Near-infrared spectroscopy combined with machine learning for plasma-based discriminant diagnosis of malignant mesothelioma: a retrospective study

Our plain-English summary of Near-infrared spectroscopy combined with machine learning for plasma-based discriminant diagnosis of malignant mesothelioma: a retrospective study, by Y. Gu, R. Wang, W. Mao, et al (PeerJ, 2025).

The study

• Authors: Y. Gu, R. Wang, W. Mao, Z. Chen
• Journal: PeerJ
• Published: 2025;13:e20503
• Read it: PubMed · Publisher · Free full text

Abstract

This study developed a novel, non-invasive platform integrating near-infrared spectroscopy (NIRS) with machine learning (ML) to address the critical clinical challenge of misdiagnosing malignant mesothelioma (MM). We analyzed plasma samples from 99 individuals (29 MM, 41 lung cancer (LC), and 29 healthy controls). A support vector machine (SVM) model perfectly discriminated MM from LC (area under the curve (AUC) = 0.827), while a partial least squares (PLS) model differentiated MM from healthy control (HC) with high accuracy (AUC = 1.0). Despite the highly promising results, this single-center study is however limited by a small sample size, inherent to the rarity of MM and the associated difficulties in patient recruitment. Our findings demonstrate the potential of the NIRS-ML platform as a highly accurate tool for improving MM diagnosis and discriminant diagnosis, meriting further validation in larger cohorts.

© 2025 Gu et al.

In plain English

This retrospective laboratory study, carried out at Zhejiang Cancer Hospital in Hangzhou, China, tested whether near-infrared spectroscopy (a light-based scan of blood plasma) combined with machine learning (computer pattern-recognition) could tell malignant mesothelioma apart from lung cancer and from healthy people. The authors scanned stored plasma from 99 people: 29 with mesothelioma, 41 with lung cancer and 29 healthy controls. They compared ten computer methods. The samples were split into a training set of 66 and a testing set of 33, of which 10 were mesothelioma samples. Two methods worked best on unprocessed spectra. On the testing set, one method (SVM) separated mesothelioma from lung cancer with an area under the curve (AUC) of 0.827. Another (PLS) separated mesothelioma from healthy controls with an AUC of 1.0. The authors concluded that the method set a foundation for future large-scale plasma screening, particularly among people with high asbestos exposure. They noted that the sample was small and came from a single hospital, and that no independent external validation was performed.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Retrospective diagnostic laboratory study: stored biobank plasma was analysed by near-infrared spectroscopy, and ten machine-learning classification methods were trained and then tested on the spectra
• Who was studied: 99 plasma samples from Zhejiang Cancer Hospital, Hangzhou, China, obtained between October 2016 and October 2023: 29 from people with pathologically diagnosed malignant mesothelioma, 41 from people with pathologically diagnosed lung cancer, and 29 from healthy controls. The site of the mesothelioma (pleural or peritoneal) is not stated. The testing set had 33 samples: 11 healthy controls, 12 lung cancer and 10 mesothelioma.
• Exposure: No asbestos exposure was measured or reported for the participants. What was studied was a diagnostic test: near-infrared spectra of plasma, analysed with machine-learning methods.
• Main results:
  • Training set, mesothelioma vs lung cancer, using unprocessed spectra: the PLS model had an accuracy of 0.8863 and a kappa of 0.7577. The SVM model had an accuracy of 0.8863 and a kappa of 0.7692.
  • Training set, mesothelioma vs healthy controls: the PLS model had an accuracy of 0.9142 and a kappa of 0.8276. The SVM model had an accuracy of 0.9705 and a kappa of 0.9427.
  • Testing set: SVM with unprocessed spectra separated mesothelioma from lung cancer with an AUC of 0.827. PLS with unprocessed spectra separated mesothelioma from healthy controls with an AUC of 1.0.
  • In pairwise comparisons, PLS with unprocessed spectra performed best for distinguishing mesothelioma from healthy controls (P = 0.006, mean difference = 0.11, Cohen's d = 1.19).
  • In 700 rounds of shuffled-label permutation testing, the models' accuracy and kappa with the true labels fell outside the range produced by the shuffled labels.
  • Feature selection repeatedly picked the spectral region between 6,530 and 7,200 cm-1 (first overtones of OH, NH and CH vibrations) as among the most relevant in all four models.
• The authors' conclusion: The authors concluded that their non-invasive platform combining near-infrared spectroscopy with machine learning set a foundation for future large-scale plasma screening for mesothelioma and for telling it apart from other diseases, particularly among populations with high asbestos exposure.
• Limitations the authors state:
  • The sample was relatively small and came from a single hospital, which may limit statistical power and how far the results apply elsewhere.
  • No independent external validation was performed.
  • There were no detailed CT scan data from the same patients, so the test could not be compared directly with imaging.
  • The method detects chemical groups rather than specific molecules, so it cannot measure specific substances in the blood.

Topics

The National Library of Medicine indexes this study under subjects our site covers: Lung cancer, Mesothelioma.

Medical terms in this study, explained in our Lexicon: Infrared Spectroscopy (IR), Mesothelioma, Lung Cancer.

Related studies

• The use of a next-generation sequencing-derived machine-learning risk-prediction model (OncoCast-MPM) for malignant pleural mesothelioma: a retrospective study — M.G. Zauderer, A. Martin, et al. (2021)
• Cancer Risk Prediction Using Machine Learning for Supporting Early Cancer Diagnosis in Symptomatic Patients: A Systematic Review of Model Types — F. Pennisi, S. Borlini, et al. (2025)
• Distant Metastatic Pattern and Its Prognostic Significance in Malignant Pleural Mesothelioma: A Population-Based Study Based on a Machine Learning Model — J. Yu, C. Peng, et al. (2025)
• Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study — M. Yakar, D. Etiz, et al. (2021)
• Identification of glycolysis genes signature for predicting prognosis in malignant pleural mesothelioma by bioinformatics and machine learning — Y. Xiao, W. Huang, et al. (2022)
• Cancer-Associated Fibroblasts Influence Survival in Pleural Mesothelioma: Digital Gene Expression Analysis and Supervised Machine Learning Model — S. Borchert, A. Mathilakathu, et al. (2023)

Browse more research in PeerJ, or return to medical research.