Predicting cancer using supervised machine learning: Mesothelioma

Our plain-English summary of Predicting cancer using supervised machine learning: Mesothelioma, by A. Choudhury (Technology and Health Care, 2021).

The study

• Author: A. Choudhury
• Journal: Technology and Health Care
• Published: 2021;29(1):45-58
• Cited by: 21 articles, as counted by NIH iCite
• Read it: PubMed · Publisher

Abstract

Background: Pleural Mesothelioma (PM) is an unusual, belligerent tumor that rapidly develops into cancer in the pleura of the lungs. Pleural Mesothelioma is a common type of Mesothelioma that accounts for about 75% of all Mesothelioma diagnosed yearly in the U.S. Diagnosis of Mesothelioma takes several months and is expensive. Given the risk and constraints associated with PM diagnosis, early identification of this ailment is essential for patient health.

Objective: In this study, we use artificial intelligence algorithms recommending the best fit model for early diagnosis and prognosis of Malignant Pleural Mesothelioma (MPM).

Methods: We retrospectively retrieved patients' clinical data collected by Dicle University, Turkey and applied multilayered perceptron (MLP), voted perceptron (VP), Clojure classifier (CC), kernel logistic regression (KLR), stochastic gradient decent (SGD), adaptive boosting (AdaBoost), Hoeffding tree (VFDT), and primal estimated sub-gradient solver for support vector machine (s-Pegasos). We evaluated the models, compared and tested them using paired t-test (corrected) at 0.05 significance based on their respective classification accuracy, f-measure, precision, recall, root mean squared error, receivers' characteristic curve (ROC), and precision-recall curve (PRC).

Results: In phase 1, SGD, AdaBoost.M1, KLR, MLP, VFDT generate optimal results with the highest possible performance measures. In phase 2, AdaBoost, with a classification accuracy of 71.29%, outperformed all other algorithms. C-reactive protein, platelet count, duration of symptoms, gender, and pleural protein were found to be the most relevant predictors that can prognosticate Mesothelioma.

Conclusion: This study confirms that data obtained from biopsy and imaging tests are strong predictors of Mesothelioma but are associated with a high cost; however, they can identify Mesothelioma with optimal accuracy.

In plain English

This study tested whether computer models (machine learning) could identify malignant pleural mesothelioma from patient records. The author used clinical data on 324 patients, 41% of them women, collected by Dicle University in Turkey. Each patient had 35 variables, including age, gender, smoking, asbestos exposure and its duration, symptoms, and blood and pleural fluid tests. Eight algorithms were compared using 10-fold cross-validation. In the first phase, the models received all variables except whether the patient died. These included a 'diagnosis method' variable that the author describes as biopsy and imaging results. Five algorithms classified patients with 100% accuracy and one came close. In the second phase, biopsy, imaging and other post-diagnosis data were removed. A selection method picked five predictors: C-reactive protein, platelet count, duration of symptoms, gender and pleural protein. Asbestos exposure was not among them, though the author noted this did not downgrade the other predictors. The best model, AdaBoost, reached 71.29% accuracy. The author concluded that the second approach could shortlist high-risk patients at lower cost, with the first used to confirm the disease.

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: Analysis of existing clinical records comparing eight machine learning classification models, in two phases, using 10-fold cross-validation
• Who was studied: 324 patients (41% female) from clinical reports generated by Dicle University, Turkey, with 35 variables each; patients were from nine different cities
• Exposure: Not an exposure study. Model inputs were clinical variables, including age, gender, city, smoking habit, asbestos exposure, duration of asbestos exposure, symptoms and blood test results. Phase 1 also included a 'diagnosis method' variable derived from biopsy and imaging tests.
• Main results:
  • Phase 1 (all predictors except 'dead or not', including biopsy and imaging results): SGD, AdaBoost.M1, KLR, MLP and VFDT achieved 100% classification accuracy
  • Phase 1: s-Pegasos came close to the optimal result
  • Phase 2 (pre-diagnosis data only): the Boruta feature selection method identified five relevant predictors: C-reactive protein, platelet count, duration of symptoms, gender, and pleural protein
  • Phase 2: AdaBoost had the highest classification accuracy, 71.29%
• The authors' conclusion: The author recommended AdaBoost and suggested using the phase 2 approach to shortlist high-risk patients, followed by the phase 1 approach (which needs biopsy and imaging data) to confirm pleural mesothelioma
• Limitations the authors state:
  • Removing biopsy and imaging data in phase 2 lowered accuracy
  • The author stated that the feature selection method does not downgrade the predictors it did not select, and does not recommend changing regular clinical procedures

Topics

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

Medical terms in this study, explained in our Lexicon: Mesothelioma, Pleural Mesothelioma, Pleura, Prognosis, Biopsy.

Related studies

• Differential diagnosis of pleural mesothelioma using Logic Learning Machine — S. Parodi, R. Filiberti, et al. (2015)
• 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)
• Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study — M. Yakar, D. Etiz, et al. (2021)
• Combination of Plasma-Based Metabolomics and Machine Learning Algorithm Provides a Novel Diagnostic Strategy for Malignant Mesothelioma — N. Li, C. Yang, et al. (2021)
• Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning — D. Chicco, A. Alameer, et al. (2022)
• Artificial intelligence-based diagnosis of asbestosis: analysis of a database with applicants for asbestosis state aid — K.B.W. Groot Lipman, C.J. de Gooijer, et al. (2023)

Browse return to medical research.