The challenges posed by neurodegenerative diseases like Alzheimer’s and Parkinson’s demand sophisticated technological solutions to improve early diagnosis and patient outcomes. Central to these efforts is the effective handling of missing data in longitudinal studies, a common issue that can significantly impact the performance of predictive models.

Alzheimer’s Disease: Enhancing Prediction through Imputation Strategies

Based on the article: “Comparison between External and Internal Imputation of Missing Values in Longitudinal Data for Alzheimer’s Disease Diagnosis”

In the article “Comparison between External and Internal Imputation of Missing Values in Longitudinal Data for Alzheimer’s Disease Diagnosis,” Dr. Federica Aracri explored the impact of various imputation techniques on the accuracy of longitudinal deep learning models designed for predicting Alzheimer’s Disease (AD) progression. Utilizing data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the study evaluated four models—Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), DeepRNN, and ODE-RGRU—coupled with six imputation strategies, including advanced methods like MissForest and Multiple Imputation by Chained Equations (MICE).

The findings revealed that models such as ODE-RGRU and DeepRNN, when paired with external imputation techniques, significantly outperformed those relying on internal imputation. For instance, the combination of ODE-RGRU with median imputation achieved an mAUC value of 0.9 ± 0.002, and DeepRNN with MissForest reached an mAUC of 0.91 ± 0.004. These results underscore the critical role that robust imputation methods play in enhancing the accuracy of AD progression models.

Based on the article: “Imputation of Missing Clinical, Cognitive and Neuroimaging Data of Dementia using MissForest, a Random Forest Based Algorithm”

Another significant contribution by Dr. Aracri is the study presented in the article “Imputation of Missing Clinical, Cognitive and Neuroimaging Data of Dementia using MissForest, a Random Forest Based Algorithm,” where she assessed the reliability of the MissForest algorithm in handling missing data from Alzheimer’s Disease (AD) and Mild Cognitive Impairment (MCI) patients. The study compared MissForest with the commonly used Mean Imputation (Imean) method by simulating increasing levels of missing data in the ADNI dataset.

The research concluded that MissForest outperformed Imean in terms of overall imputation accuracy, particularly when considering the average error across all features. However, it was noted that MissForest had slightly higher errors than Imean for specific cognitive tests. These insights highlight the effectiveness of MissForest in handling missing data in dementia research, while also cautioning against its use with highly skewed variables.

Parkinson’s Disease: Classifying Phenotypes with Machine Learning

Based on the article: “Impact of Imputation Methods on Supervised Classification: A Multiclass Study on Patients with Parkinson’s Disease and Subjects with Scans Without Evidence of Dopaminergic Deficit”

Expanding on this work, Dr. Aracri also investigated the impact of imputation methods on supervised classification in the context of Parkinson’s Disease (PD). This study, detailed in the article “Impact of Imputation Methods on Supervised Classification: A Multiclass Study on Patients with Parkinson’s Disease and Subjects with Scans Without Evidence of Dopaminergic Deficit,” focused on the classification of PD, healthy controls, and a unique subgroup known as Scans Without Evidence of Dopaminergic Deficit (SWEDD). Two imputation approaches—MissForest and Mean Imputation (Imean)—were compared to assess their influence on the performance of tree-based algorithms, including Random Forest, XGBoost, and LightGBM.

The results demonstrated that while Mean Imputation occasionally led to overfitting, MissForest consistently retained more accurate information, proving to be the superior method for handling missing data in this context. This finding is particularly valuable for research into rare phenotypes of Parkinson’s Disease, where the accurate imputation of missing data is crucial for reliable classification outcomes.

Broader Implications and Future Directions

These works, conducted by Dr. Federica Aracri under my supervision, contribute significantly to the optimization of machine learning models for neurodegenerative disease research. The insights gained from these studies not only advance the understanding and prediction of diseases like Alzheimer’s and Parkinson’s but also have broader implications for other fields, particularly telemedicine. As healthcare continues to evolve with the integration of telehealth platforms, the methodologies developed in these studies could greatly enhance the reliability and utility of patient data collected remotely.

Moving forward, research will focus on incorporating additional biomarkers and conducting more extensive analyses to further refine these models. The ultimate goal is to improve early detection and personalized treatment strategies for neurodegenerative diseases, thereby enhancing patient outcomes on a global scale.