# Classification of Alzheimer’s disease in a mixed clinical cohort using biofluid Raman spectroscopy

## TL;DR

This study shows that Raman spectroscopy can quickly and accurately classify Alzheimer's disease in a diverse group of patients using cerebrospinal fluid.

## Contribution

The study demonstrates the first real-world application of Raman spectroscopy for Alzheimer's diagnosis in a clinically heterogeneous cohort.

## Key findings

- AD was classified with 93% accuracy using Raman spectroscopy and machine learning.
- Spectral biomarkers were linked to protein-derived aromatic amino acids and correlated with known AD biomarkers like amyloid-β 42 and phosphorylated-tau 181.
- The method is rapid (<1 hour), reagentless, and does not require specialized labs.

## Abstract

There is a critical unmet need for scalable, accessible and objective diagnostic tests for stratification in dementia. Biofluid Raman spectroscopy (RS) due to its simplicity, holistic and label-free nature, is a powerful approach that has the potential to offer differential diagnosis across dementia types including Alzheimer’s disease (AD). RS is a laser-based optical method that can rapidly provide chemically rich information (‘spectral biomarkers’) from biofluids but its utility for AD diagnosis has not been established in a ‘real-world’ context, specifically from a clinically heterogenous cohort of patients. We carried out RS measurements on cerebrospinal fluid (CSF) samples of patients from a mixed clinical cohort (N = 143). All patients reported cognitive complaints and were clinically diagnosed over 2 years with conditions including AD and other neurodegenerative diseases, as well as developmental and long-term chronic conditions. Machine-learning algorithms were trained, optimised and evaluated on Raman spectra to classify AD from non-AD. AD was classified with 93% accuracy for patients in the testing set. Time from sample to classification was < 1 h. Spectral biomarkers explaining AD classification were identified and primarily assigned to protein-derived aromatic amino acids, representing a difference in proteome signature between AD and non-AD groups. Signals from a subset of spectral biomarkers directly correlated with pathological CSF biomarker concentrations including amyloid-β 42, phosphorylated-tau 181, and total tau. This pre-clinical study is a first step towards realising the real-world application of RS for dementia diagnosis. Compared to current and emerging methods, RS does not require sophisticated instrumentation or specialised labs. It is reagentless and simple, offering unprecedented rapidity, scalability, accessibility for dementia diagnosis.

The online version contains supplementary material available at 10.1186/s13195-025-01879-4.