Deep neural network modeling for brain tumor classification using magnetic resonance spectroscopic imaging
Erin B. Bjørkeli, Knut Johannessen, Jonn Terje Geitung, Anna Karlberg, Live Eikenes, Morteza Esmaeili
Abstract
This study is driven by the complex and specialized nature of magnetic resonance spectroscopy imaging (MRSI) data processing, particularly within the scope of brain tumor assessments. Traditional methods often involve intricate manual procedures that demand considerable expertise. In response, we investigate the application of deep neural networks directly to raw MRSI data in the time domain.
Introduction
Brain tumors present a significant public health concern, necessitating early detection and classification for effective treatment. While conventional magnetic resonance imaging (MRI) is widely employed for this purpose, it faces limitations in accuracy and efficiency. Magnetic resonance spectroscopic imaging (MRSI), an advanced technique that provides insight into tissue chemical composition, holds promise as a valuable tool in the diagnostic and treatment assessment of gliomas.
Methods
Dataset simulation followed the approach by Lee and Kim, in order to generate ground truth target spectra from a spectral basis set consisting of key brain metabolites. The metabolites included in the basis set were aspartate (Asp), creatine (Cr), gamma-aminobutyric acid (GABA), glutamate (Glu), glutamine (Gln), glutathione (GSH), glycine (Gly), glycerophosphocholine (GPC), glycerophosphoethanolamine (GPE), Myo-inositol (Ins), lactate (Lac), N-acetyl-aspartate (NAA), N-acetyl-aspartyl glutamate (NAAG), phosphocholine (PCh), phosphocreatine (PCr), phosphorylethanolamine (PEtn), scyllo-Inositol (ScyIns), serine (Ser), and taurine (Tau).
Results
The domain transformation was conducted by the CNN block of the model. Synthetic, phantom, and in vivo MRSI data were employed to train and evaluate the model performance.
Discussion
In this study, we proposed a supervised deep learning model for classifying unprocessed raw MR spectra of healthy tissues from those of glioma tissues. Direct use of time-domain spectra acquired from the scanner without preprocessing can improve accessibility and potentially bypass the need for additional preprocessing. Our findings highlight the significant potential of deep learning in the realm of MRSI for brain tumor classification. The demonstrated success of our model in distinguishing tumor tissue from healthy tissue based on raw MRSI data in the time domain underscores the promise of automated, data-driven approaches to enhance diagnosis and streamline the analysis pipeline.
Conclusion
In conclusion, the demonstrated robustness in domain transformation, spanning from synthetic spectra to in vivo data and refined through the fine-tuning process, underscores the adaptability of our model to diverse datasets and real-world scenarios. The model’s efficacy in distinguishing glioma from healthy MR spectra further positions it as a valuable tool with potential clinical applications. The integration of MR spectroscopy with deep learning algorithms holds immense promise for improving the accuracy of in vivo investigation of lower-grade gliomas. As these techniques progress and gain broader acceptance in clinical settings, they have the potential to streamline the diagnostic and prognostic evaluation of glioma patients, providing valuable guidance for clinical decision-making.
Citation: Bjørkeli EB, Johannessen K, Geitung JT, Karlberg A, Eikenes L, Esmaeili M (2025) Deep neural network modeling for brain tumor classification using magnetic resonance spectroscopic imaging. PLOS Digit Health 4(4): e0000784. https://doi.org/10.1371/journal.pdig.0000784
Editor: Peter H. Charlton, University of Cambridge, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND
Received: September 10, 2024; Accepted: February 14, 2025; Published: April 9, 2025
Copyright: © 2025 Bjørkeli 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: MR spectra analyzed in this manuscript were obtained from publicly available repositories: MRS repository on NITRC:
\https://www.nitrc.org/projects/fmrs_2020/ (retrieved February 2024); MRSI repository on Zenodo: https://zenodo.org/records/3904584 (retrieved February 2024):
NOTE: Investigators should first fill out a form with their name and email, then they will be redirected to the Zenodo platform to create a username. Upon registration and email verification, the user will be granted access to Zenodo and the data used in this study. Due to ethical and legal restrictions, as well as compliance with the European Union General Data Protection Regulation (GDPR), the dataset acquired and analyzed from St. Olav Hospital and NTNU cannot be made publicly available, as its release would compromise patient privacy. However, investigators may request access to this dataset by contacting Øivind Rognmo at St. Olav Hospital and the Faculty of Medicine and Health Sciences, NTNU, Trondheim. Øivind Rognmo (oivind.rognmo@ntnu.no) is the institutional contact for data access and is involved in ethical oversight of the dataset. Requests will be subject to a data licensing agreement and institutional approval. Further details on the data transfer agreement process can be found here: (https://i.ntnu.no/wiki/-/wiki/Norsk/Dataoverf ø ringsavtale/.
Funding: This study was funded by the Southern Eastern Norway Regional Health Authority (Helse Sør-Øst RHF, HSØ; ME: grant number HSØ 2018047; E.B.B. HSØ 2021023). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
