Advantages of integrating artificial intelligence and spectral CT for lung nodule classification and prognostic judgment: a narrative review

Introduction

Lung cancer continues to rank among the leading causes of cancer-related deaths worldwide. Its persistently high mortality is largely attributable to late-stage diagnosis, which drastically narrows treatment possibilities and leads to unfavorable survival outcomes. The global burden of the disease remains substantial, underscoring an urgent need for more effective and accessible interventions (1,2).

The disease remains one of the primary contributors to cancer-related fatalities around the world. Lung cancer’s high mortality rate persists due to frequent late-stage diagnosis, which severely limits therapeutic options and contributes to poor survival outcomes. Current treatment strategies—including surgery, radiotherapy, chemotherapy, targeted therapy, and immunotherapy—face significant challenges such as tumor heterogeneity, drug resistance, and access to personalized treatment regimens. These hurdles underscore the critical need for innovative strategies that enhance early detection and expand accessible, effective therapies to reduce the global burden of the disease (3-5). This situation highlights the vital necessity of early disease detection. Identifying lung cancer in its initial stages can markedly enhance the chances of effective treatment, ultimately leading to better patient outcomes. Furthermore, early diagnosis has the potential to reduce mortality rates associated with this disease, highlighting the vital role that timely screening and intervention play in managing lung cancer cases successfully (6,7). It is important to note that the presence of multiple lung nodules can result in a high degree of similarity in their imaging appearances. Whereas early imaging manifestations of lung cancer are dominated by lung nodules, qualitative imaging studies of lung nodules are very difficult to perform (8). Regarding the different characteristics of pulmonary nodules, we defined them into the following four simple categories: morphological characteristics (e.g., lobulation, spiculation, pleural pulling); attenuation (solid, ground glass, part-solid); positional characteristics (central, peripheral); and dynamic characteristics [growth rate, trend in iodine concentration (IC)] (9-11) (Figure 1). This phenomenon, known as “different diseases with the same image”, can easily occur in such cases. This predicament presents a significant challenge to clinical diagnosis, treatment selection, and prognosis assessment. The partial subjectivity of some imaging physicians in morphologic judgment, coupled with the absence of sufficiently quantitative and objective criteria, contributes to this challenge.

Figure 1 Brief classification of lung nodules with different characteristics.

The advent of artificial intelligence (AI) algorithms, particularly in the domain of deep learning and substantial progress in image recognition, has precipitated a marked escalation in the utilisation of AI in the field of medical imaging in recent years (12-14). Recent advancements in the field of AI have led to significant progress in the study of lung nodule differentiation, both benign and malignant. The integration of AI with precision recognition of morphological signs, which are often imperceptible to the unassisted eye, has enabled the accurate identification of the benign and malignant nature of these nodules, as well as their potential regression (15,16). Convolutional neural networks (CNNs), as one of the deep learning algorithms, have been proven to efficiently and accurately extract complex features related to lung nodule morphology. This technological breakthrough holds promise for distinguishing benign from malignant lung nodules while also aiding in the classification of pathological subtypes (17). The maturation of its use in medical imaging has also established a solid foundation for the screening of individuals at high risk of malignant lung nodules, as well as for accurate diagnosis and treatment.

Spectral computed tomography (CT) is an advanced imaging technology that overcomes the limitations of conventional single-parameter methods. This technology offers several key advantages, including enhanced image quality and reduced radiation dose. Consequently, it has become a widely adopted modality in medical imaging, particularly for diagnostic procedures such as urinary stone analysis, tumour detection, cardiovascular system assessment, and various other medical conditions (18-23). It has been demonstrated by certain scholars that spectral CT is capable of facilitating the qualitative diagnosis of lung nodules (24,25). Furthermore, AI technology has been shown to enhance the accuracy of qualitative judgements made by spectral CT, due to the quantitative parameter data provided by AI technology (26). A considerable number of studies have utilized a single modality, either spectral CT or AI, to predict the characteristics of lung nodules (27-29). However, there is a scarcity of research that effectively integrates these two techniques. In this paper, we review the advancements in studies that combine spectral CT and AI to analyze various characteristics of lung nodules. We present this article in accordance with the Narrative Review reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1246/rc).

Methods

This study is a narrative review and did not use the standardized search and screening process of a systematic review. For specific screening criteria, refer to Table 1.

Search strategy

We selected PubMed/MEDLINE, Web of Science, and Google Scholar for a comprehensive search. Keywords included “spectral CT”, “pulmonary nodule” and “artificial intelligence”. Prioritize retrieving literature from January 2010 to September 2025 to reflect the latest advancements in AI and spectral CT technology.

Inclusion criteria

Research topics focused on the following areas: applications of spectral CT in distinguishing benign from malignant pulmonary nodules, analysis of pathological subtypes, or assessment of prognosis; development or clinical validation of AI algorithms for pulmonary nodule image analysis; and exploratory studies on the combined use of AI and spectral CT. Priority was given to original research studies (such as cohort studies and diagnostic test accuracy investigations), high-quality reviews, and key technology validation papers.

Exclusion criteria

We excluded the following literature: non-lung nodule-related studies (e.g., tumors in other organs, non-neoplastic lesions); studies involving only conventional CT or a single modality (not combined with AI or spectroscopic CT); non-academic literature such as conference abstracts, case reports, and technical specifications; and studies with incomplete data. Two reviewers independently screened articles and extracted data using standardized forms. Disagreements were resolved by a third reviewer. Ultimately, we included and screened relevant literature for corresponding analyses.

Results

A total of 25 studies were included in this review. The findings are synthesized into three key areas. Table 2 (30-33) serves as a representative example for analysis and reference in this study.

Differentiation of benign and malignant lung nodules

Due to the complexity and diversity of lung nodules, manual reviews can result in missed diagnoses and misdiagnoses, which complicates the early diagnosis of lung cancer. Recent advancements in AI systems have shown promise in detecting and identifying malignant lung nodules from chest CT images (34-37). The aforementioned studies employed deep learning models, a subset of AI, to assist radiologists in identifying benign and malignant lung nodules. Specifically, these models are based on artificial neural network architectures. In a more focused investigation, one study developed a 3D DenseSharp network—a specific type of deep learning model designed for multi-task learning. This model was trained on a single cohort dataset and demonstrated higher accuracy in lesion identification and prediction than practicing radiologists (38). However, it has been shown that the recognition specificity of the AI system is lower than that of manual recognition (39). This may be due to the high rate of false positives in the AI model due to tissues such as blood vessels, bronchi, and lymph nodes within the lung tissue. Dobbins III et al. developed an automated lung segmentation method and nodule detection technique that ensures anatomical segmentation of the lung and correctly identifies all nodules (40). AI technology cannot fully replace human sensory perception in the early diagnosis of benign and malignant pulmonary nodules. However, it is increasingly being used by radiologists as an auxiliary tool in the early diagnosis of benign and malignant pulmonary nodules due to its ability to enhance the diagnostic efficiency of radiologists (41-43). In summary, while contemporary AI systems exhibit superior sensitivity in nodule detection and offer valuable assistance to radiologists, their clinical application remains constrained by limitations in specificity and interpretability. Future developments in this field should concentrate on integrating multimodal data, enhancing model transparency, and validating performance in a range of clinical settings. This will help to bridge the gap between technical capability and clinical implementation.

Presumed pathological findings in pulmonary nodules

On top of the AI determination of the benign and malignant nature of lung nodules, the correlation between imaging features and pathological subtypes was further explored, and lung nodules were classified and identified by morphological and other features on CT imaging (44). Li et al. employed the K-nearest neighbors (KNN) algorithm for lung nodule segmentation. Evaluation of this algorithm on a given dataset revealed that for solid lesions, perivascular lesions, and perivascular solid lesions, the average Tanimoto/Jaccard error values were all below 0.25. These findings provide a theoretical basis for the precise classification of lung nodules across different pathological subtypes and the application of the Lung-RADS system (33). The AI combined with the imaging histology model proposed by Lin et al. in their recent study provided accurate classification of lung nodules for different pathological subtypes and the Lung-RADS system. In addition to this, it has been shown that the introduction of Refined Radiomics and Deep Learning Features-Guided CatBoost (RRDLC)-Classifier in the pathological evaluation of stage I lung adenocarcinoma in 371 patients by machine learning and imaging histology demonstrated excellent pathological diagnostic efficacy (32,33).

Progression of lung nodules and assessment of prognosis

AI is in full swing to identify benign and malignant lung nodules and to predict pathological types, but the reliability of single metrics for prognostic assessment of lung nodules is far from adequate for today’s clinical needs (45-47). So some scholars began to use AI to start predicting the growth trend and regression of lung nodules (48,49). Notably, one study accurately predicted EGFR mutations in lung adenocarcinoma as well as squamous cell carcinoma by using AI to stage lung cancer in combination with PET and other imaging tools (50). Furthermore, a deep learning model has been developed for N2 lymph node metastasis prediction and prognostic stratification in stage I NSCLC. This has led to a more detailed and reliable prognostic assessment of NSCLC due to pulmonary malignant nodules (51). With the development of AI, the ITEN (Impact of Non-Small Cell Lung Cancer Therapy Evolution) model has been further refined in deep learning to become a reliable tool for predicting the impact of non-small cell lung cancer (NSCLC) therapies on cost as well as survival (52,53).

We have compiled data on AI techniques for diagnosis and treatment of lung nodules, as shown in Table 3 (22,29,37,54). Summarising the above studies, it is not difficult to conclude that AI can play an important role in the differential diagnosis of benign and malignant lung nodules, the prediction of pathological results, and the assessment of the progression and prognosis of lung nodules through the analysis of lung imaging data and deep learning. AI can assist imaging physicians and clinicians in the full cycle of assessment of lung nodules, from nature to prognosis, from screening to treatment, reducing the burden on physicians while improving the accuracy of treatment, making the diagnosis and treatment of lung nodules precise and individualised.

Energy spectrum CT in the diagnosis of lung nodules

Spectral CT for identification of benign and malignant lung nodules

We summarized for spectral CT identification of benign and malignant lung nodules as shown in Table 4 (46,48,50,51,55). Clinical classification of pulmonary nodules includes solid pulmonary nodules, ground glass nodules (GGNs), etc. The misdiagnosis rate is high in routine pulmonary nodule imaging based solely on the morphological features of the nodules. However, the multiparameter and quantitative analysis of spectral CT provides the possibility of evaluating the benign or malignant nature of lung nodules and assessing the degree of malignancy by grading. Based on the existing in-depth studies on the diagnosis of lung nodules by spectral CT, we believe that the value of spectral CT in the study of benign and malignant differentiation of lung nodules has theoretical basis. Since the imaging parameters of spectral CT are related to iodine content, its IC, normalised iodine concentration (NIC) and slope of spectral attenuation curves (λHU), which reflect the blood supply as well as compositional information. Among them, spectral CT is a definite reference for the identification of benign and malignant GGNs (55). The blood supply and infiltration correlation of GGN can be reflected by the energy spectrum CT iodogram (56). In addition, the variability of IC, NIC, and λHU can provide a basis for the differentiation of benign and malignant pulmonary nodules, Son et al. compared the spectral CT parameters of inflammatory, malignant neoplastic, and tuberculosis-induced isolated pulmonary nodules. The results showed higher values of IC, NIC, and λHU in malignant isolated pulmonary nodules compared to inflammatory isolated pulmonary nodules, while IC, NIC, and λHU were lower in tuberculous isolated pulmonary nodules compared to inflammatory isolated pulmonary nodules (30). For spectral CT, IC is used as an indicator to reflect the blood supply of lung nodules, and the change of IC value can reflect different pathological types. Since the blood vessels in the tumour tissue are tortuous and dilated, and the gap in the endothelium of the blood vessels is enlarged, the blockage of blood flow in malignant nodules is less significant than that in benign nodules, and from this we can preliminarily deduce the benignness of the lung nodule by the IC value. However, some scholars have shown that the IC, NIC, and λHU of malignant tumours at 40 kiloelectronvolt (keV) are lower than those of lung inflammation, which may be related to the excessive growth of lung cancer lesions and the immature development of blood vessels leading to necrosis within the tumour (57). It may also be due to the high IC, NIC, and λHU caused by the inflammatory factors contained in acute inflammation stimulating the massive proliferation of blood vessels in the lesion. In chronic inflammation, the blood vessels are destroyed by inflammation and blood flow is reduced, resulting in low IC, NIC, and λHU. The aforementioned study confirms that spectral CT holds potential advantages in distinguishing benign from malignant pulmonary nodules.

Spectral CT for identification of pathological subtypes and assessment of invasiveness of malignant lung nodules

In order to meet the clinical needs, based on the identification of benign and malignant lung nodules by spectral CT, it has become a hot research direction to evaluate the subtype identification of lung cancer, invasiveness of malignant lung nodules and the degree of differentiation by spectral CT. Shi et al. found that the combined tumour markers of spectral CT differed between squamous cell carcinomas, adenocarcinomas, and neuroendocrine tumours, and that the IC of VP and AP, as well as λHU, differed between the groups, and could be relied upon to provide a preliminary differentiation between the above three different pathologic types of tumours (54). However, in other scholars’ studies, adenocarcinoma had a higher IC than squamous cell carcinoma (58). However, the difference in these parameters between the two types of cancer was only significant in the venous phase, but not in the arterial phase (59). The study by Lin et al. addressed the association between Ki-67 expression and EGFR mutation status with energetic CT parameters in NSCLC and showed that tumour grade and slope of the energetic CT curve in the venous stage were independent factors affecting Ki-67, and the possible factors for the differences in the expression of Ki-67 could be due to the different composition and components of tumour tissues (60). The elevation of its index is related to the degree of malignancy of the tumour, which may be due to the complexity of the internal environment of the tumour, such as the abnormality of the nucleoplasmic ratio of the cells, the compactness of the extracellular space as well as the elevation of the content of macromolecular proteins. In addition, studies have developed spectral CT scanning-derived parameters and tumor abnormal protein levels to predict the aggressiveness of mixed gross glass nodules, and multiple quantitative and functional parameters were derived to predict the pathologic aggressiveness of mGGNs, with Zeff a playing a prominent role (61) his study demonstrates the feasibility of combining tumor markers with spectral CT for the study of lung nodule aggressiveness.

Prognostic assessment of lung cancer and prediction of metastasis by spectral CT

As the use of spectral CT is becoming more sophisticated in the assessment of benign and malignant lung nodules and the determination of pathological subtypes, there is an increasing demand for the use of spectral CT for the prediction of metastasis and prognostic assessment of lung cancer. In the study of Dewaguet et al., the energy spectrum CT parameters of lung metastases of renal cell carcinoma, colorectal cancer, head and neck cancer, etc. were investigated to provide data support for the quantitative IC reference range of lung metastases (62). In the metastasis of lung cancer, the role of lymph nodes is crucial, and the number of lymph node metastases is also related to the stage of lung cancer, which is directly related to the prognosis of patients. Among the indices of spectral CT, the lymph nodes of lung cancer (lymph node) can be evaluated by the values of lymph node λHU as well as IC. IC as well as λHU were significantly lower in enlarged lymph node as compared to normal lymph nodes. Kaup et al. found that the energy spectrum CT-derived iodine content differed significantly between normal, inflammatory and metastatic squamous cell carcinoma of the lung lymph nodes. When the NIC value was 0.43, the sensitivity of detecting lymph node metastasis was 80% and 75%, the specificity reached 70% and 75%, the positive predictive value was 70% and 75%, and the negative predictive value was 76% and 75%, respectively, with an accuracy of up to 73% and 75% (63). The contradictory findings regarding spectral CT parameters, such as IC values, across different studies may be attributed to several factors. Firstly, variations in study populations and lesion characteristics—such as differences in the proportion of inflammatory activities, the degree of tumour necrosis, or vascularization patterns—could significantly influence quantitative measurements. Secondly, technical factors, including variations in CT scanner models, imaging protocols, contrast agent administration regimens, and post-processing methods, may contribute to discrepancies between studies. Furthermore, the temporal parameters of measurement relative to contrast injection (for example, arterial vs. venous phase) and region-of-interest placement strategies may also affect parameter quantification. These inconsistencies underscore the necessity for standardised imaging protocols and rigorous adjustment for confounding variables in future studies.

The above study, in terms of benign and malignant differentiation of lung nodules, determination of pathological subtypes, and assessment of prognosis and metastasis by spectroscopic CT, illustrates that spectroscopic CT carries out multiparametric imaging from quantitative material research to conduct an all-round and full-cycle study of lung nodules. With the continuous development of spectral CT technology, spectral CT will play an important role in the field of thoracic tumours. Although spectral CT is ahead of other imaging examinations in the qualitative material and multi-parameter level, in the present time of diversified information processing, the combination with AI in the accurate quantitative will be more accurate diagnosis.

Application of AI combined with energetic spectrum CT in the diagnosis of lung nodules

The analysis of the aforementioned two components demonstrates that AI technology possesses remarkable capabilities in data quantification. Additionally, the imaging benefits of energy spectrum CT create a solid foundation that enables AI to acquire more precise quantitative data. Notably, certain scholars have suggested a novel diagnostic method that integrates energy spectrum CT with AI for imaging the spinal cord. Furthermore, they have introduced age- and gender-specific bone marrow attenuation curves based on a substantial cohort of healthy individuals, which serves as a valuable reference for future investigations into trabecular bone structures (64). Furthermore, the integration of AI with energy-spectrum CT extends beyond its applications within the skeletal system. We posit that this powerful combination could significantly enhance the detection of both benign and malignant lung nodules and infiltrates. Additionally, it holds promise for improving prognostic and preventive screening measures in lung health. In a study conducted by Wang et al., the authors explored the effectiveness of merging deep learning techniques with spectral CT in evaluating lymph node metastasis associated with primary lung tumors. The findings of the study indicate that the five-fold cross-validation method is more effective in accurately predicting metastatic lymph nodes, particularly at lower energy levels. This suggests that advancements in imaging technology, when coupled with sophisticated AI methods, can greatly improve diagnostic accuracy in oncology (65). Research indicates that energy-spectral CT offers substantial enhancements in the sensitivity of AI when it comes to diagnosing lung nodules. Compared to traditional single-energy CT, energy-spectral CT proves to be a more effective diagnostic tool. This advancement not only allows for more accurate identification of lung nodules but also facilitates the efficient screening of a larger patient population who may be affected by this condition. The implications of such a technology could lead to earlier detection and intervention, ultimately improving patient outcomes in the management of lung nodules (66). This study evaluates the utility of SDCT-Zeff-based imageomics, deep learning, and clinical features in distinguishing between GGNs characteristic precursor glandular lesions (PGLs) and adenocarcinomas, which further illustrates the promise of AI combined with spectroscopic CT in the diagnosis and treatment of lung nodules (67). The dual approach is integrated into the routine process of lung cancer screening through the development of standardized workflows (e.g., AI-assisted spectral CT parameter analysis platform). AI can automatically extract spectral CT multi-parameter data (e.g., IC, attenuation curves), generate visual reports, and assist imaging physicians in quickly determining the nature of nodules. Initially, it can be preferentially applied to high-risk population screening, and subsequently optimize the algorithm generalizability through multi-center cooperation.

Discussion

Lung cancer is a major cause of cancer-related mortality worldwide, with a high mortality rate largely attributable to late-stage diagnosis. This has the effect of significantly restricting treatment options and leading to a poor prognosis. Notwithstanding the advances in current therapeutic approaches—including surgery, radiotherapy, chemotherapy, targeted therapy and immunotherapy—there remain numerous challenges, including tumour heterogeneity, drug resistance and accessibility to personalised treatment regimens. It is therefore evident that the primary method by which to reduce the burden of lung cancer is to achieve early diagnosis and effective intervention.

Precise management of pulmonary nodules is crucial for early lung cancer diagnosis. The present review focuses on the application of AI combined with spectral CT in pulmonary nodule assessment. The findings of this study demonstrate that this dual-technology approach shows significant potential in distinguishing benign from malignant nodules, classifying pathological subtypes, and predicting prognosis. Spectral CT provides quantitative parameters, including IC and spectral curve slope, which reflect nodule blood supply and composition information. Concurrently, AI has been demonstrated to efficiently extract deep features from images, thereby enabling precise nodule characterisation. The combination of these elements is poised to enhance diagnostic objectivity and accuracy, while also demonstrating the potential to optimise clinical workflows and facilitate comprehensive lifecycle management of pulmonary nodules.

However, extant studies are largely confined to single-centre, small-sample designs and lack direct comparisons with conventional CT combined with AI approaches, thus limiting comprehensive validation of the added value of spectral CT. In addition, the interpretability of AI models is inadequate, and the financial burden of acquiring spectral CT equipment is a significant obstacle to its clinical implementation. It is recommended that future efforts concentrate on multi-centre, prospective studies with the aim of further validating the efficacy of the treatment and exploring feasible pathways to lower the application.

Limitations

Limitations of this review

This review has several limitations that warrant consideration. First, as a narrative synthesis, the literature selection process lacked systematic search protocols (e.g., PRISMA guidelines), potentially introducing selection bias. Second, the methodological quality of included studies was not formally assessed using tools such as QUADAS-2, limiting the evaluation of evidence reliability. Importantly, most studies have failed to adjust for key confounders (e.g., age, smoking status, heterogeneity of CT scanners) as well as relevant influences on manual interpretation (68-70), which may affect the validity of diagnostic performance indicators. Compared to similar literature, our study is more confined to the application of spectral CT and AI in the diagnosis of pulmonary nodules, whereas related literature focuses on the discussion of deep learning for lung cancer. This also represents one of our limitations.

Limitations of the study

Additionally, only 33% of studies validated findings against histopathological standards, raising concerns about diagnostic accuracy. All AI models were trained and tested within single-institution cohorts, with no external validation to ensure generalizability across diverse populations. Notably, in tuberculosis-endemic regions, a lot of AI models misclassified inflammatory lesions as malignant due to insufficient training on infectious etiologies.

Technical limitations include the absence of radiation dose comparisons between spectral and conventional CT, as well as insufficient data on long-term clinical outcomes—only two studies reported ≥12-month patient follow-up. Furthermore, correlations between imaging features (71-73) and actionable biomarkers (e.g., EGFR/Ki-67) remain quantitatively undefined, hindering clinical translation. These gaps underscore the need for multicenter trials with standardized protocols to establish robust evidence for clinical adoption.

We believe that the black box problem can be solved by enhancing the interpretability of the AI model, which can enhance doctors’ trust in the AI through interpretable technology, and also assist teaching and multidisciplinary consultation, and alleviating black box anxiety in the future is an important aspect of the technology towards the clinic (74-79) addition, the high acquisition cost of spectral CT equipment and the arithmetic requirements of AI algorithms may limit its popularity in primary care. However, imaging physicians need to receive interdisciplinary training to understand the quantitative parameters (e.g, λHU, NIC) of AI output and their clinical significance. Low-cost AI deployment options (e.g., cloud-based analysis) with standardized training systems need to be explored in the future.

Conclusions

In conclusion, the integration of spectral CT and AI demonstrates preliminary potential for evaluating lung nodules across their lifecycle, particularly in distinguishing benign versus malignant lesions, identifying pathological subtypes, and assessing disease progression. However, the current evidence remains insufficient to establish robust clinical utility. Key limitations—such as small sample sizes, inconsistent pathological subtyping, lack of standardized diagnostic criteria, and limited external validation—significantly constrain the generalizability of findings. Although preliminary data suggest that AI combined with spectral CT has potential for clinical application, its diagnostic efficacy needs to be established through multicenter prospective studies, standardized pathological controls, and cross-population validation.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1246/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1246/prf

Funding: This study was supported by the Innovation and Entrepreneurship Program for College Students in Heilongjiang Province (No. S202411230046) and the Scientific Research Project of Heilongjiang Provincial Health and Health Commission (No. 20220909010622).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1246/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

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. Luo G, Zhang Y, Rumgay H, et al. Estimated worldwide variation and trends in incidence of lung cancer by histological subtype in 2022 and over time: a population-based study. Lancet Respir Med 2025;13:348-63. [Crossref] [PubMed]
2. Thanoon MA, Zulkifley MA, Mohd Zainuri MAA, et al. A Review of Deep Learning Techniques for Lung Cancer Screening and Diagnosis Based on CT Images. Diagnostics (Basel) 2023;13:2617. [Crossref] [PubMed]
3. Meyer ML, Peters S, Mok TS, et al. Lung cancer research and treatment: global perspectives and strategic calls to action. Ann Oncol 2024;35:1088-104. [Crossref] [PubMed]
4. Leiter A, Veluswamy RR, Wisnivesky JP. The global burden of lung cancer: current status and future trends. Nat Rev Clin Oncol 2023;20:624-39. [Crossref] [PubMed]
5. Smolarz B, Łukasiewicz H, Samulak D, et al. Lung Cancer-Epidemiology, Pathogenesis, Treatment and Molecular Aspect (Review of Literature). Int J Mol Sci 2025;26:2049. [Crossref] [PubMed]
6. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2020. CA Cancer J Clin 2020;70:7-30. [Crossref] [PubMed]
7. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
8. Mazzone PJ, Lam L. Evaluating the Patient With a Pulmonary Nodule: A Review. JAMA 2022;327:264-73. [Crossref] [PubMed]
9. Zhao WH, Zhang LJ, Li X, et al. Clinical and Computed Tomography Characteristics of Inflammatory Solid Pulmonary Nodules with Morphology Suggesting Malignancy. Acad Radiol 2025;32:1067-77. [Crossref] [PubMed]
10. Christensen J, Prosper AE, Wu CC, et al. ACR Lung-RADS v2022: Assessment Categories and Management Recommendations. Chest 2024;165:738-53. [Crossref] [PubMed]
11. Piskorski L, Debic M, von Stackelberg O, et al. Malignancy risk stratification for pulmonary nodules: comparing a deep learning approach to multiparametric statistical models in different disease groups. Eur Radiol 2025;35:3812-22. [Crossref] [PubMed]
12. Roberts GS, Peret A, Jonaitis EM, et al. Normative Cerebral Hemodynamics in Middle-aged and Older Adults Using 4D Flow MRI: Initial Analysis of Vascular Aging. Radiology 2023;307:e222685. [Crossref] [PubMed]
13. Zhao M, Xue G, He B, et al. Integrated multiomics signatures to optimize the accurate diagnosis of lung cancer. Nat Commun 2025;16:84. [Crossref] [PubMed]
14. McKinney SM, Sieniek M, Godbole V, et al. International evaluation of an AI system for breast cancer screening. Nature 2020;577:89-94. [Crossref] [PubMed]
15. Zhang Y, Jiang B, Zhang L, et al. Lung Nodule Detectability of Artificial Intelligence-assisted CT Image Reading in Lung Cancer Screening. Curr Med Imaging 2022;18:327-34. [Crossref] [PubMed]
16. Zhang L, Shao Y, Chen G, et al. An artificial intelligence-assisted diagnostic system for the prediction of benignity and malignancy of pulmonary nodules and its practical value for patients with different clinical characteristics. Front Med (Lausanne) 2023;10:1286433. [Crossref] [PubMed]
17. Wan YL, Wu PW, Huang PC, et al. The Use of Artificial Intelligence in the Differentiation of Malignant and Benign Lung Nodules on Computed Tomograms Proven by Surgical Pathology. Cancers (Basel) 2020;12:2211. [Crossref] [PubMed]
18. Hong Y, Zhong L, Lv X, et al. Application of spectral CT in diagnosis, classification and prognostic monitoring of gastrointestinal cancers: progress, limitations and prospects. Front Mol Biosci 2023;10:1284549. [Crossref] [PubMed]
19. Li M, Zheng X, Gao F, et al. Spectral CT imaging of intranodular hemorrhage in cases with challenging benign thyroid nodules. Radiol Med 2016;121:279-90. [Crossref] [PubMed]
20. Wu F, Zhou H, Li F, et al. Spectral CT Imaging of Lung Cancer: Quantitative Analysis of Spectral Parameters and Their Correlation with Tumor Characteristics. Acad Radiol 2018;25:1398-404. [Crossref] [PubMed]
21. Li X, Tang P, Liang F, et al. Machine learning based multi-label classification of single or mixed-composition urinary stones in in vivo spectral CT. Med Phys 2023;50:661-74. [Crossref] [PubMed]
22. Qu M, Ramirez-Giraldo JC, Leng S, et al. Dual-energy dual-source CT with additional spectral filtration can improve the differentiation of non-uric acid renal stones: an ex vivo phantom study. AJR Am J Roentgenol 2011;196:1279-87. [Crossref] [PubMed]
23. Schicchi N, Fogante M, Esposto Pirani P, et al. Third-generation dual-source dual-energy CT in pediatric congenital heart disease patients: state-of-the-art. Radiol Med 2019;124:1238-52. [Crossref] [PubMed]
24. Wang Y, Chen H, Chen Y, et al. A semiautomated radiomics model based on multimodal dual-layer spectral CT for preoperative discrimination of the invasiveness of pulmonary ground-glass nodules. J Thorac Dis 2023;15:2505-16. [Crossref] [PubMed]
25. Yu J, Zhong Y, Wang Y, et al. The value of dual-energy spectral CT in differentiating the pathological grades of T1-size lung adenocarcinoma. J Thorac Dis 2025;17:4858-71. [Crossref] [PubMed]
26. Kawai T, Shibamoto Y, Hara M, et al. Can dual-energy CT evaluate contrast enhancement of ground-glass attenuation? Phantom and preliminary clinical studies. Acad Radiol 2011;18:682-9. [Crossref] [PubMed]
27. Kang S, Zhang H, Lei L, et al. Dual-layer spectral detector CT electron density imaging: impressive technology for imaging characteristics of ground-glass nodules. Quant Imaging Med Surg 2025;15:7441-52. [Crossref] [PubMed]
28. Wei Y, Zhou Q, Wu J, et al. Review of Artificial Intelligence in Lung Nodule Risk Assessment. IEEE Reviews in Biomedical Engineering 2025;1-16. [Crossref] [PubMed]
29. Liu W, Wu Y, Zheng Z, et al. Enhancing Diagnostic Accuracy of Lung Nodules in Chest Computed Tomography Using Artificial Intelligence: Retrospective Analysis. J Med Internet Res 2025;27:e64649. [Crossref] [PubMed]
30. Son JY, Lee HY, Kim JH, et al. Quantitative CT analysis of pulmonary ground-glass opacity nodules for distinguishing invasive adenocarcinoma from non-invasive or minimally invasive adenocarcinoma: the added value of using iodine mapping. Eur Radiol 2016;26:43-54. [Crossref] [PubMed]
31. Li B, Chen Q, Peng G, et al. Segmentation of pulmonary nodules using adaptive local region energy with probability density function-based similarity distance and multi-features clustering. Biomed Eng Online 2016;15:49. [Crossref] [PubMed]
32. Lin CY, Guo SM, Lien JJ, et al. Combined model integrating deep learning, radiomics, and clinical data to classify lung nodules at chest CT. Radiol Med 2024;129:56-69. [Crossref] [PubMed]
33. Zheng H, Chen W, Liu J, et al. Predicting High-Grade Patterns in Stage I Solid Lung Adenocarcinoma: A Study of 371 Patients Using Refined Radiomics and Deep Learning-Guided CatBoost Classifier. Technol Cancer Res Treat 2024;23:15330338241308610. [Crossref] [PubMed]
34. Ardila D, Kiraly AP, Bharadwaj S, et al. End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med 2019;25:954-61. [Crossref] [PubMed]
35. Ziegelmayer S, Marka AW, Strenzke M, et al. Speed and efficiency: evaluating pulmonary nodule detection with AI-enhanced 3D gradient echo imaging. Eur Radiol 2025;35:2237-44. [Crossref] [PubMed]
36. Nam JG, Hwang EJ, Kim J, et al. AI Improves Nodule Detection on Chest Radiographs in a Health Screening Population: A Randomized Controlled Trial. Radiology 2023;307:e221894. [Crossref] [PubMed]
37. Kotoulas SC, Spyratos D, Porpodis K, et al. A Thorough Review of the Clinical Applications of Artificial Intelligence in Lung Cancer. Cancers (Basel) 2025;17:882. [Crossref] [PubMed]
38. Wang J, Dobbins JT 3rd, Li Q. Automated lung segmentation in digital chest tomosynthesis. Med Phys 2012;39:732-41. [Crossref] [PubMed]
39. Liu D, Zhao Y, Liu B. The effectiveness of deep learning model in differentiating benign and malignant pulmonary nodules on spiral CT. Technol Health Care 2024;32:5129-40. [Crossref] [PubMed]
40. Zhang Y, Zhang F, Shen C, et al. Multi-omics model is an effective means to diagnose benign and malignant pulmonary nodules. Clinics (Sao Paulo) 2025;80:100599. [Crossref] [PubMed]
41. Li Y, Zheng H, Huang X, et al. Research on lung nodule recognition algorithm based on deep feature fusion and MKL-SVM-IPSO. Sci Rep 2022;12:17403. [Crossref] [PubMed]
42. Gu Y, Chi J, Liu J, et al. A survey of computer-aided diagnosis of lung nodules from CT scans using deep learning. Comput Biol Med 2021;137:104806. [Crossref] [PubMed]
43. Li X, Wei Q, Wang T, et al. Narrative review of the application of artificial intelligence-related technologies in the diagnosis of pulmonary nodules with recommendations for clinical practice and future research. J Thorac Dis 2025;17:6326-38. [Crossref] [PubMed]
44. Liang H, Hu M, Ma Y, et al. Performance of Deep-Learning Solutions on Lung Nodule Malignancy Classification: A Systematic Review. Life (Basel) 2023;13:1911. [Crossref] [PubMed]
45. Huang Z, Hu C, Chi C, et al. An Artificial Intelligence Model for Predicting 1-Year Survival of Bone Metastases in Non-Small-Cell Lung Cancer Patients Based on XGBoost Algorithm. Biomed Res Int 2020;2020:3462363. [Crossref] [PubMed]
46. Zhou T, Zhu P, Xia K, et al. A Predictive Model Integrating AI Recognition Technology and Biomarkers for Lung Nodule Assessment. Thorac Cardiovasc Surg 2025;73:174-81. [Crossref] [PubMed]
47. Yang D, Miao Y, Liu C, et al. Advances in artificial intelligence applications in the field of lung cancer. Front Oncol 2024;14:1449068. [Crossref] [PubMed]
48. Liao RQ, Li AW, Yan HH, et al. Deep learning-based growth prediction for sub-solid pulmonary nodules on CT images. Front Oncol 2022;12:1002953. [Crossref] [PubMed]
49. Mao Y, Xu N, Wu Y, et al. Assessments of lung nodules by an artificial intelligence chatbot using longitudinal CT images. Cell Rep Med 2025;6:101988. [Crossref] [PubMed]
50. Shao X, Ge X, Gao J, et al. Transfer learning-based PET/CT three-dimensional convolutional neural network fusion of image and clinical information for prediction of EGFR mutation in lung adenocarcinoma. BMC Med Imaging 2024;24:54. [Crossref] [PubMed]
51. Zhong Y, She Y, Deng J, et al. Deep Learning for Prediction of N2 Metastasis and Survival for Clinical Stage I Non-Small Cell Lung Cancer. Radiology 2022;302:200-11. [Crossref] [PubMed]
52. Moldaver D, Hurry M, Evans WK, et al. Development, validation and results from the impact of treatment evolution in non-small cell lung cancer (iTEN) model. Lung Cancer 2020;139:185-94. [Crossref] [PubMed]
53. Nadler E, Espirito JL, Pavilack M, et al. Treatment Patterns and Clinical Outcomes Among Metastatic Non-Small-Cell Lung Cancer Patients Treated in the Community Practice Setting. Clin Lung Cancer 2018;19:360-70. [Crossref] [PubMed]
54. Shi J, Schmid-Bindert G, Fink C, et al. Dynamic volume perfusion CT in patients with lung cancer: baseline perfusion characteristics of different histological subtypes. Eur J Radiol 2013;82:e894-900. [Crossref] [PubMed]
55. Chen ML, Li XT, Wei YY, et al. Can spectral computed tomography imaging improve the differentiation between malignant and benign pulmonary lesions manifesting as solitary pure ground glass, mixed ground glass, and solid nodules?. Thorac Cancer 2019;10:234-42. [Crossref] [PubMed]
56. Li Q, Li X, Li XY, et al. Spectral CT in Lung Cancer: Usefulness of Iodine Concentration for Evaluation of Tumor Angiogenesis and Prognosis. AJR Am J Roentgenol 2020;215:595-602. [Crossref] [PubMed]
57. Li X, Meng X, Ye Z. Iodine quantification to characterize primary lesions, metastatic and non-metastatic lymph nodes in lung cancers by dual energy computed tomography: An initial experience. Eur J Radiol 2016;85:1219-23. [Crossref] [PubMed]
58. Sudarski S, Hagelstein C, Weis M, et al. Dual-energy snap-shot perfusion CT in suspect pulmonary nodules and masses and for lung cancer staging. Eur J Radiol 2015;84:2393-400. [Crossref] [PubMed]
59. Zhang Z, Zou H, Yuan A, et al. A Single Enhanced Dual-Energy CT Scan May Distinguish Lung Squamous Cell Carcinoma From Adenocarcinoma During the Venous phase. Acad Radiol 2020;27:624-9. [Crossref] [PubMed]
60. Lin L, Cheng J, Tang D, et al. The associations among quantitative spectral CT parameters, Ki-67 expression levels and EGFR mutation status in NSCLC. Sci Rep 2020;10:3436. [Crossref] [PubMed]
61. Fan Z, Yue Y, Lu X, et al. Predicting the Invasiveness of Mixed Ground-Glass Nodules Based on Spectral Computed Tomography-Derived Parameters and Tumor Abnormal Protein Levels: Development and Validation of a Model. Acad Radiol 2025;32:2990-3005. [Crossref] [PubMed]
62. Werner S, Krauss B, Horger M. Dual-energy CT based monitoring of treatment-induced bone marrow changes in lung cancer patients: preliminary results. Quant Imaging Med Surg 2022;12:1871-81. [Crossref] [PubMed]
63. Kaup M, Scholtz JE, Engler A, et al. Dual-Energy Computed Tomography Virtual Monoenergetic Imaging of Lung Cancer: Assessment of Optimal Energy Levels. J Comput Assist Tomogr 2016;40:80-5. [Crossref] [PubMed]
64. Fervers P, Fervers F, Weisthoff M, et al. Dual-Energy CT, Virtual Non-Calcium Bone Marrow Imaging of the Spine: An AI-Assisted, Volumetric Evaluation of a Reference Cohort with 500 CT Scans. Diagnostics (Basel) 2022;12:671. [Crossref] [PubMed]
65. Wang YW, Chen CJ, Wang TC, et al. Multi-energy level fusion for nodal metastasis classification of primary lung tumor on dual energy CT using deep learning. Comput Biol Med 2022;141:105185. [Crossref] [PubMed]
66. Zhu X, Zhu L, Song D, et al. Comparison of single- and dual-energy CT combined with artificial intelligence for the diagnosis of pulmonary nodules. Clin Radiol 2023;78:e99-105. [Crossref] [PubMed]
67. Wang T, Fan Z, Yue Y, et al. Spectral dual-layer detector CT-based radiomics-deep learning for predicting pathological aggressiveness of stage I lung adenocarcinoma: discrimination of precursor glandular lesions and invasive adenocarcinomas. Transl Lung Cancer Res 2025;14:431-48. [Crossref] [PubMed]
68. Yoon SH, Kim YJ, Doh K, et al. Interobserver variability in Lung CT Screening Reporting and Data System categorisation in subsolid nodule-enriched lung cancer screening CTs. Eur Radiol 2021;31:7184-91. [Crossref] [PubMed]
69. Koike S, Shimizu K, Ide S, et al. Is using a consolidation tumor ratio 0.5 as criterion feasible in daily practice? Evaluation of interobserver measurement variability of consolidation tumor ratio of lung cancer less than 3 cm in size. Thorac Cancer 2022;13:3018-24. [Crossref] [PubMed]
70. Bolte H, Jahnke T, Schäfer FK, et al. Interobserver-variability of lung nodule volumetry considering different segmentation algorithms and observer training levels. Eur J Radiol 2007;64:285-95. [Crossref] [PubMed]
71. Qin ZZ, Ahmed S, Sarker MS, et al. Tuberculosis detection from chest x-rays for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence algorithms. Lancet Digit Health 2021;3:e543-54. [Crossref] [PubMed]
72. Qin ZZ, Sander MS, Rai B, et al. Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems. Sci Rep 2019;9:15000. [Crossref] [PubMed]
73. Nxumalo ZZ, Irusen EM, Allwood BW, et al. The utility of artificial intelligence in identifying radiological evidence of lung cancer and pulmonary tuberculosis in a high-burden tuberculosis setting. S Afr Med J 2024;114:e1846. [Crossref] [PubMed]
74. Poon AIF, Sung JJY. Opening the black box of AI-Medicine. J Gastroenterol Hepatol 2021;36:581-4. [Crossref] [PubMed]
75. Chan B. Black-box assisted medical decisions: AI power vs. ethical physician care. Med Health Care Philos 2023;26:285-92. [Crossref] [PubMed]
76. Karim MR, Islam T, Shajalal M, et al. Explainable AI for Bioinformatics: Methods, Tools and Applications. Brief Bioinform 2023;24:bbad236. [Crossref] [PubMed]
77. Najjar R. Redefining Radiology: A Review of Artificial Intelligence Integration in Medical Imaging. Diagnostics (Basel) 2023;13:2760. [Crossref] [PubMed]
78. De-Giorgio F, Benedetti B, Mancino M, et al. The need for balancing ‘black box’ systems and explainable artificial intelligence: A necessary implementation in radiology. Eur J Radiol 2025;185:112014. [Crossref] [PubMed]
79. Saw SN, Yan YY, Ng KH. Current status and future directions of explainable artificial intelligence in medical imaging. Eur J Radiol 2025;183:111884. [Crossref] [PubMed]