In response to the high false positive rate of traditional Low-Dose Computed Tomography (LDCT) in diagnosing pulmonary malignant nodules, this study aimed to investigate the effectiveness of scoring of blood-based non-invasive biological metabolite detection combined with artificial intelligent scoring of non-invasive imaging in the clinical diagnosis of Pulmonary Nodules (PNs). In this retrospective study, risk scoring was performed in patients positive for pulmonary nodules and subsequently, PNs were sampled by invasive procedures for pathological examinations. The pathological classification was used as the gold standard, and statistical and machine learning methods showed, that in 210 patients (23 benign PN and 187 malignant PN), the risk score of Metabonomics, radiomics, and multi-omics had different levels of performance in different risk groups based on various predictive models. The Area Under the receiver operating Characteristic Curve (AUC) of the multi-omics model was 0.823. The present results indicate that a multi-omics model is more effective than a single model in the non-invasive diagnosis of pulmonary malignant nodules.