DIAGNOSTICS OF NEURODEGENERATIVE DISEASES BY MACHINE LEARNING APPROACH: NEW LABORATORY AND CLINICAL BIOMARKER SELECTION

Objectives Low sensitivity and specificity of current diagnostic methodologies, lead to frequent dementia misdiagnosis. This raises the need of more efficient integration of biomarkers with multidimensional clinical data. We aim to: -Collect real word data from clinical centres: Alzheimer's (AD) and Parkinson's diseases (PD) and controls; -Integrate real word data to a larger database based on AddNeuroMed, ADNI and PPMI projects; -Measure innovative biomarkers in clinical biological samples; -Apply Machine Learning methodologies (ML) to integrate biomarkers with clinical features of AD and PD patients, for patient stratification and refined diagnosis. Methods Hundreds of variables in AddNeuroMed and ADNI databases were correlated with real world data from excellence Italian dementia centres. ML to combine continuous, discrete and categorical data, generated diagnostic models based on few discriminant variables, to classify patients into disease categories with probabilistic scores. New high-throughput NGF and proNGF protein measures and transcriptomics by RNA-Seq NGS in fluid biosamples, of AD, PD, controls, were correlated with clinical data by ML. Models are being continuously updated using new patients data fed by clinicians. Results The models provided reliable diagnoses, with good sensitivity and higher accuracy also for newly diagnosed patients. We extracted new diagnostic knowledge integrating new potential laboratory biomarkers and clinical variables. Conclusions The diagnostic models allowed to focus on essential clinical variables and support clinicians in diagnostic process. This approach will be the framework to integrate new clinical protocols and will allow to design new faster and potentially cheaper diagnostic work-flows for healthcare services.