abstract

• Hyperspectral imaging (HSI) is a versatile modality that can provide a noninvasive tool for cancer detection, image-guided surgery, tissue classification, among other applications. In this work, we demonstrate the integration of hyperspectral imaging, spectral unmixing, and machine learning models to accurately classify tumor and non-tumor tissue from histopathology samples. The studied database contains 494 images from 13 different patients diagnosed with glioblastoma. Our approach is based on identifying characteristic spectral signatures for each hyperspectral image by spectral unmixing, and using them as an input feature vector for machine learning models: support vector machine, random forest, and a voting ensemble. The resulting average accuracy in our evaluation of four folds was 89.4%25 that improves the reference value of 85.5%25, which was the best performance in the state-of-the-art. In this way, our proposed methodology shows promising results to assist in pathological analysis and provide support to healthcare professionals. © 2024, The Author(s), under exclusive license to Springer Nature Switzerland AG.

authors

• Campos Delgado David Leonardo
• Cruz-Guerrero I.A.

publication date

• 2024-01-01

published in

• IFMBE Proceedings  Proceedings

keywords

• Binary Classification; Glioblastoma; Hyperspectral imaging; Machine Learning; Spectral Unmixing Image classification; Learning algorithms; Learning systems; Patient monitoring; Support vector machines; Tumors; Binary classification; Glioblastomas; HyperSpectral; Machine learning algorithms; Machine learning models; Machine-learning; Noninvasive tools; Spectral unmixing; Tissue classification; Tumour tissue; Hyperspectral imaging

Digital Object Identifier (DOI)

• 10.1007/978-3-031-46933-6_11

PubMed ID

start page

• 100

end page

• 111

volume

• 96

issue