Artificial Intelligence for Alzheimer’s Detection through Advanced Diagnostic Profiling and Multimodal Fusion

Alzheimer’s disease is a complex neurodegenerative condition that presents significant challenges in early diagnosis and progression prediction for faster results. In this work, BioML-AD is introduced as the first multimodal machine learning framework for integrating and analyzing a range of biomarker profiles, including genetic, neuroimaging, and cognitive data, to improve early detection and prediction of disease progression. The method is composed of three main layers CNN feature extraction for neuroimaging data, temporal cognitive assessment with RNN, and genetic biomarkers with random forests. The outputs are then fused through a multimodal attention network to enhance the accuracy of clinical predictions. The features are then combined through a Multimodal Attention Network (MMAN) to focus on the few biomarkers that contribute most to the prediction of early-stage AD and then additionally build a hybrid Long Short-Term Memory (LSTM) and transformer-based model to predict progression from MCI to AD. The performance of BioML-AD is evaluated across five key measures early detection accuracy, identifying early-stage AD, biomarker contribution score, estimating biomarker importance, progression prediction reliability, providing stable predictions over time, model interpretability, using shape values for feature explanation, and generalization capability, assessing performance across diverse patient data. It has been tested on an MRI dataset for early diagnosis and clinical decision support, and BioML-AD proves to be an efficient and effective tool, offering highly sensitive and reliable results for clinical applications.

Keywords:

Cognitive Assessment , neuroimaging , Machine Learning , Memory , Predictive Modeling