Machine learning models to predict neuropsychiatric disorders in various brain tumors

Area of Science:

• Neuroscience
• Oncology
• Computer Science

Background:

• Neuropsychiatric disorders are common in brain tumor patients, yet difficult to predict.
• Early identification of these disorders is crucial for effective management.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) decision algorithms in predicting neuropsychiatric problems in diverse brain tumor types.
• To identify the most accurate ML model for this predictive task.

Main Methods:

• Trained and validated four supervised ML classification tree models (J48, Random Forest, Random Tree, Hoeffding Tree) on data from 145 primary brain tumor patients.
• Utilized patient attributes including age, gender, depression, dementia, and tumor type as predictors.
• Performed multi-target classification to predict various neuropsychiatric conditions.

Main Results:

• High prevalence of depression (86%) and dementia (55%) observed in brain tumor patients.
• Anger (92.41%), sleep disorders (83%), apathy (80%), and mood swings (76.55%) were common neuropsychiatric conditions.
• Glioblastoma patients showed higher rates of depression and dementia.
• Random Forest and Random Tree models achieved up to 94% accuracy in predicting neuropsychiatric disorders, with excellent precision, recall, and F-measure.

Conclusions:

• Random Forest Trees demonstrate high accuracy in predicting neuropsychiatric illnesses in brain tumor patients.
• ML-based decision trees can assist physicians in pre-diagnosing mental health issues and optimizing therapeutic strategies.
• This approach aids in proactively managing and preventing neuropsychiatric complications in brain tumor patients.