. sourcePartly supported

  The Discussion (page not provided) calls for dynamic models that capture adaptive learning and says several studies omit indirect costs and integration expenses. It also identifies insufficient attention to equity in economic evaluations. But the paper says such modeling would improve estimates of long-term cost-effectiveness; it does not establish that considering adaptive learning and workflow integration maximizes long-term value. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

A Chinese retrospective study of 100 patients with solitary sub-solid nodules confirmed pathologically as minimally invasive or invasive adenocarcinoma found that an integrated model combining CT-based features and radiomic signatures showed strong discrimination in the training set (AUC 0.943) and validation set (AUC 0.912).

“An integrated model combining CT-based features such as nodule size, shape, margins, and radiomic signatures was constructed, showing strong discrimination in both the training set (AUC 0.943) and validation set (AUC 0.912) [30].”

• cites [30] Chen, B.T., Chen, Z., Ye, N., Mambetsariev, I., Fricke, J., Daniel, E., Wang, G., Wong, C.W., Rockne, R.C., and Colen, R.R. (2020). Differentiating peripherally located small-cell lung cancer from non-small cell lung cancer using a CT radiomic approach. Front. Oncol., 10. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

AI represents a promising solution for reducing observational variability and supporting more consistent staging practices.

“Minimizing observational variability is essential, and AI represents a promising solution for reducing such errors and supporting more consistent staging practices [41].”

• cites [41] Zhao, 2018, 3D deep learning from CT scans predicts tumor invasiveness of subcentimeter pulmonary adenocarcinomas, Cancer Res. sourceNot checked — no full text of the source

May combine two claims.

Coudray et al. predicted the ten most frequent mutations in lung adenocarcinoma by training neural networks, and six mutations (KRAS, STK11, TP53, EGFR, SETBP1, and FAT1) were identifiable from pathology images with accuracies ranging from 73.3% to 85.6%.

“Coudray et al. proposed that specific gene alterations might influence the morphological appearance of cancer cells on histologic slides; by training neural networks, they were able to predict the ten most frequent mutations in lung adenocarcinoma [36].”

• Also stated in: § 4.3. Biomarkers · plain

  “Six of these mutations (KRAS, STK11, TP53, EGFR, SETBP1, and FAT1) were identifiable from pathology images with accuracies ranging from 73.3% to 85.6% [36].”

• cites [36] Coudray, 2018, Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning, Nat. Med. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

ML methods have been used to optimize beam configurations, estimate dose-volume histograms, evaluate radiation dose and toxicity, and support clinical decision-making.

“ML methods have been used to optimize beam configurations, estimate dose–volume histograms, evaluate radiation dose and toxicity, and support clinical decision-making [42].”

• cites [42] Kureshi, 2016, A predictive model for personalized therapeutic interventions in non-small cell lung cancer, IEEE J. Biomed. Health Inform. sourceNot checked — no full text of the source

May combine two claims.

A deep convolutional neural network on 301 lung carcinoma CT images identified lung cancer with sensitivity 0.93, precision 0.82, and overall accuracy approximately 71%, comparable to cytotechnologists and pathologists.

“Probabilities for three cancer types were estimated, and threefold cross-validation reported a sensitivity of 0.93, precision of 0.82, and overall accuracy of approximately 71%, comparable to that of cytotechnologists and pathologists.”

• cites [31] Teramoto, 2017, Automated classification of lung cancer types from cytological images using deep convolutional neural networks, BioMed Res. Int. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Deep-learning-based analysis has the potential to support pathologists in identifying tumor subtypes and genetic mutations, and may ultimately assist clinicians in the early diagnosis and screening of lung cancer.

“Overall, these findings suggest that deep-learning-based analysis has the potential to support pathologists in identifying tumor subtypes and genetic mutations, and may ultimately assist clinicians in the early diagnosis and screening of lung cancer.”

May combine two claims.

AI algorithms based on deep learning and convolutional neural networks have demonstrated high sensitivity up to 94.6% and specificity up to 93.6% for lung nodule detection, often matching or surpassing radiologist performance, especially for small nodules.

“AI algorithms, especially those based on deep learning and convolutional neural networks, have demonstrated high sensitivity (up to 94.6%) and specificity (up to 93.6%) for lung nodule detection, often matching or surpassing radiologist performance, especially for small nodules [13,14].”

• cites [13] Thong, 2023, Diagnostic test accuracy of artificial intelligence-based imaging for lung cancer screening: A systematic review and meta-analysis, Lung Cancer sourceNot checked — no full text of the source
• cites [14] Adams, 2023, Lung cancer screening, Lancet sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Postoperative management benefits from AI-driven predictive models and wearable monitoring devices, enabling early complication detection and improved patient follow-up.

“Postoperative management benefits from AI-driven predictive models and wearable monitoring devices, enabling early complication detection and improved patient follow-up [10,11].”

• cites [10] Kim, 2021, Applications of artificial intelligence in the thorax: A narrative review focusing on thoracic radiology, J. Thorac. Dis. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites [11] Yang, D., Miao, Y., Liu, C., Zhang, X., Wang, L., Li, J., Zhao, H., Chen, Q., Zhou, Y., and Sun, K. (2024). Advances in artificial intelligence applications in the field of lung cancer. Front. Oncol., 14. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

The quote only partly states this claim.

A predictive model incorporating clinical parameters, radiologic characteristics of pulmonary nodules, LDCT-derived AI outputs, and liquid biopsy data demonstrated sensitivity of 89.53%, specificity of 81.31%, and an AUC of 0.880 in the training cohort.

“In another prospective investigation, a predictive model incorporating clinical parameters (age and smoking status), radiologic characteristics of pulmonary nodules (including nodule size, number, upper-lobe location, malignant edge features, and subsolid appearance), LDCT-derived AI outputs, and liquid biopsy data demonstrated optimal performance in the training cohort, with a sensitivity of 89.5”

• cites [34] Scott, 2008, Biomarkers in lung cancer: From early detection to novel therapeutics and decision making, Biomark. Med. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Integrating multiple biomarkers such as HE4, sVCAM-1, TTR, ApoA2, and CEA can markedly enhance diagnostic performance; one such model reached a sensitivity of 93.33%, specificity of 92.00%, and an AUC of 0.988.

“Additional studies have reported that integrating multiple biomarkers—such as human epididymis protein 4 (HE4), soluble vascular cell adhesion molecule-1 (sVCAM-1), transthyretin (TTR), apolipoprotein A2 (ApoA2), and the carcinoembryonic antigen CEA—can markedly enhance diagnostic performance. One such model reached a sensitivity of 93.33%, specificity of 92.00%, and an AUC of 0.988, indicating”

• cites [24] Huang, 2023, Artificial intelligence in lung cancer diagnosis and prognosis: Current application and future perspective, Semin. Cancer Biol. sourceNot checked — no full text of the source

May combine two claims.

Using AI as a prescreener, where radiologists only review AI-positive scans, can reduce workload and maintain sensitivity, while AI as an assistant or backup may increase recall rates but also false positives and interpretation time.

“Implementation scenarios vary: using AI as a prescreener (where radiologists only review AI-positive scans) can reduce workload and maintain sensitivity, while AI as an assistant or backup may increase recall rates but also false positives and interpretation time [17].”

• cites [17] Lee, 2025, Artificial intelligence for low-dose CT lung cancer screening: Comparison of utilization scenarios, Am. J. Roentgenol. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Staging accuracy can be influenced by the radiologist's experience, and the complex appearance of lung nodules on CT increases the risk of interpretive errors.

“However, staging accuracy can be influenced by the radiologist’s experience, and the complex appearance of lung nodules on CT increases the risk of interpretive errors.”

• cites [40] Baker, 2013, Malpractice suits in chest radiology: An evaluation of the histories of 8265 radiologists, J. Thorac. Imaging sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI technologies are well suited to managing large volumes of repetitive, image-based tasks, making them valuable tools for assisting clinicians in visually intensive workflows.

“AI technologies are well suited to managing large volumes of repetitive, image-based tasks, making them valuable tools for assisting clinicians in visually intensive workflows.”

May combine two claims.

Machine learning applied to FDG-PET imaging for lung cancer detection achieved sensitivities of 95.9% and 91.5% and specificities of 98.1% and 94.2% for standard and ultralow-dose scans, respectively, suggesting that ML approaches can detect lung cancer even at minimal radiation exposure of 0.11 mSv.

“Additionally, machine learning (ML) has been applied to FDG-PET imaging for lung cancer detection, achieving sensitivities of 95.9% and 91.5% and specificities of 98.1% and 94.2% for standard and ultralow-dose scans, respectively, suggesting that ML approaches can detect lung cancer even at minimal radiation exposure of 0.11 mSv [28].”

• cites [28] Sun, 2020, Radiomics for lung adenocarcinoma manifesting as pure ground-glass nodules: Invasive prediction, Eur. Radiol. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Kureshi et al. proposed a data-driven predictive model for assessing tumor response to EGFR-TKI treatment in advanced NSCLC; incorporating clinical profiles, environmental exposures, and EGFR mutation status, the model reached an accuracy of 76%.

“Kureshi et al. proposed a data-driven predictive model for assessing tumor response to EGFR-TKI treatment in advanced NSCLC; by incorporating clinical profiles, environmental exposures, and EGFR mutation status, the model reached an accuracy of 76% [42].”

• cites [42] Kureshi, 2016, A predictive model for personalized therapeutic interventions in non-small cell lung cancer, IEEE J. Biomed. Health Inform. sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

By expediting interpretation of CT scans and pathology slides, AI has the potential to streamline and enhance the accuracy of lung cancer staging.

“By expediting interpretation of CT scans and pathology slides, AI has the potential to streamline and enhance the accuracy of lung cancer staging.”

May combine two claims.

Manual review may lead to missed or incorrect diagnoses, complicating efforts to identify lung cancer early.

“Manual review may therefore lead to missed or incorrect diagnoses, complicating efforts to identify lung cancer early [40].”

• cites [40] Baker, 2013, Malpractice suits in chest radiology: An evaluation of the histories of 8265 radiologists, J. Thorac. Imaging sourceNot checked — no full text of the source

May combine two claims.

When applied as a second reader for CT and PET imaging, AI can reduce radiologists' workload while improving the detection of suspicious nodules.

“When applied as a second reader for CT and PET imaging, AI can reduce radiologists’ workload while improving the detection of suspicious nodules.”

• cites [39] Masood, 2018, Computer-assisted decision support system in pulmonary cancer detection and stage classification on CT images, J. Biomed. Inform. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

ML and radiomic analyses offer non-invasive means to better evaluate the tumor and its microenvironment, thereby aiding in anticipating treatment responses.

“ML and radiomic analyses, however, offer non-invasive means to better evaluate the tumor and its microenvironment, thereby aiding in anticipating treatment responses.”

• cites [42] Kureshi, 2016, A predictive model for personalized therapeutic interventions in non-small cell lung cancer, IEEE J. Biomed. Health Inform. sourceNot checked — no full text of the source

May combine two claims.

The AI approach demonstrated markedly superior discrimination among minimally invasive adenocarcinoma (MIA), invasive adenocarcinoma (IA), and normal lung tissue, achieving an overall accuracy of 84.9%, whereas frozen sections reached only 20%.

“The AI approach demonstrated markedly superior discrimination among minimally invasive adenocarcinoma (MIA), invasive adenocarcinoma (IA), and normal lung tissue, achieving an overall accuracy of 84.9%, whereas frozen sections reached only 20%.”

• cites [52] Liu, H.C., Lin, M.H., Chang, W.C., Zeng, R.-C., Wang, Y.-M., and Sun, C.-W. (2023). Rapid on-site AI-assisted grading for lung surgery based on optical coherence tomography. Cancers, 15. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

AI integration into surgical planning and intraoperative guidance can reduce operative times and complication rates, which translates to lower direct healthcare costs and improved quality-adjusted life years (QALYs).

“AI integration into surgical planning and intraoperative guidance can reduce operative times and complication rates, which translates to lower direct healthcare costs and improved quality-adjusted life years (QALYs).”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

AI has the potential to transform surgical practice by offering real-time analysis during operations, supporting enhanced decision-making, and improving overall surgical outcomes.

“AI has the potential to transform surgical practice by offering real-time analysis during operations, supporting enhanced decision-making, and improving overall surgical outcomes.”

May combine two claims.

Real-time histologic assessment allows surgeons to adjust operative strategy as needed, promoting complete tumor removal while avoiding unnecessary excision of healthy structures.

“Such real-time histologic assessment allows surgeons to adjust their operative strategy as needed, promoting complete tumor removal while avoiding unnecessary excision of healthy structures.”

• cites [53] Pao, 2023, Predicting EGFR mutational status from pathology images using a real-world dataset, Sci. Rep. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Sensitivity and specificity were higher for MIA (89% and 82.7%) and IA (94% and 80.6%) with the AI approach.

“Sensitivity and specificity were also higher for both MIA (89% and 82.7%) and IA (94% and 80.6%), respectively.”

• cites [52] Liu, H.C., Lin, M.H., Chang, W.C., Zeng, R.-C., Wang, Y.-M., and Sun, C.-W. (2023). Rapid on-site AI-assisted grading for lung surgery based on optical coherence tomography. Cancers, 15. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

Generating extensive annotations depends heavily on expert pathologists, making the process slow and resource-intensive.

“In practice, however, generating extensive annotations depends heavily on expert pathologists, making the process slow and resource-intensive.”

May combine two claims.

While current clinical use of robotic platforms remains at the level of robot assistance (autonomy level 1), research is progressing toward higher autonomy, with conditional autonomy demonstrated in preclinical models.

“While current clinical use remains at the level of robot assistance (autonomy level 1), research is progressing toward higher autonomy, with conditional autonomy demonstrated in preclinical models [50]”

• cites [50] Vasey, 2023, Intraoperative Applications of Artificial Intelligence in Robotic Surgery: A Scoping Review of Current Development Stages and Levels of Autonomy, Ann. Surg. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

No clinical investigations of autonomous surgical AI on human patients have been reported, with all studies remaining at preclinical stages using in silico simulations, ex vivo models, or animal experiments.

“Critically, no clinical investigations of autonomous surgical AI on human patients have been reported, with all studies remaining at preclinical stages using in silico simulations, ex vivo models, or animal experiments [54].”

• cites [54] Lee, 2024, Levels of autonomy in FDA-cleared surgical robots: A systematic review, NPJ Digit. Med. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI-driven tools now enhance tumor detection, risk stratification, intraoperative navigation, and postoperative monitoring, improving precision and workflow efficiency.

“AI-driven tools now enhance tumor detection, risk stratification, intraoperative navigation, and postoperative monitoring, improving precision and workflow efficiency.”

May combine two claims.

When AI outputs appear opaque or difficult to interpret, it may undermine clinician confidence and slow down broader clinical adoption.

“When AI outputs appear opaque or difficult to interpret, it may undermine clinician confidence and slow down broader clinical adoption.”

May combine two claims.

Li et al. introduce a strategy that merges 3D printing with augmented-reality visualization; when patient-specific 3D-printed lung models are paired with AR during surgery, spatial orientation improves significantly, contributing to reduced operative duration, decreased blood loss, and shorter hospitalization.

“Li et al. introduce an innovative strategy that merges three-dimensional (3D) printing with augmented-reality (AR) visualization.”

• Also stated in: § 6.2. Intraoperative Support · plain

  “When these models are paired with AR during surgery, spatial orientation improves significantly, contributing to reduced operative duration, decreased blood loss, and shorter hospitalization [49].”

• cites [49] Li, 2021, Augmented reality and 3-dimensional printing technologies for guiding complex thoracoscopic surgery, Ann. Thorac. Surg. sourceNot checked — no full text of the source

May combine two claims.

The initial investment in AI infrastructure-including software, hardware, and staff training-remains substantial, and the economic impact is sensitive to implementation costs and local resource constraints.

“However, the initial investment in AI infrastructure—including software, hardware, and staff training—remains substantial, and the economic impact is sensitive to implementation costs and local resource constraints.”

May combine two claims.

Esteva H et al. investigated artificial neural networks for estimating postoperative outcomes following lung resection; neural networks demonstrated greater flexibility and individualized predictive power, achieving nearly 100% sensitivity and specificity for patient outcomes.

“Esteva H, et al. investigated the use of artificial NN—modeled after the human neural system—for estimating postoperative outcomes following lung resection, comparing them to conventional risk indices. NN demonstrated greater flexibility and individualized predictive power, achieving nearly 100% sensitivity and specificity for patient outcomes [44].”

• cites [44] Esteva, 2002, Neural networks as a prognostic tool of surgical risk in lung resections, Ann. Thorac. Surg. sourceNot checked — no full text of the source

May combine two claims.

Pao et al.'s study highlights that deep learning models can detect fine-grained pathological patterns during surgery and deliver immediate feedback about tissue characteristics.

“Their study highlights how deep learning models can detect fine-grained pathological patterns during surgery and deliver immediate feedback about tissue characteristics.”

• cites [53] Pao, 2023, Predicting EGFR mutational status from pathology images using a real-world dataset, Sci. Rep. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

May combine two claims.

These economic benefits are driven by earlier detection, fewer missed diagnoses, and streamlined care pathways.

“These economic benefits are driven by earlier detection, fewer missed diagnoses, and streamlined care pathways.”

May combine two claims.

Traditional risk assessment tools such as the Goldman index for cardiac risk and the Torrington index for respiratory risk are effective in categorizing patients into broad risk categories but fall short in providing precise, individualized predictions of operative risk.

“Traditional risk assessment tools, such as the Goldman index for cardiac risk and the Torrington index for respiratory risk, are effective in categorizing patients into broad risk categories but fall short in providing precise, individualized predictions of operative risk [44].”

• cites [44] Esteva, 2002, Neural networks as a prognostic tool of surgical risk in lung resections, Ann. Thorac. Surg. sourceNot checked — no full text of the source

May combine two claims.

AI in thoracic surgery demonstrates potential for cost-effectiveness primarily by improving diagnostic accuracy, reducing unnecessary procedures, and optimizing resource utilization.

“Artificial intelligence in thoracic surgery demonstrates potential for cost-effectiveness primarily by improving diagnostic accuracy, reducing unnecessary procedures, and optimizing resource utilization.”

• cites [59] 2025, Systematic review of cost effectiveness and budget impact of artificial intelligence in healthcare, NPJ Digit. Med. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Although many models demonstrate strong performance in classification tasks, little progress has been made in improving common-sense reasoning, particularly understanding subtle cellular or tissue-level characteristics.

“Although many models demonstrate strong performance in classification tasks, little progress has been made in improving common-sense reasoning—particularly when it comes to understanding subtle cellular or tissue-level characteristics.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI is rapidly transforming oncologic thoracic surgery, advancing from decision support in diagnostics and surgical planning to integration with robotic-assisted and image-guided interventions.

“AI is rapidly transforming oncologic thoracic surgery, advancing from decision support in diagnostics and surgical planning to integration with robotic-assisted and image-guided interventions.”

May combine two claims.

Current surgical robots in thoracic surgery operate predominantly at Level 1 (Robot Assistance), with 86% of FDA-cleared surgical robots functioning at this level.

“Current surgical robots in thoracic surgery operate predominantly at Level 1 (Robot Assistance), with 86% of FDA-cleared surgical robots functioning at this level.”

• cites [54] Lee, 2024, Levels of autonomy in FDA-cleared surgical robots: A systematic review, NPJ Digit. Med. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• AssumptionRests on an assumption: The five-level LASR classification is the most widely adopted classification system for autonomy in surgical robotics.

May combine two claims.

The HMI can highlight suspicious areas on live images and automatically assess tumor grade, supporting intraoperative decision-making.

“This HMI can highlight suspicious areas on live images and automatically assess tumor grade, thereby supporting intraoperative decision-making [52].”

• cites [52] Liu, H.C., Lin, M.H., Chang, W.C., Zeng, R.-C., Wang, Y.-M., and Sun, C.-W. (2023). Rapid on-site AI-assisted grading for lung surgery based on optical coherence tomography. Cancers, 15. sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.