Abstract:
With no medication currently available and a clinical trial failure rate of 99.6% for
Alzheimer’s disease (AD) , early diagnosis is crucial to prevent its progression. MCI has
been identified as a transitional stage between healthy aging and AD, making it
promising for early detection. In this study, we propose a machine learning (ML) based
survival analysis approach to predict the time to AD conversion in early MCI and late
MCI stages separately, as we found that the progression rate varies in both stages. Unlike
typical ML classifiers, ML-based survival analysis models can provide information about
the timing and likelihood of disease progression. We employed multiple ML survival
models, including Random Survival Forest (RSF), Extra Survival Trees (XST), Gradient
Boosting Survival Analysis (GB), Survival Tree (ST), Cox-net, and Cox Proportional
Hazard (CoxPH), on 291 eMCI and 546 lMCI subjects. The study also compared
different data modalities, such as cognitive tests, neuroimaging tests, and cerebrospinal
fluid (CSF) biomarkers, both individually and in combination to identify the most
influential features for the models' performance. The results show that RSF outperformed
traditional CoxPH and other ML models used in this study. For the eMCI dataset, RSF
trained on multimodal data achieved a C-Index of 0.96 and an IBS of 0.02. For the lMCI
dataset, the C-Index was 0.82 and the IBS was 0.16. Additionally, the multimodal
analysis highlighted the importance of cognitive tests, as they exhibited a statistically
significant improvement over other modalities and multimodal data, demonstrating their
reliability in predicting AD progression. Finally, individual survival curves were
generated using RSF on baseline data to predict the probability of early onset of AD in
patients. This facilitates clinical decision-making by assisting clinicians in developing
personalized treatment strategies and implementing preventive measures to slow down or
potentially stop the progression of AD during its early stages.
