AI-driven breath biopsy from a case-control study assists in the early detection of paediatric brain tumours.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
BACKGROUND: Paediatric brain tumours (PBT) are among the deadliest childhood cancers, with delayed diagnosis often limiting early intervention. This study explores exhaled volatile metabolites as potential biomarkers for PBT, employing AI to identify diagnostic markers and develop a predictive risk assessment model. METHODS: We conducted a case-control study using untargeted volatile metabolomics on exhaled breath samples from 161 patients with primary paediatric brain tumours (PBT) and 140 non-tumour controls, analysed via a custom-built analysis platform. Machine learning, combined with univariate analysis, identified volatile organic compound (VOC) biomarkers linked to PBT. Diagnostic performance was evaluated using four supervised learning classifiers: Support Vector Machine (SVM), Naive Bayes (NB), Logistic Regression (LR), and Random Forest (RF), and model robustness was validated in an independent internal cohort of 32 participants. Incorporating PBMCs transcriptome for multi-omics analysis to construct a gene-VOC interaction network for investigating the biological origins of VOC biomarkers. An ensemble learning model was developed to enhance risk assessment by integrating clinical indicators. FINDINGS: Significant differences in exhaled volatile metabolomics were observed between PBT patients and controls. A panel of 12 VOC biomarkers was identified, achieving a best performed AUC of 0.81 (95% CI: 0.72-0.91) on SVM classifier, with all AUC >0.70 for all classifiers in the discovery cohort, highlighting strong diagnostic potential. Multi-omics analysis revealed VOC alterations linked to immune dysregulation, with indole and NOX components NCF1/2 emerging as key regulatory factors. An AI-assisted risk assessment model, incorporating immune-inflammatory clinical indicators, achieved high sensitivity (0.90, 95% CI: 0.77-0.90), specificity (0.86, 95% CI: 0.76-0.90), and accuracy (0.85, 95% CI: 0.80-0.89), outperforming traditional diagnostic models by nearly 20%. INTERPRETATION: This study underscores the potential of AI-driven analysis of exhaled volatile metabolites for identifying biomarkers of paediatric brain tumours. The VOC biomarker panel, integrated with clinical indicators, offers a valuable biomarker resource for non-invasive early diagnosis and risk stratification in PBT, highlighting the potential of breath biopsy as a promising clinical tool. FUNDING: This work was supported by the National Natural Science Foundation of China (No. 22476023 and No. 22276038), the Fundamental Research Funds for the Central Universities (No. KLSB2023KF-06), AI for Science Foundation of Fudan University (No. FudanX24Al026), Agilent Research Gift (No. 4956) and the Foundation of Xinhua Hospital (No. GD202501).