CoChat

Multimodal deep learning model for prediction of prognosis in central nervous system inflammation

Open access
Choi, Bo Kyu, Choi, Yoonhyeok, Jang, Sooyoung, Ha, Woo-Seok, Cho, Soomi, Chang, Kimoon, Sohn, Beomseok, Kim, Kyung Min, Park, Yu Rang · 2025 · Brain Communications

The DOI matches its Crossref record; 37 of the 37 references it cites were found in a registry or an index.

Abstract

Inflammatory diseases of the CNS impose a substantial disease burden, necessitating prompt and appropriate prognosis prediction. We developed a multimodal deep learning model integrating clinical features and brain MRI data to enhance early prognosis prediction of CNS inflammation. This retrospective study used thin-cut T1-weighted brain MRI scans and the clinical variables of patients with CNS inflammation who were admitted to a tertiary referral hospital between January 2010 and December 2023. Data collected after January 2022 served as the external test set. 3D MRI images were first segmented into 43 brain regions using the FastSurfer library. The segmented images were then processed through a 3D convolutional neural network model for feature extraction and vectorization, after which they were integrated with clinical features for prediction. The performance of each artificial intelligence model was assessed using accuracy, F1 score, area under the receiver operating characteristic curve and area under the precision-recall curve. The internal dataset comprised 413 images from 291 patients (mean age, 45.5 years ± 19.3 [SD]; 151 male patients; 54 with poor prognosis). The external dataset comprised 210 images from 106 patients (mean age, 45.5 years ± 18.9 [SD]; 59 male patients; 31 with poor prognosis). The multimodal deep learning model outperformed unimodal models across all aetiological groups, achieving area under the receiver operating characteristic curve values of 0.8048 for autoimmune, 0.9107 for bacterial, 1.0000 for tuberculosis and 0.9242 for viral infections. Furthermore, artificial intelligence assistance improved clinicians' prognostic accuracy, as demonstrated in comparisons with neurologists, paediatricians and radiologists. Our findings demonstrate that the multimodal deep learning model enhances artificial intelligence-assisted prognosis prediction in CNS inflammation, improving both model performance and clinician decision-making.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1093/braincomms/fcaf179

Cites

37 of 37 references verified0 citation errors · 0 not checkable

Statistics

Figures

Graphical Abstract
Graphical Abstract CC BY · PubMed Central
Figure 1 The schematic diagram of the overall training process for multimodal deep learning. DL, deep learning; MLP, multilayer perceptron; DenseNet, densely connected convolutional networks.
Figure 1 The schematic diagram of the overall training process for multimodal deep learning. DL, deep learning; MLP, multilayer perceptron; DenseNet, densely connected convolutional networks.CC BY · PubMed Central
Figure 2 Hierarchical clustering results based on predictive performance of unimodal models using brain MRI. (A) The 43 brain regions were divided into 6 clusters. (B–D) Visualization of hierarchical clustering of each brain region overlaid on brain MRI images. (B) Coronal, (C) sagittal and (D) axial section. Hierarchical clustering was performed using an agglomerative approach with average linkage and Euclidean distance as the dissimilarity metric. Clustering was conducted on an internal dataset derived from the results of the MRI unimodal model. The dataset included autoimmune (n = 45), bacterial (n = 46), tuberculosis (n = 30) and viral (n = 170) cases.
Figure 2 Hierarchical clustering results based on predictive performance of unimodal models using brain MRI. (A) The 43 brain regions were divided into 6 clusters. (B–D) Visualization of hierarchical clustering of each brain region overlaid on brain MRI images. (B) Coronal, (C) sagittal and (D) axial section. Hierarchical clustering was performed using an agglomerative approach with average linkage and Euclidean distance as the dissimilarity metric. Clustering was conducted on an internal dataset derived from the results of the MRI unimodal model. The dataset included autoimmune (n = 45), bacterial (n = 46), tuberculosis (n = 30) and viral (n = 170) cases.CC BY · PubMed Central
Figure 3 Top 20 variables crucially utilized in multimodal deep learning for each aetiology in the external test set and their SHAP values. A batch-sampling strategy was employed for SHAP estimation, selecting one-fifth of the test dataset at a time to compute SHAP values while balancing memory efficiency and estimation performance. (A) Autoimmune (n = 29), (B) bacteria (n = 8), (C) tuberculosis (n = 3) and (D) virus (n = 66). CSF, cerebrospinal fluid; WBC, white blood cell; PMN, polymorphonuclear.
Figure 3 Top 20 variables crucially utilized in multimodal deep learning for each aetiology in the external test set and their SHAP values. A batch-sampling strategy was employed for SHAP estimation, selecting one-fifth of the test dataset at a time to compute SHAP values while balancing memory efficiency and estimation performance. (A) Autoimmune (n = 29), (B) bacteria (n = 8), (C) tuberculosis (n = 3) and (D) virus (n = 66). CSF, cerebrospinal fluid; WBC, white blood cell; PMN, polymorphonuclear.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 37 references

Research works printed but not linked: 1

  1. Ester, 1996, Proceedings—2nd international conference on knowledge discovery and data mining, KDD 1996, 226

Metadata

Authors
Choi, B. K., Choi, Y., Jang, S., Ha, W.-S., Cho, S., Chang, K., Sohn, B., Kim, K. M., & Park, Y. R.
Affiliations
  • Yonsei University (KR)
  • Eulji University (KR)
  • Samsung Medical Center (KR)
Corresponding author
  • Kim, Kyung Min
  • Park, Yu Rang
Venue
Brain Communications
Published
2025
Volume
7
Issue
3
Pages
fcaf179-fcaf179
ISSN
2632-1297 (ISSN-L)
OpenAlex
W4410225724
Cited by
1
References
35
Citations by year
2026: 1
FWCI
0.40
Type
journal article
Language
en
Open access
Locations
  • Brain Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • Hyundai Motor Chung Mong-Koo Foundation
  • Korea Health Industry Development Institute
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
Keywords
neuroinflammation, outcome prediction, multimodal deep learning, brain MRI, T1-weighted imaging, brain structure segmentation, 3D convolutional neural networks, clinical features, autoimmune encephalitis, bacterial meningitis, tuberculous meningitis, viral encephalitis, AI-assisted diagnosis, clinician decision-making

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.