Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands
Brain tumors, particularly gliomas, pose a significant clinical challenge with rising incidence rates and high mortality. Artificial intelligence combined with Hyperspectral Imaging (HSI) offers promising tools to improve surgical precision and patients’ outcomes. HSI offers unique advant...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11077150/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1839625378992226304 |
|---|---|
| author | Domenico Ragusa Marco Gazzoni Emanuele Torti Elisa Marenzi Francesco Leporati |
| author_facet | Domenico Ragusa Marco Gazzoni Emanuele Torti Elisa Marenzi Francesco Leporati |
| author_sort | Domenico Ragusa |
| collection | DOAJ |
| description | Brain tumors, particularly gliomas, pose a significant clinical challenge with rising incidence rates and high mortality. Artificial intelligence combined with Hyperspectral Imaging (HSI) offers promising tools to improve surgical precision and patients’ outcomes. HSI offers unique advantages, including non-invasiveness and detailed spectral data, for enhanced tumor tissue differentiation. This study tailored the Vision Transformer (ViT) with techniques from remote sensing to segment nineteen spectral images of low- and high-grade gliomas with limited spectral bands. The choice of the ViT was motivated by its attention mechanism, enabling fine-grained distinction of subtle details. Segmentation focused on four classes: healthy tissue, tumor tissue, blood vessels and dura mater. A careful hyperparameter optimization was performed, resulting in the selection of two models based on a defined quality index, which were evaluated using three experimental methodologies, achieving up to <inline-formula> <tex-math notation="LaTeX">$98.24\pm 2.50$ </tex-math></inline-formula>% average Overall Accuracy (OACC) and <inline-formula> <tex-math notation="LaTeX">$99.61\pm 0.66$ </tex-math></inline-formula>% average Area Under the Curve (AUC) in intra-patient classification. For inter-patient classification, the models achieved an average OACC up to <inline-formula> <tex-math notation="LaTeX">$53.56\pm 24.91$ </tex-math></inline-formula>% and an average AUC score up to <inline-formula> <tex-math notation="LaTeX">$79.27\pm 10.43$ </tex-math></inline-formula>%, highlighting areas of improvement. Comparable or improved performance was demonstrated versus other deep learning techniques applied to the same dataset, proving effectiveness with few spectral bands. Some results were lower than a similar application with more bands, but they also underscore the adaptability and potential of the ViT to handle challenging datasets. Insights from hyperparameter optimization shows the ViT’s promise as a robust tool for tumor identification, paving the way for integration into real-time clinical workflows and advancing precision medicine. |
| format | Article |
| id | doaj-art-a1ca84641300428b9904a2f13a9c42f8 |
| institution | Matheson Library |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-a1ca84641300428b9904a2f13a9c42f82025-07-17T23:00:47ZengIEEEIEEE Access2169-35362025-01-011312170412171910.1109/ACCESS.2025.358800111077150Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral BandsDomenico Ragusa0https://orcid.org/0009-0004-9482-1877Marco Gazzoni1https://orcid.org/0000-0003-4213-8270Emanuele Torti2https://orcid.org/0000-0001-8437-8227Elisa Marenzi3https://orcid.org/0000-0003-4537-5618Francesco Leporati4https://orcid.org/0000-0003-2901-4935Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, ItalyDepartment of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, ItalyDepartment of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, ItalyDepartment of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, ItalyDepartment of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, ItalyBrain tumors, particularly gliomas, pose a significant clinical challenge with rising incidence rates and high mortality. Artificial intelligence combined with Hyperspectral Imaging (HSI) offers promising tools to improve surgical precision and patients’ outcomes. HSI offers unique advantages, including non-invasiveness and detailed spectral data, for enhanced tumor tissue differentiation. This study tailored the Vision Transformer (ViT) with techniques from remote sensing to segment nineteen spectral images of low- and high-grade gliomas with limited spectral bands. The choice of the ViT was motivated by its attention mechanism, enabling fine-grained distinction of subtle details. Segmentation focused on four classes: healthy tissue, tumor tissue, blood vessels and dura mater. A careful hyperparameter optimization was performed, resulting in the selection of two models based on a defined quality index, which were evaluated using three experimental methodologies, achieving up to <inline-formula> <tex-math notation="LaTeX">$98.24\pm 2.50$ </tex-math></inline-formula>% average Overall Accuracy (OACC) and <inline-formula> <tex-math notation="LaTeX">$99.61\pm 0.66$ </tex-math></inline-formula>% average Area Under the Curve (AUC) in intra-patient classification. For inter-patient classification, the models achieved an average OACC up to <inline-formula> <tex-math notation="LaTeX">$53.56\pm 24.91$ </tex-math></inline-formula>% and an average AUC score up to <inline-formula> <tex-math notation="LaTeX">$79.27\pm 10.43$ </tex-math></inline-formula>%, highlighting areas of improvement. Comparable or improved performance was demonstrated versus other deep learning techniques applied to the same dataset, proving effectiveness with few spectral bands. Some results were lower than a similar application with more bands, but they also underscore the adaptability and potential of the ViT to handle challenging datasets. Insights from hyperparameter optimization shows the ViT’s promise as a robust tool for tumor identification, paving the way for integration into real-time clinical workflows and advancing precision medicine.https://ieeexplore.ieee.org/document/11077150/Vision transformerhyperspectral imagingglioblastoma detectionhyperparameter optimizationdeep learningmedical image analysis |
| spellingShingle | Domenico Ragusa Marco Gazzoni Emanuele Torti Elisa Marenzi Francesco Leporati Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands IEEE Access Vision transformer hyperspectral imaging glioblastoma detection hyperparameter optimization deep learning medical image analysis |
| title | Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands |
| title_full | Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands |
| title_fullStr | Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands |
| title_full_unstemmed | Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands |
| title_short | Vision Transformer for Brain Tumor Detection Using Hyperspectral Images With Reduced Spectral Bands |
| title_sort | vision transformer for brain tumor detection using hyperspectral images with reduced spectral bands |
| topic | Vision transformer hyperspectral imaging glioblastoma detection hyperparameter optimization deep learning medical image analysis |
| url | https://ieeexplore.ieee.org/document/11077150/ |
| work_keys_str_mv | AT domenicoragusa visiontransformerforbraintumordetectionusinghyperspectralimageswithreducedspectralbands AT marcogazzoni visiontransformerforbraintumordetectionusinghyperspectralimageswithreducedspectralbands AT emanueletorti visiontransformerforbraintumordetectionusinghyperspectralimageswithreducedspectralbands AT elisamarenzi visiontransformerforbraintumordetectionusinghyperspectralimageswithreducedspectralbands AT francescoleporati visiontransformerforbraintumordetectionusinghyperspectralimageswithreducedspectralbands |