Biomarkers.

Area of Science:

• Biomedical Optics
• Spectroscopy
• Machine Learning in Medicine

Background:

• Clinical Alzheimer's Disease (AD) diagnosis accuracy varies due to inconsistency and patient heterogeneity.
• Raman spectroscopy offers rapid, label-free chemical analysis of biofluids.
• Previous AD classification studies lacked clinical representativeness and statistical power.

Purpose of the Study:

• To evaluate the efficacy of Raman spectroscopy and machine learning for AD diagnosis in a representative clinical cohort.
• To assess the performance of support-vector machine (SVM) and convolutional neural network (CNN) models for AD classification.
• To identify key spectral features indicative of AD and correlate them with established biomarkers.

Main Methods:

• Cerebrospinal fluid (CSF) samples from 141 patients (66 AD, 75 non-AD) were analyzed using Raman Spectroscopy.
• Machine learning models (SVM, CNN) were trained and optimized on 80% of the data and tested on 20%.
• Classifier performance was evaluated using Area Under the Receiver Operating Characteristic Curve (AUROC), sensitivity, and specificity.

Main Results:

• The SVM model achieved 93% classification accuracy with an AUROC of 0.92.
• Strong correlations were observed between classifier scores and patient ATN biomarker status.
• Key spectral features included protein-derived aromatic amino acids (phenylalanine, tyrosine).

Conclusions:

• Raman spectroscopy, combined with ML, accurately identifies AD in CSF from a mixed clinical cohort.
• This method correctly classified other neurodegenerative diseases as non-AD.
• Further validation in larger studies is needed to confirm population-level accuracy for potential clinical translation.