Skip to content

About

Performing clustering on stroke patients using NIHSS data.

Topics

Resources

Stars

4 stars

Watchers

1 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

Performing clustering on stroke patients based on NIHSS data.

This is the official repository for the research paper

Behavioral Clusters and Lesion Distributions in Ischemic Stroke, based on NIHSS Similarity

Research use only. Not a medical device. This software was developed in the HEREDITARY project as a research prototype. It has not been assessed or certified under the EU Medical Device Regulation (EU) 2017/745. It must not be used to diagnose, treat, monitor or make any other decision about individual patients.

1. Overview

Field Value
Name RSC Stroke
Version 0.1
Type Software or library
Owner partner and contact University of Padua, louisfabrice.tshimanga@unipd, andrea.zanola@phd.unipd.it
License SPDX identifier (see LICENSE file)
Repository and DOI 10.1007/s41666-025-00197-6
Release date 13.03.2025
Related task or deliverable T4.2
How to cite 10.1007/s41666-025-00197-6

2. Intended use

Primary intended use. Stroke is a leading cause of death and disability. The subsequent dysfunctions relate to brain lesion location. The complex relationship between behavioral symptoms and lesion location is essential for care and rehabilitation, and useful to understand the healthy brain. Such complexity eludes linear methods. Differently from previous works, this study clusters patients by recurrent symptoms profiles, then retrieves their neural correlates.

Intended users. biomedical researchers, data scientists.

Out-of-scope uses. This release must not be used for

  • clinical diagnosis, prognosis or treatment of individual patients;
  • decisions about individuals, such as insurance, employment or access to care;
  • re-identification of individuals or linkage with other data sources;
  • populations, devices or data types not described in Section 3.

3. Data provenance

Dataset Source partner Type (public, synthetic, pseudonymized, anonymized) Population (size, age, sex, country, condition) Ethics approval or access conditions
UNIPD and SL Stroke dataset University of Padua anonymized 172 subjects (87M), 68.9 ± 15.0 y for Padua (IT) and 55.3 ± 11.7 y for St. Louis (US), stroke Written informed consent was obtained from all participants in accordance with the Declaration
of Helsinki, retrospective study

This repository and the released files contain no personal data, pseudonymized records, identifiers or pseudonymization keys.

4. Method

  • Algorithm or architecture. Repeated Spectral Clustering
  • Training setting. Centralized
  • Privacy techniques. none

5. Performance

  • Evaluation data. internal validation.
  • Metrics and results. 3 CLusters with FWER at alpha=0.05 significant different lesions.
  • Results by subgroup. NA.

6. Known limitations and risks

  • Generalizability. Small cohort of ischemic stroke only.
  • Known biases and confounders. Small sample, mostly caucasian descent.

7. Responsible use and reporting

8. Acknowledgement

Funded by the European Union under grant agreement No 101137074 (HEREDITARY). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HaDEA). Neither the European Union nor the granting authority can be held responsible for them.

Methods: Prevalence, severity and associations of symptoms affecting patients are condensed in a similarity measure between their profiles. The computation specifically adheres to the ordinal nature of NIHSS data. A network model links patients with strength proportional to the behavioral similarity. The network nodes are then classified through a novel spectral clustering variation. Lesions from each cluster's patients are used in a voxel-wise analysis to statically validate the findings.

Results: The behavioral clusters express symptoms co-occurring in accordance with the literature on behavioral deficits covariance. Moreover, they show coherent groups of brain lesions in the associated CT and MRI scans. The anatomical position of the significant voxels found are aligned with the literature. Even when lesion density maps overlap, the significant voxels are well separated.

Conclusion: The presented workflow offers concrete, clinically useful correlations between patients profiles, individual patients-to-group associations, in-group co-occurrences of symptoms, and brain lesion correlates. The underlying mathematical assumptions, and the unsupervised machine learning techniques offer statistically robust results and are being further applied to other multimodal biomedical data.

Authors and Citation

If you find codes and results useful for your research, please concider citing our work. It would help us to continue our research.

Contributors:

  • M.Sc. Louis Fabrice Tshimanga
  • M.Sc. Andrea Zanola
  • Dr. Silvia Facchini
  • Dr. Antonio Luigi Bisogno
  • PhD. Lorenzo Pini
  • Prof. Manfredo Atzori
  • Prof. Maurizio Corbetta

License

The code is released under the MIT License

About

Performing clustering on stroke patients using NIHSS data.

Topics

Resources

Stars

4 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages