Predicting cancer using supervised machine learning: Mesothelioma

Abstract

Insights

Artificial intelligence models can aid in the early diagnosis of Malignant Pleural Mesothelioma (MPM). Adaptive Boosting (AdaBoost) demonstrated the highest accuracy, identifying key predictors for Mesothelioma prognosis.

Area of Science:

  • Oncology
  • Medical Informatics
  • Machine Learning

Background:

  • Malignant Pleural Mesothelioma (MPM) is a rare but aggressive lung cancer.
  • Early diagnosis is crucial for patient outcomes but current methods are time-consuming and costly.
  • MPM accounts for approximately 75% of all Mesothelioma cases in the U.S.

Purpose of the Study:

  • To identify the optimal artificial intelligence (AI) model for early diagnosis and prognosis of Malignant Pleural Mesothelioma (MPM).
  • To compare the performance of various machine learning algorithms in predicting MPM.

Main Methods:

  • Retrospective analysis of clinical data from Dicle University, Turkey.
  • Application and evaluation of multiple AI algorithms including multilayered perceptron (MLP), kernel logistic regression (KLR), and adaptive boosting (AdaBoost).
  • Model comparison using paired t-tests based on classification accuracy, f-measure, precision, recall, ROC, and PRC.

Main Results:

  • Several AI models, including SGD, AdaBoost.M1, KLR, MLP, and VFDT, showed optimal performance in initial phases.
  • Adaptive Boosting (AdaBoost) achieved the highest classification accuracy at 71.29% in the second phase.
  • Key predictors for MPM prognosis identified include C-reactive protein, platelet count, symptom duration, gender, and pleural protein levels.

Conclusions:

  • AI models can effectively aid in the early diagnosis and prognosis of MPM.
  • While biopsy and imaging are accurate, AI offers a potentially more accessible and efficient approach.
  • Identifying key clinical predictors enhances the ability to prognosticate Mesothelioma.