AI-Driven Exhalation Biopsy Technology for the Detection of Gastrointestinal Cancer

Authors: Shangzhewen (1)*, Yongqian Liu (1), Jian Chen (1), Yongyan Ji (1), Xiang Li (1) (corresponding author)

Affiliations: (1）Department of Environmental Science & Engineering, Fudan University, Shanghai 200438, P.R. China. 

Poster Available Soon

Abstract

Gastrointestinal cancer (GI cancer), including colorectal cancer (CRC) and gastric cancer (GC), is a highly lethal cancer worldwide. Early diagnosis is critical for improving clinical outcomes by facilitating timely intervention. However, the gold standard for diagnosing GI cancer, which are endoscopic examination, collection of blood markers and CT imaging, are both invasive and costly. Consequently, there is an urgent need for non-invasive detection approaches with high sensitivity and specificity. Even though extensive research has confirmed the significant potential of exhaled VOCs as biomarkers for the non-invasive detection of cancer, the lack of standardized protocols for breath sample collection and analysis, along with the unexplored origins of these compounds, has hindered their broader clinical adoption. In response to these challenges, our study constructed a standardized sampling and analysis system for exhaled VOCs based on ReCIVA breath sampler coupled with a comprehensive mass spectrometry-based procedure, which precisely identified 82 VOCs across 351 participants (101 CRC, 73 GC and 177 controls), significantly surpassing similar research efforts. Utilizing topological data analysis (TDA), we demonstrated that exhaled VOCs could distinguish not only between patients and healthy individuals but also among various gastrointestinal conditions. Subsequently, we introduced non-invasive techniques combining breath biopsy with AI-enhanced integrated learning models to identified sets of breath biomarkers capable of accurately distinguishing CRC and GC. The diagnostic signatures effectively discriminated CRC and GC patients from normal controls with high AUC of 0.91 and 0.92 separately. The analysis of intestinal flora and single-cell sequencing were employed to carry out the in-depth in vivo traceability of identified markers, underscoring the reliability of exhaled VOCs as biomarkers for GI cancer. In order to indicate gastrointestinal disease severity intuitively, a Lasso regression model were applied to establish a VOC scoring system showcasing over 90% diagnostic sensitivity and specificity. Besides, the degree of inflammation was found to have a significantly higher effect on VOC scoring system than other pathways by constructing an SEM model, with effect sizes ranging from 3-fold to 10-fold. Our approach advances the design of exhaled analysis for GI cancer detection and holds promise as a non-invasive method to the clinic.