Deep learning in Myocarditis: A novel approach to severity assessment

Makoto Nishimori, Tomoyuki Otani, Yasuhide Asaumi,Keiko Ogo, Yoshihiko Ikeda, Kisaki Amemiya, Teruo Noguchi, Chisato Izumi, Masakazu Shinohara, Kinta Hatakeyama, Kunihiro Nishimura

Abstract
Myocarditis is life-threatening in the acute phase, yet biopsy—the diagnostic gold standard—lacks an objective method to quantify cardiomyocyte damage. We developed deep learning models to derive a pathology-based severity index for myocarditis from whole-slide biopsy images.

Introduction
Myocarditis is a very serious and high-risk disease carrying the risk of circulatory fluctuations during the acute phase [1–3]. Although clinical assessment typically involves symptoms, laboratory findings, and electrocardiograms, the definitive diagnosis relies on histopathological evaluation of myocardial biopsy [4]. Early assessment of clinical disease severity is also essential for guiding treatment and estimating prognosis [5].

Methods

We included 305 consecutive patients for whom at least one digitized H&E whole-slide image (WSI) of the index endomyocardial biopsy was available between 2002–2021 (1,056 WSIs). Digitization was part of routine archiving/batch scanning and independent of clinical outcomes. Within this eligible set, 145 had biopsy-proven myocarditis and 160 did not; the myocarditis cohort was used for severity-score modeling.

Results
The study period spanned from January 2002 to December 2021 at the NCVC. A total of 145 patients diagnosed with myocarditis, including children, were eligible for inclusion. These patients had undergone myocardial biopsy and were confirmed to have myocarditis. No specific exclusion criteria were set.

Discussion
In this study, we developed two AI models to evaluate the severity of myocarditis. Model 1, based on lymphocytes density, achieved an AUROC of 0.809, providing novel evidence that lymphocytes count correlate strongly with disease severity.

Conclusion
We introduce a pathology-based severity index for myocarditis that aligns with short-term clinical severity. The index offers a standardized histologic indicator of tissue injury and complements clinical assessment, providing a foundation for future multimodal prognostic modeling.

Citation: Nishimori M, Otani T, Asaumi Y, Ogo K, Ikeda Y, Amemiya K, et al. (2026) Deep learning in Myocarditis: A novel approach to severity assessment. PLoS One 21(8): e0354714. https://doi.org/10.1371/journal.pone.0354714

Editor: Muhammad Iqhrammullah, Universitas Muhammadiyah Aceh, INDONESIA

Received: January 11, 2026; Accepted: July 10, 2026; Published: August 12, 2026

Copyright: © 2026 Nishimori et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data underlying this study consist of sensitive human participant clinical and pathology data. Although the data used in the analyses were de-identified (pseudonymized), the individual-level dataset cannot be made publicly available without restriction because of the risk of participant re-identification. These restrictions were imposed by the ethics and institutional governance framework of the National Cerebral and Cardiovascular Center (NCVC), under Ethics Approval No. R21078-2. Requests for access to the minimal de-identified dataset may be directed to the NCVC Hospital Ethics Committee Secretariat (Clinical Ethics Office; hec@ncvc.go.jp), a non-author institutional contact, and will be considered in accordance with NCVC ethics and institutional governance requirements. The authors did not have access to identifiable information, and the re-identification key is retained by NCVC and was not provided to the authors.

Funding: The author(s) received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Abbreviations: AI, Artificial Intelligence; MIL, Multiple Instance Learning; WSIs, Whole-Slide pathology Images; AUROC, Area Under the Receiver Operating Characteristic curve; AUPRC, Area Under the Precision-Recall Curve; RNN, Recurrent Neural Network.