Predicting isocitrate dehydrogenase status in glioma using hierarchical attention-based deep 3D multiple instance learning

Sep 28, 2026

Uploaded by Asher

Hierarchical Attention-Based Multiple Instance Learning (HAB-MIL) predicts isocitrate dehydrogenase (IDH) status in glioma using conventional preoperative MRI, targeting the clinical need for accurate noninvasive prognostic biomarkers. The approach addresses limitations of invasive surgical determination by training on 345 retrospectively collected patients and 495 additional cases from TCIA. HAB-MIL incorporates auxiliary positional encoding and evaluates performance via five-fold cross-validation with ROC, AUC, sensitivity, and specificity.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Background

Materials and methods

Results

Conclusion

Keywords

TYPE Original Research PUBLISHED 18 December 2025 DOI 10.3389/fonc.2025.1665690 OPEN ACCESS EDITED BY Francesco Bruno, University and City of Health and Science Hospital, Italy REVIEWED BY Giulia Berzero, IRCCS Ospedale San Raffaele, Italy Yu Liang, Harbin University of Science and Technology, China *CORRESPONDENCE Chen Niu [niuchen.xjtu@xjtu.edu.cn](niuchen.xjtu@xjtu.edu.cn) RECEIVED 16 July 2025 REVISED 02 December 2025 ACCEPTED 03 December 2025 PUBLISHED 18 December 2025 CITATION Xie Q, Sun Y, Liang Y, Shang Y, Wang H, Wang F, Wei R, Chen B, Zhang M and Niu C (2025) Predicting isocitrate dehydrogenase status in glioma using hierarchical attentionbased deep 3D multiple instance learning. Front. Oncol. 15:1665690 . doi: 10.3389/fonc.2025.1665690 COPYRIGHT © 2025 Xie, Sun, Liang, Shang, Wang, Wang, Wei, Chen, Zhang and Niu. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. Predicting isocitrate dehydrogenase status in glioma using hierarchical attentionbased deep 3D multiple instance learning Qinqin Xie 1,2, Yongheng Sun 3, Yuxia Liang 1, Yu Shang 1, Haifeng Wang 3, Fan Wang 3, Rong Wei 2, Bin Chen 4, Ming Zhang 1 and Chen Niu 1,2* 1 PET-CT Center, The First Afﬁliated Hospital of Xi ’an Jiaotong University, Xi ’an, China, 2 Department of Network Information, The First Afﬁliated Hospital of Xi’an Jiaotong University, Xi’an, China, 3School of Mathematics and Statistics, Xi’an Jiaotong University, Xi ’an, China, 4 Data Center, Hangzhou First People’s Hospital, Hangzhou, China Background: According to the 2021 WHO classiﬁcation of tumors of the central nervous system, isocitrate dehydrogenase (IDH) status serve an independent prognostic biomarker and is closely associated with tumor diagnosis and treatment response. At present, the determination of IDH status still relies on invasive surgical procedures. Method: A total of 345 patients with pathologically conﬁrmed gliomas diagnosed at the First Afﬁliated Hospital of Xi’an Jiaotong University between October 2019 and October 2024 were retrospectively included, comprising 148 (42 . 9%) IDHwild and 197 (57 . 1%) IDH-mutant. An additional 495 glioma patients were obtained from the public TCIA dataset. Patients were randomly split into training, validation, and test cohorts 6:2:2 . A Hierarchical Attention-Based Multiple Instance Learning (HAB-MIL) framework was developed, integrating auxiliary positional encoding into feature maps to capture spatially speciﬁc information and generate reﬁned 3D lesion representations . Model performance was evaluated using ﬁve-fold cross-validation, with receiver operating characteristic (ROC) curves, area under the curve (AUC), sensitivity, and speciﬁcity as assessment metrics. Result: HAB-MIL achieved competitive performance, with AUCs of 0 . 917 and 0. 892 on the glioma datasets from TCIA and the First Afﬁliated Hospital of Xi’an Jiaotong University. Additionally, our work achieves results that are comparable to the state-of-the-art methods in TCIA dataset and demonstrates that multiple instance learning has great potential for IDH prediction. Conclusion: The proposed HAB-MIL achieved IDH classiﬁcation based on conventional preoperative MRI images, eliminating the need for pixel-level annotations and signiﬁcantly reducing the annotation burden for doctors. KEYWORDS dynamic gated attention, glioma, IDH, location encoding, Multiple Instance Learning Frontiers in Oncology 01 [frontiersin.org](frontiersin.org) 1 Introduction Gliomas, the most prevalent malignant tumor in the central nervous system,

FAQs

Does the method require pixel-level annotations?

No. The model eliminates the need for pixel-level annotations by using multiple instance learning, which significantly reduces the annotation burden for clinicians.

Sep 28, 2026

Uploaded by Asher