# Diagnostic value of combined CT artificial intelligence (AI) system and lung cancer biomarkers in pulmonary nodule evaluation

## TL;DR

Combining CT AI with lung cancer biomarkers improves the accuracy of diagnosing pulmonary nodules compared to using AI alone.

## Contribution

Demonstrates that combining CT AI with biomarkers significantly enhances diagnostic performance for pulmonary nodules.

## Key findings

- The CT/AI system combined with biomarkers achieved 93% accuracy, compared to 80.5% with AI alone.
- Biomarkers like CEA, NSE, and SCCA showed strong predictive value for malignancy.
- Age, smoking, and chronic lung disease were significant risk factors for malignant nodules.

## Abstract

To analyze the diagnostic value of a Computed Tomography (CT) artificial intelligence (AI) system combined with lung cancer biomarkers for pulmonary nodules.

A retrospective analysis was conducted on 200 patients with pulmonary nodules treated at our hospital from February 2021 to January 2025. Based on pathological results, patients were divided into a benign group and a malignant group. The two groups were compared in terms of baseline data and lung cancer biomarkers, including carcinoembryonic antigen (CEA), neuron-specific enolase (NSE), cytokeratin 19 fragment 21–1 (CYFRA 21–1), squamous cell carcinoma antigen (SCCA), and pro-gastrin-releasing peptide (ProGRP). The sensitivity, specificity, accuracy, misdiagnosis rate, and missed diagnosis rate of the CT/AI system alone and in combination with lung cancer biomarkers were analyzed.

There were no statistically significant differences between the benign group (134 cases) and malignant group (66 cases) regarding sex, lobulation sign, spiculation sign, solitary pulmonary nodule (SPN), or mean CT value (P > 0.05). However, the benign group had significantly lower age, years of smoking, chronic lung disease, pure ground-glass nodules (pGGN), nodule diameter, irregular nodules, bronchial changes, and vascular changes compared to the malignant group (P < 0.05). Levels of CEA, NSE, CYFRA 21–1, SCCA, and ProGRP were also significantly lower in the benign group than in the malignant group (P < 0.05). Taking pathology as the reference standard, the CT/AI system alone had a sensitivity of 71.21% (47/66), specificity of 85.07% (114/134), accuracy of 80.50% (161/200), misdiagnosis rate of 19.50% (39/200), and missed diagnosis rate of 28.79% (19/66). In contrast, the CT/AI system combined with lung cancer biomarkers had a sensitivity of 92.42% (61/66), specificity of 93.28% (125/134), accuracy of 93.00% (186/200), misdiagnosis rate of 7.00% (14/200), and missed diagnosis rate of 7.58% (5/66), with all diagnostic parameters significantly improved compared with the CT/AI system alone (P < 0.05). Logistic regression analysis showed that age, smoking for >20 years, chronic lung disease, nodule diameter, irregular nodules, bronchial changes, vascular changes, NSE, CYFRA 21–1, and SCCA were all risk factors for malignant pulmonary nodules (P < 0.05). Receiver operating characteristic (ROC) curve analysis demonstrated that age, nodule type, chronic lung disease, nodule morphology, bronchial changes, and vascular changes had modest value for predicting malignant pulmonary nodules, with AUCs of 0.586, 0.750, 0.707, 0.601, 0.580, and 0.565, respectively. Smoking, nodule diameter, CEA, NSE, CYFRA 21–1, SCCA, and ProGRP had better predictive value, with AUCs of 0.840, 0.944, 0.958, 0.922, 0.856, 0.978, and 0.990, respectively. The combined diagnosis of all indicators achieved an AUC of 0.993.

The CT/AI system combined with lung cancer biomarkers demonstrates high sensitivity and specificity in diagnosing the nature of pulmonary nodules. Moreover, the occurrence of malignant pulmonary nodules is significantly associated with factors such as age, smoking, and chronic lung disease.