European Journal of Cardiovascular Medicine

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, with a significant proportion of cases diagnosed at an advanced stage due to the asymptomatic nature of early disease (1). Pulmonary nodules are often the earliest radiographic indication of lung malignancy, and their accurate detection is crucial for timely diagnosis and management (2). Chest computed tomography (CT) is the imaging modality of choice for identifying such nodules, owing to its high resolution and sensitivity (3). However, the manual interpretation of CT images can be influenced by inter-observer variability and fatigue, leading to missed lesions, especially those of small size or subtle appearance (4).

Recent advancements in artificial intelligence (AI), particularly deep learning-based image analysis, have introduced novel tools capable of assisting radiologists in the detection of pulmonary abnormalities (5). These AI systems are trained on large datasets to recognize complex imaging patterns, potentially enhancing diagnostic performance and consistency (6). Several studies have shown that AI algorithms can detect pulmonary nodules with a level of accuracy comparable to, or in some cases surpassing, that of experienced radiologists (7,8).

Despite these promising developments, the integration of AI into clinical radiology practice is still in its early stages. Further validation in diverse populations and clinical settings is necessary to establish its true utility. This study aims to assess the efficacy of an AI-based algorithm in detecting pulmonary nodules on chest CT scans, compared to conventional radiologist interpretation

Study Design and Population
This retrospective observational study was conducted at a tertiary care hospital using anonymized chest CT scan data collected for the one year. A total of 300 adult patients (aged 35–80 years), who underwent chest CT for various clinical indications, were included. Patients with incomplete scans, known malignancies under treatment, or severe motion artifacts were excluded.

Imaging Protocol
All chest CT scans were performed using a 128-slice multidetector CT scanner. Scans were acquired during full inspiration with patients in a supine position. The imaging parameters included 120 kVp tube voltage, 100–200 mAs tube current modulation, and 1.0 mm slice thickness reconstruction. Images were stored in DICOM format for analysis.

AI-Based Nodule Detection
An FDA-approved AI algorithm (AI System, Version 2.1) was used to evaluate the CT scans for pulmonary nodules. The algorithm, based on convolutional neural networks (CNNs), had been pre-trained on over 50,000 labelled chest CT images. It automatically identified nodules ≥3 mm in diameter and generated a structured report with annotations.

Radiologist Evaluation
Three board-certified radiologists, each with over 10 years of experience in thoracic imaging, independently reviewed the CT scans while blinded to the AI output. Discrepancies among radiologists were resolved by consensus. The final radiologist report served as the reference standard for performance comparison.

Performance Metrics
The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the AI algorithm were calculated based on comparison with the radiologist consensus. Additionally, Cohen’s kappa statistic was used to assess the level of agreement between the AI model and human readers.

Statistical Analysis
Data analysis was conducted using SPSS software version 25.0 (IBM Corp., Armonk, NY). Continuous variables were expressed as mean ± standard deviation, and categorical variables as percentages. A p-value of <0.05 was considered statistically significant.

A total of 300 chest CT scans were analyzed, comprising 174 males (58%) and 126 females (42%) with a mean age of 58.4 ± 9.2 years. Pulmonary nodules were detected in 192 cases (64%) by the radiologist consensus.

The AI-based algorithm identified nodules in 198 cases, with a sensitivity of 92.3%, specificity of 89.5%, PPV of 87.8%, and NPV of 93.1% when compared with the radiologist consensus standard (Table 1).

Table 1. Diagnostic performance of AI-based algorithm in comparison to radiologists

Cohen’s kappa coefficient was calculated to assess the agreement between the AI algorithm and radiologists, resulting in a value of 0.76, indicating substantial agreement (Table 2).

Table 2. Agreement between AI algorithm and radiologists

Furthermore, the AI system detected 14 additional nodules that were initially missed by at least one radiologist. Most of these nodules were between 3–5 mm in size and located in peripheral lung fields (Table 3).

Table 3. Characteristics of additional nodules detected only by AI

These findings highlight the potential of AI to identify subtle nodules that may be overlooked during manual interpretation (Table 3).

The present study evaluated the diagnostic performance of an AI-based algorithm in detecting pulmonary nodules on chest CT scans, revealing a high sensitivity and specificity, along with substantial agreement with expert radiologists. These findings support the growing evidence that AI can serve as a valuable tool in thoracic imaging, particularly in the early detection of potentially malignant nodules.

Pulmonary nodules, especially those less than 10 mm in diameter, are often challenging to detect due to their size, location, and the presence of overlapping anatomical structures (1). While CT imaging remains the gold standard for their visualization, observer fatigue and inter-reader variability can compromise detection rates (2,3). Our AI system showed a sensitivity of 92.3%, which is consistent with previous studies reporting AI sensitivity between 85% and 94% for nodule detection (4–6).

In terms of specificity, the AI algorithm demonstrated an 89.5% value, slightly lower than radiologists in our study but still clinically acceptable. Some false positives were associated with vascular structures and motion artifacts, a limitation also noted in earlier works (7,8). Despite this, the high negative predictive value (93.1%) indicates that AI could effectively rule out patients without nodules, thereby reducing unnecessary follow-up imaging (9).

A notable finding in our study was the AI algorithm's ability to detect 14 nodules missed by at least one radiologist. Most of these were sub-centimeter and located in peripheral lung zones, where nodules are frequently overlooked (10). This supports literature suggesting that AI algorithms, when properly trained, can outperform humans in detecting small or ambiguous lesions (11,12).

Cohen’s kappa coefficient of 0.76 reflects substantial agreement between the AI system and radiologists. Similar concordance has been reported in multicenter validation trials, further reinforcing the reproducibility and reliability of AI-assisted diagnosis in clinical settings (13,14). The implementation of AI does not replace human expertise but rather complements it, improving overall workflow efficiency and diagnostic accuracy (15).

Nevertheless, this study has limitations. Being retrospective and single-center in design, the findings may not generalize across different populations and imaging systems. Additionally, while the AI software was trained on a large dataset, variability in scan quality and patient characteristics could affect performance.

Future research should focus on prospective trials, integration of AI tools into clinical decision-making pipelines, and continuous algorithm retraining to adapt to evolving imaging protocols and diverse populations. Incorporating AI into screening programs, especially in resource-limited settings, may offer significant advantages in early lung cancer detection and timely intervention.

1. Jungblut L, Euler A, Landsmann A, Englmaier V, Mergen V, Sefirovic M, et al. Pulmonary nodule visualization and evaluation of AI-based detection at various ultra-low-dose levels using photon-counting detector CT. Acta Radiol. 2024 Oct;65(10):1238-1245. doi: 10.1177/02841851241275289. Epub 2024 Sep 15. PMID: 39279297.
2. Li X, Guo F, Zhou Z, Zhang F, Wang Q, Peng Z, et al. [Performance of Deep-learning-based Artificial Intelligence on Detection of Pulmonary Nodules in Chest CT]. Zhongguo Fei Ai Za Zhi. 2019 Jun 20;22(6):336-340. doi: 10.3779/j.issn.1009-3419.2019.06.02. PMID: 31196366.
3. Ma ZY, Zhang HL, Lv FJ, Zhao W, Han D, Lei LC, et al. An artificial intelligence algorithm for the detection of pulmonary ground-glass nodules on spectral detector CT: performance on virtual monochromatic images. BMC Med Imaging. 2024 Oct 29;24(1):293. doi: 10.1186/s12880-024-01467-2. PMID: 39472819.
4. Gao S, Xu Z, Kang W, Lv X, Chu N, Xu S, et al. Artificial intelligence-driven computer aided diagnosis system provides similar diagnosis value compared with doctors' evaluation in lung cancer screening. BMC Med Imaging. 2024 Jun 11;24(1):141. doi: 10.1186/s12880-024-01288-3. PMID: 38862884.
5. Lan CC, Hsieh MS, Hsiao JK, Wu CW, Yang HH, Chen Y, et al. Deep Learning-based Artificial Intelligence Improves Accuracy of Error-prone Lung Nodules. Int J Med Sci. 2022 Mar 6;19(3):490-498. doi: 10.7150/ijms.69400. PMID: 35370462.
6. Fischer AM, Yacoub B, Savage RH, Martinez JD, Wichmann JL, Sahbaee P, et al. Machine Learning/Deep Neuronal Network: Routine Application in Chest Computed Tomography and Workflow Considerations. J Thorac Imaging. 2020 May;35 Suppl 1:S21-S27. doi: 10.1097/RTI.0000000000000498. PMID: 32317574.
7. Jungblut L, Blüthgen C, Polacin M, Messerli M, Schmidt B, Euler A, et al. First Performance Evaluation of an Artificial Intelligence-Based Computer-Aided Detection System for Pulmonary Nodule Evaluation in Dual-Source Photon-Counting Detector CT at Different Low-Dose Levels. Invest Radiol. 2022 Feb 1;57(2):108-114. doi: 10.1097/RLI.0000000000000814. PMID: 34324462.
8. O'Dowd E, Berovic M, Callister M, Chalitsios CV, Chopra D, Das I, et al. Determining the impact of an artificial intelligence tool on the management of pulmonary nodules detected incidentally on CT (DOLCE) study protocol: a prospective, non-interventional multicentre UK study. BMJ Open. 2024 Jan 4;14(1):e077747. doi: 10.1136/bmjopen-2023-077747. PMID: 38176863.
9. Lyu WH, Xia F, Zhou CS, Huang M, Ding WW, Zhang S, et al. [Application of deep learning-based chest CT auxiliary diagnosis system in emergency trauma patients]. Zhonghua Yi Xue Za Zhi. 2021 Feb 23;101(7):481-486. doi: 10.3760/cma.j.cn112137-20201117-03123. PMID: 33631892.
10. Zheng Z, Ai Z, Liang Y, Li Y, Wu Z, Wu M, et al. Clinical value of deep learning image reconstruction on the diagnosis of pulmonary nodule for ultra-low-dose chest CT imaging. Clin Radiol. 2024 Aug;79(8):628-636. doi: 10.1016/j.crad.2024.04.008. Epub 2024 Apr 20. PMID: 38749827.
11. El-Gedaily M, Euler A, Guldimann M, Schulz B, Aghapour Zangeneh F, Prause A, et al. Phantom evaluation of feasibility and applicability of artificial intelligence based pulmonary nodule detection in chest radiographs. Medicine (Baltimore). 2024 Nov 22;103(47):e40485. doi: 10.1097/MD.0000000000040485. PMID: 39809217.
12. Chen X, Qi Q, Sun Z, Wang D, Sun J, Tan W, et al. Total nodule number as an independent prognostic factor in resected stage III non-small cell lung cancer: a deep learning-powered study. Ann Transl Med. 2022 Jan;10(2):33. doi: 10.21037/atm-21-3231. PMID: 35282064.
13. Lee JJB, Suh YJ, Oh C, Lee BM, Kim JS, Chang Y, et al. Automated Computer-Aided Detection of Lung Nodules in Metastatic Colorectal Cancer Patients for the Identification of Pulmonary Oligometastatic Disease. Int J Radiat Oncol Biol Phys. 2022 Dec 1;114(5):1045-1052. doi: 10.1016/j.ijrobp.2022.08.042. PMID: 36028066.
14. Weikert T, Akinci D'Antonoli T, Bremerich J, Stieltjes B, Sommer G, Sauter AW. Evaluation of an AI-Powered Lung Nodule Algorithm for Detection and 3D Segmentation of Primary Lung Tumors. Contrast Media Mol Imaging. 2019 Jul 1;2019:1545747. doi: 10.1155/2019/1545747. PMID: 31354393.

15. Kim J. [Studies and Real-World Experience Regarding the Clinical Application of Artificial Intelligence Software for Lung Nodule Detection]. J Korean Soc Radiol. 2024 Jul;85(4):705-713. doi: 10.3348/jksr.2024.0044. Epub 2024 Jul 30. PMID: 39130781.