Individuals with Mild Cognitive Impairment (MCI) are at greater risk of developing Alzheimer’s disease (AD). Early detection of those more likely to develop AD holds great potential for preventative and/or therapeutic interventions. This thesis aims to explore demographic, neuroimaging, and neuropsychiatric data to identify the most informative factors influencing MCI to AD conversion. Using machine learning models, the research seeks to distinguish between MCI converters and non-converters, with a specific focus on exploring the potential association between choroid plexus volume and conversion. Choroid plexus volume was introduced as a representative measure of choroid plexus morphology, which may be linked to choroid plexus function and/or cerebrospinal fluid dynamics. The data involved come from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database and only data from baseline visits are considered. The dataset includes 393 MCI patients who converted to AD and 130 MCI patients who did not convert to AD. The first step involves replicating the methodology of Matthew Velasquez et al. (2021). This process includes training different machine learning models (random forest, logistic regression, SVM, XGBoost) on distinct feature subsets and assessing their importance in predicting MCI conversion to AD. Choroid plexus volume measures are then incorporated into the most effective models to investigate their impact on predictive outcomes. Feature importance is explored with permutation importance, LIME, and SHAP algorithms. For the classification task without the choroid plexus feature, SVM and logistic regression gave the best performance, especially in terms of specificity (SVM6 = 0.942, LR6 = 0.885, SVM13 = 0.885, LR13 = 0.857). The most important features of these models were ADAS13, FAQ, Hippocampus, and APOE4. When including the choroid plexus variable in the four best models, performances in specificity slightly decreased (SVM7 = 0.885, LR6 = 0.885, SVM13 = 0.857, LR13 = 0.828), suggesting that choroid plexus volume may not provide useful information for predicting conversion to AD. From feature importance analysis, ADAS13, FAQ, Hippocampus, and APOE4 confirmed their importance in the prediction, while the choroid plexus volume had little impact on the model decision. This study showed that the choroid plexus volume may not play a central role in predicting the conversion of MCI patients to AD. It may be possible that disease-related changes to the choroid plexus that may be useful in prediction are not captured in the volume measure, and other measures of choroid plexus function or cerebrospinal fluid dynamics may be more informative. Further investigations into the role of the choroid plexus in MCI and AD are needed.

Individuals with Mild Cognitive Impairment (MCI) are at greater risk of developing Alzheimer’s disease (AD). Early detection of those more likely to develop AD holds great potential for preventative and/or therapeutic interventions. This thesis aims to explore demographic, neuroimaging, and neuropsychiatric data to identify the most informative factors influencing MCI to AD conversion. Using machine learning models, the research seeks to distinguish between MCI converters and non-converters, with a specific focus on exploring the potential association between choroid plexus volume and conversion. Choroid plexus volume was introduced as a representative measure of choroid plexus morphology, which may be linked to choroid plexus function and/or cerebrospinal fluid dynamics. The data involved come from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database and only data from baseline visits are considered. The dataset includes 393 MCI patients who converted to AD and 130 MCI patients who did not convert to AD. The first step involves replicating the methodology of Matthew Velasquez et al. (2021). This process includes training different machine learning models (random forest, logistic regression, SVM, XGBoost) on distinct feature subsets and assessing their importance in predicting MCI conversion to AD. Choroid plexus volume measures are then incorporated into the most effective models to investigate their impact on predictive outcomes. Feature importance is explored with permutation importance, LIME, and SHAP algorithms. For the classification task without the choroid plexus feature, SVM and logistic regression gave the best performance, especially in terms of specificity (SVM6 = 0.942, LR6 = 0.885, SVM13 = 0.885, LR13 = 0.857). The most important features of these models were ADAS13, FAQ, Hippocampus, and APOE4. When including the choroid plexus variable in the four best models, performances in specificity slightly decreased (SVM7 = 0.885, LR6 = 0.885, SVM13 = 0.857, LR13 = 0.828), suggesting that choroid plexus volume may not provide useful information for predicting conversion to AD. From feature importance analysis, ADAS13, FAQ, Hippocampus, and APOE4 confirmed their importance in the prediction, while the choroid plexus volume had little impact on the model decision. This study showed that the choroid plexus volume may not play a central role in predicting the conversion of MCI patients to AD. It may be possible that disease-related changes to the choroid plexus that may be useful in prediction are not captured in the volume measure, and other measures of choroid plexus function or cerebrospinal fluid dynamics may be more informative. Further investigations into the role of the choroid plexus in MCI and AD are needed.

Data-Driven Predictive Modeling for Alzheimer's Disease Progression: Integrating Choroid Plexus Volume Analysis

ROVERONI, EMMA
2023/2024

Abstract

Individuals with Mild Cognitive Impairment (MCI) are at greater risk of developing Alzheimer’s disease (AD). Early detection of those more likely to develop AD holds great potential for preventative and/or therapeutic interventions. This thesis aims to explore demographic, neuroimaging, and neuropsychiatric data to identify the most informative factors influencing MCI to AD conversion. Using machine learning models, the research seeks to distinguish between MCI converters and non-converters, with a specific focus on exploring the potential association between choroid plexus volume and conversion. Choroid plexus volume was introduced as a representative measure of choroid plexus morphology, which may be linked to choroid plexus function and/or cerebrospinal fluid dynamics. The data involved come from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database and only data from baseline visits are considered. The dataset includes 393 MCI patients who converted to AD and 130 MCI patients who did not convert to AD. The first step involves replicating the methodology of Matthew Velasquez et al. (2021). This process includes training different machine learning models (random forest, logistic regression, SVM, XGBoost) on distinct feature subsets and assessing their importance in predicting MCI conversion to AD. Choroid plexus volume measures are then incorporated into the most effective models to investigate their impact on predictive outcomes. Feature importance is explored with permutation importance, LIME, and SHAP algorithms. For the classification task without the choroid plexus feature, SVM and logistic regression gave the best performance, especially in terms of specificity (SVM6 = 0.942, LR6 = 0.885, SVM13 = 0.885, LR13 = 0.857). The most important features of these models were ADAS13, FAQ, Hippocampus, and APOE4. When including the choroid plexus variable in the four best models, performances in specificity slightly decreased (SVM7 = 0.885, LR6 = 0.885, SVM13 = 0.857, LR13 = 0.828), suggesting that choroid plexus volume may not provide useful information for predicting conversion to AD. From feature importance analysis, ADAS13, FAQ, Hippocampus, and APOE4 confirmed their importance in the prediction, while the choroid plexus volume had little impact on the model decision. This study showed that the choroid plexus volume may not play a central role in predicting the conversion of MCI patients to AD. It may be possible that disease-related changes to the choroid plexus that may be useful in prediction are not captured in the volume measure, and other measures of choroid plexus function or cerebrospinal fluid dynamics may be more informative. Further investigations into the role of the choroid plexus in MCI and AD are needed.
2023
Data-Driven Predictive Modeling for Alzheimer's Disease Progression: Integrating Choroid Plexus Volume Analysis
Individuals with Mild Cognitive Impairment (MCI) are at greater risk of developing Alzheimer’s disease (AD). Early detection of those more likely to develop AD holds great potential for preventative and/or therapeutic interventions. This thesis aims to explore demographic, neuroimaging, and neuropsychiatric data to identify the most informative factors influencing MCI to AD conversion. Using machine learning models, the research seeks to distinguish between MCI converters and non-converters, with a specific focus on exploring the potential association between choroid plexus volume and conversion. Choroid plexus volume was introduced as a representative measure of choroid plexus morphology, which may be linked to choroid plexus function and/or cerebrospinal fluid dynamics. The data involved come from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database and only data from baseline visits are considered. The dataset includes 393 MCI patients who converted to AD and 130 MCI patients who did not convert to AD. The first step involves replicating the methodology of Matthew Velasquez et al. (2021). This process includes training different machine learning models (random forest, logistic regression, SVM, XGBoost) on distinct feature subsets and assessing their importance in predicting MCI conversion to AD. Choroid plexus volume measures are then incorporated into the most effective models to investigate their impact on predictive outcomes. Feature importance is explored with permutation importance, LIME, and SHAP algorithms. For the classification task without the choroid plexus feature, SVM and logistic regression gave the best performance, especially in terms of specificity (SVM6 = 0.942, LR6 = 0.885, SVM13 = 0.885, LR13 = 0.857). The most important features of these models were ADAS13, FAQ, Hippocampus, and APOE4. When including the choroid plexus variable in the four best models, performances in specificity slightly decreased (SVM7 = 0.885, LR6 = 0.885, SVM13 = 0.857, LR13 = 0.828), suggesting that choroid plexus volume may not provide useful information for predicting conversion to AD. From feature importance analysis, ADAS13, FAQ, Hippocampus, and APOE4 confirmed their importance in the prediction, while the choroid plexus volume had little impact on the model decision. This study showed that the choroid plexus volume may not play a central role in predicting the conversion of MCI patients to AD. It may be possible that disease-related changes to the choroid plexus that may be useful in prediction are not captured in the volume measure, and other measures of choroid plexus function or cerebrospinal fluid dynamics may be more informative. Further investigations into the role of the choroid plexus in MCI and AD are needed.
Alzheimer's Disease
Machine learning
Choroid plexus
Random forest
Features importance
File in questo prodotto:
File Dimensione Formato  
Roveroni_Emma.pdf

accesso riservato

Dimensione 1.36 MB
Formato Adobe PDF
1.36 MB Adobe PDF

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/64796