# From scales to circuits: integrating behavioral diagnosis and neural biomarkers for improved classification in disorders of consciousness

## TL;DR

This study combines behavioral assessments and brain imaging to better classify patients with disorders of consciousness, improving accuracy in distinguishing between different patient groups.

## Contribution

A novel data-driven method integrating behavioral and neuroimaging data to classify disorders of consciousness with high accuracy.

## Key findings

- 31 out of 58 UWS and 23 out of 30 MCS patients were identified as representative with 90.2% classification accuracy.
- Classification accuracy dropped significantly when including nonrepresentative patients.
- Altered DMN connectivity patterns were consistent in representative patients but not in excluded ones.

## Abstract

In this study, we propose a data-driven approach that integrates behavioral diagnosis with neuroimaging features to identify representative UWS and MCS patients from a large inpatient cohort.

Clinical information was extracted using a subset of UWS patients with CRS-R scores ≤ 5. Neuroimaging biomarkers were established as the increased and decreased functional connectivity indices of anatomically defined regions covering the whole brain. The algorithm was implemented through an iterative refinement process that converged on a division of UWS and MCS patients into representative and excluded (or nonrepresentative) patient groups.

Thirty-one out of 58 UWS patients and 23 out of 30 MCS patients were identified as representative, with an average classification accuracy of 90.2% in differentiating between the two groups. In contrast, differentiating between excluded UWS patients (n = 27) and representative MCS patients (n = 23) and between all UWS (n = 58) and MCS (n = 30) patients produced average classification accuracies of 50.9 and 64.3%, respectively. Furthermore, altered DMN functional connectivity between representative UWS and MCS patients revealed a consistent pattern as shown in prior studies, while comparisons involving excluded patients did not.

These results highlight the value of integrating behavioral scores and neural connectivity features for DOC classification, providing a more coherent basis for downstream analysis and machine-learning applications in DOC classification.