A new rapid diagnostic system with ambient mass spectrometry and machine learning for colorectal liver metastasis

Abstract

Background: Probe electrospray ionization-mass spectrometry (PESI-MS) can rapidly visualize mass spectra of small, surgically obtained tissue samples, and is a promising novel diagnostic tool when combined with machine learning which discriminates malignant spectrum patterns from others. The present study was performed to evaluate the utility of this device for rapid diagnosis of colorectal liver metastasis (CRLM). Methods: A prospectively planned study using retrospectively obtained tissues was performed. In total, 103 CRLM samples and 80 non-cancer liver tissues cut from surgically extracted specimens were analyzed using PESI-MS. Mass spectra obtained by PESI-MS were classified into cancer or non-cancer groups by using logistic regression, a kind of machine learning. Next, to identify the exact molecules responsible for the difference between CRLM and non-cancerous tissues, we performed liquid chromatography-electrospray ionization-MS (LC-ESI-MS), which visualizes sample molecular composition in more detail. Results: This diagnostic system distinguished CRLM from non-cancer liver parenchyma with an accuracy rate of 99.5%. The area under the receiver operating characteristic curve reached 0.9999. LC-ESI-MS analysis showed higher ion intensities of phosphatidylcholine and phosphatidylethanolamine in CRLM than in non-cancer liver parenchyma (P < 0.01, respectively). The proportion of phospholipids categorized as monounsaturated fatty acids was higher in CRLM (37.2%) than in non-cancer liver parenchyma (10.7%; P < 0.01). Conclusion: The combination of PESI-MS and machine learning distinguished CRLM from non-cancer tissue with high accuracy. Phospholipids categorized as monounsaturated fatty acids contributed to the difference between CRLM and normal parenchyma and might also be a useful diagnostic biomarker and therapeutic target for CRLM.