Multicenter Prospective Validation of AI Models for Malignancy Risk Prediction in Pulmonary Nodules

A Multicenter Prospective Diagnostic Accuracy Study of Three CT-Based Artificial Intelligence Models for Predicting Malignancy Risk in Pulmonary Nodules Using Pathology as the Gold Standard

This multicenter prospective diagnostic accuracy study will compare the performance of three artificial intelligence (AI) models (MVCS, LungDoc, and a United Imaging AI model) for predicting the malignancy risk of pulmonary nodules on chest CT. All enrolled patients will have pulmonary nodules ≤3 cm on CT and a definitive postoperative or biopsy pathological diagnosis. The AI models will generate continuous malignancy probability scores based only on CT images. Pathology will serve as the gold standard.

The primary objective is to compare the area under the receiver operating characteristic curve (AUC) for malignancy prediction among the three AI models. Secondary objectives include comparison of sensitivity, specificity, positive and negative predictive values, accuracy, F1 score, and calibration. Exploratory analyses will evaluate the MVCS model for predicting pathological invasion degree (pre-invasive, minimally invasive, and invasive adenocarcinoma) and an extended MVCSN model that incorporates clinical and imaging features in a data-complete subset.

Study Overview

• Status: Not yet recruiting
• Conditions: Pulmonary Nodules

Detailed Description

Lung cancer is the leading cause of cancer-related morbidity and mortality worldwide. Low-dose CT (LDCT) screening improves early detection but generates a high prevalence of indeterminate pulmonary nodules and substantial false positives, leading to unnecessary invasive procedures and anxiety while still risking missed early cancers. Traditional radiologic assessment of pulmonary nodules relies on visual features and clinical risk factors, with limited performance and substantial reader variability, especially for small or subsolid nodules.

Recent advances in deep learning enable AI models to extract high-dimensional imaging features from CT scans and to predict nodule malignancy and risk stratification. Several commercial AI systems for pulmonary nodule assessment have been approved and deployed in clinical practice, and academic groups have proposed novel algorithms such as the multi-view coupled self-attention (MVCS) model. However, most prior studies have been single-center, retrospective, and used reference standards such as imaging follow-up or expert reading rather than biopsy pathology. Head-to-head comparisons of different AI models in prospective, multicenter real-world populations with pathological gold standard are lacking.

This study is a multicenter, prospective, diagnostic accuracy comparison of three purely imaging-based AI models-MVCS, LungDoc (Shukun Technology), and a United Imaging AI model-for predicting the malignancy of pulmonary nodules. Eligible patients are adults (≥18 years) with at least one pulmonary nodule ≤3 cm on CT, who undergo surgical or biopsy pathology with a definitive benign or malignant diagnosis, and with a CT-pathology interval ≤6 months. CT images in DICOM format will be collected using standardized acquisition parameters across centers and processed by the three AI models, which output continuous malignancy probabilities or suspicion scores. Investigators will be blinded to AI outputs.

The primary endpoint is the AUC for malignancy prediction for each model, and pairwise AUC comparisons using DeLong's test. Secondary endpoints include binary performance metrics (sensitivity, specificity, PPV, NPV, accuracy, F1 score) at model-native thresholds and optimal Youden index thresholds, as well as calibration (calibration curves, intercept, slope). Prespecified subgroup analyses will examine performance by age, sex, smoking status, nodule size, morphology, and study center, and random-effects methods will be used to assess center effects.

An exploratory aim will validate the MVCS model for predicting pathological invasion degree by classifying nodules into pre-invasive lesions (atypical adenomatous hyperplasia [AAH] / adenocarcinoma in situ [AIS]), minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (ADC), using metrics such as multi-class accuracy, weighted F1 score, confusion matrix, Matthews correlation coefficient, and AUC for predefined binary sub-tasks. Another exploratory analysis will evaluate the MVCSN model, which incorporates CT images plus clinical and radiologic features, in a subset with complete data.

The study plans to enroll 3,000 pathologically confirmed pulmonary nodules across five centers in China over approximately 30 months. No experimental treatment is administered; all clinical management, imaging, and pathology follow standard of care. Risks are limited to those associated with clinically indicated pathology procedures (e.g., surgery or biopsy).

• Study Type: Observational
• Enrollment (Estimated): 3000

Contacts and Locations

• Study Contact: Name: Yijing Feng, PhD; Phone Number: 8613650882360; Email: yfeng@g.harvard.edu

Study Locations

• China
  • Chongqing Municipality
    • Chongqing, Chongqing Municipality, China, 400037
      • Second Affiliated Hospital of Army Medical University (Xinqiao Hospital)
      • Contact: Jigang Dai; Phone Number: 86 23 6875 5114; Email: cqdaijigangzhushou@163.com
  • Guangdong
    • Guangzhou, Guangdong, China, 510000
      • Guangdong Provincial People's Hospital
      • Contact: Yijing Feng, PhD; Phone Number: 8613650882360; Email: yfeng@g.harvard.edu
      • Principal Investigator: Wenzhao Zhong, PhD
    • Guangzhou, Guangdong, China, 510000
      • Zhujiang Hospital, Southern Medical University
      • Contact: Guibin Qiao; Phone Number: 86 20 61643888; Email: qiaoguibin@smu.edu.cn
  • Jiangsu
    • Xuzhou, Jiangsu, China, 221006
      • Affiliated Hospital of Xuzhou Medical University
      • Contact: Hao Zhang; Phone Number: 0516 8560 9999; Email: haozhang_xz@163.com
  • Zhejiang
    • Hangzhou, Zhejiang, China, 310003
      • Zhejiang University
      • Contact: Junqiang Fan; Phone Number: 86 57187951111; Email: zrxwk@zju.edu.cn

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

Adults (≥18 years) undergoing routine clinical care at five tertiary hospitals in China who have at least one pulmonary nodule (≤3 cm) detected on chest CT and receive surgical or biopsy pathology with a definitive benign or malignant diagnosis. All participants have adequate CT image quality and complete clinicopathologic information for AI model validation

Description

Inclusion Criteria:

• Age ≥ 18 years, any sex.
• At least one pulmonary nodule detected on chest CT, with initial nodule diameter ≤ 3 cm.
• The nodule undergoes surgical resection or biopsy with a definitive benign or malignant pathological diagnosis.
• Time interval between CT examination and pathological examination ≤ 6 months.
• Availability of complete CT imaging data in DICOM format with adequate image quality (no severe artifacts), meeting input requirements of all three AI models.

Availability of complete clinicopathologic information including histologic type and grade, with clear pathological diagnosis suitable as gold standard labels for AI validation.

-The patient (or legally authorized representative) is willing and able to sign written informed consent.

Exclusion Criteria:

• Pathological results are unclear, inconclusive, or disputed; nodule nature or grade cannot be reliably determined.
• The patient receives treatments between CT and pathology that may significantly alter nodule appearance (e.g., chemotherapy, radiotherapy, targeted therapy).
• CT imaging data are incomplete (missing essential series) or have severe motion, metal, or other artifacts preventing accurate AI analysis.
• Required metadata for any AI model are missing and cannot be imputed. History of other malignant tumors (malignancies other than the index non-small cell lung cancer).
• Severe psychiatric illness, cognitive impairment, or other conditions that prevent cooperation with study-related procedures and follow-up.

Participation in another clinical study that may interfere with the results of this research.

-The patient or legal representative refuses participation.

Exclusion (Post-Enrollment / Removal from Analysis)

Participants already enrolled may be excluded from the analysis set if:

• They are later found not to meet inclusion criteria or to meet exclusion criteria.
• No usable data are available after enrollment.
• Required AI model assessments are not completed (e.g., technical failure to generate outputs).
• Critical data are missing, preventing contribution to primary analysis.
• The interval between CT and pathology exceeds 6 months.

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort
Single group; This study has only one group.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Area Under the ROC Curve (AUC) for Malignancy Prediction	For each pure imaging AI model (MVCS, LungDoc, United Imaging model), the AUC of the receiver operating characteristic curve for predicting malignant versus benign pulmonary nodules, based on continuous malignancy probabilities or suspicion scores. AUCs will be reported with 95% confidence intervals, and pairwise comparisons will be conducted using DeLong's test.	At the time of availability of pathology results, up to 6 months after index chest CT

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity and Specificity for Malignancy Prediction	Sensitivity and specificity for classifying nodules as malignant vs benign for each AI model, using both (a) model-predefined thresholds and (b) optimal cut-off points determined by maximizing the Youden index. 95% confidence intervals will be reported; paired comparisons will use McNemar's test.	At the time of availability of pathology results, up to 6 months after index chest CT
Positive Predictive Value (PPV) and Negative Predictive Value (NPV)	PPV and NPV for malignancy prediction for each AI model at the same thresholds as above, with 95% confidence intervals.	At the time of availability of pathology results, up to 6 months after index chest CT
Overall Diagnostic Accuracy and F1 Score	Proportion of correctly classified nodules (accuracy) and F1 score for each AI model in the binary task of malignant versus benign nodules, with 95% confidence intervals.	At the time of availability of pathology results, up to 6 months after index chest CT
Calibration Metrics	Calibration performance of each AI model will be evaluated by calibration plots, calibration intercept, and calibration slope for predicted malignancy probability versus observed malignant proportion. Hosmer-Lemeshow goodness-of-fit test will be reported.	At the time of availability of pathology results, up to 6 months after index chest CT

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Multi-Class Accuracy of MVCS for Pathological Invasion Degree	For nodules with specific pathological subtypes (AAH, AIS, MIA, ADC), performance of the MVCS model in three-class classification: pre-invasive lesions (AAH-AIS), minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (ADC). Metrics include overall three-class accuracy and weighted F1 score.	At the time of availability of pathology results, up to 6 months after index chest CT
Matthews Correlation Coefficient (MCC) and Confusion Matrix for Invasion Classification		At the time of availability of pathology results, up to 6 months after index chest CT
Exploratory Performance of MVCSN Model for Malignancy Prediction	In a subset with complete clinical and imaging feature data, the MVCSN model (CT + clinical + imaging features) will be evaluated for malignancy prediction using similar metrics (AUC, sensitivity, specificity, PPV, NPV, accuracy, F1, calibration). Results will be used exploratorily and not for primary hypothesis testing.	At the time of availability of pathology results, up to 6 months after index chest CT

Collaborators and Investigators

• Sponsor: Guangdong Provincial People's Hospital

Study record dates

Study Major Dates

• Study Start (Estimated): July 1, 2026
• Primary Completion (Estimated): March 1, 2028
• Study Completion (Estimated): December 1, 2028

Study Registration Dates

• First Submitted: July 18, 2026
• First Submitted That Met QC Criteria: July 22, 2026
• First Posted (Actual): July 27, 2026

Study Record Updates

• Last Update Posted (Actual): July 27, 2026
• Last Update Submitted That Met QC Criteria: July 22, 2026
• Last Verified: July 1, 2026

More Information

Terms related to this study

• Keywords: Artificial intelligence; Diagnostic accuracy; Deep learning; Pulmonary nodule
• Additional Relevant MeSH Terms: Neoplasms by Site; Neoplasms; Respiratory Tract Diseases; Lung Diseases; Respiratory Tract Neoplasms; Thoracic Neoplasms; Lung Neoplasms; Multiple Pulmonary Nodules
• Other Study ID Numbers: KY2026-585-02; 2024ZD0529400 (Other Grant/Funding Number: National Science and Technology Major Project of China)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: UNDECIDED

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No