Hyperspectral Tissue Image Segmentation Using Semi-Supervised NMF and Hierarchical Clustering.

Area of Science:

• Biomedical optics
• Computational pathology
• Spectroscopy

Background:

• Hyperspectral imaging (HSI) in the mid-infrared (mid-IR) range offers sub-cellular resolution for tissue analysis.
• Analyzing mid-IR spectra of cellular components is key for disease state assessment.
• Pixel classification in unstained HSI images is challenging due to noisy spectral data and limited annotations.

Purpose of the Study:

• To develop a hyperspectral imaging pixel classification method for unstained tissue.
• To address challenges in spectral dimension reduction and hierarchical pixel clustering.
• To improve the accuracy of classifying sub-cellular components for disease analysis.

Main Methods:

• Developed a semi-supervised learning approach for HSI pixel classification.
• Integrated spectral dimension reduction and hierarchical pixel clustering.
• Utilized both labeled and unlabeled pixels for training.

Main Results:

• Achieved an F1 score of 71.18% on twofold cross-validation across 20 tissue images.
• The method effectively accounts for spectral variations within cell types.
• Demonstrated straightforward disease classification post-segmentation.

Conclusions:

• The proposed semi-supervised HSI pixel classification method enhances sub-cellular component identification.
• This approach overcomes limitations of traditional supervised methods for noisy spectral data.
• HSI shows significant potential for disease classification with accurate image segmentation.