Hyperspectral Imaging for Subcutaneous Vein Detection

Subcutaneous vein detection is critical in medical procedures like venipuncture and catheter placement. This PhD thesis introduces comprehensive research on Hyperspectral Imaging (HSI)-based vein detec tion. A Hyperspectral (HS) image dataset, called HyperVein, for vein detection is presented. The effectiveness of several state-of-the-art dimensionality reduction techniques, including Principal Com ponent Analysis (PCA), Folded Principal Component Analysis (FPCA), and Ward’s Linkage Strategy using Mutual Information (WaLuMI), along with Support Vector Machine (SVM) classification for vein detection, is investigated using the HyperVein dataset. Results show that FPCA-based methods generate more accurate results than WaLuMI and PCA-based methods. An effective dimensionality reduction method for vein detection from HS images, called Inter Band Correlation and Clustering (ICC), was developed. The proposed ICC method normalizes each spectral band, computes a correlation matrix across normalized HS images, and uses a subset of the correlation matrix’s eigenvectors to create a feature space by projecting the input HS image data onto the eigenvectors. Clustering is then applied to the resulting feature space coefficients to map them into a more effective feature space, generating the dimensionality reduced representation of the input HS image. The reduced HS image and SVM classification algorithm were used for vein detection. Experimental results show that the proposed method outperforms existing methods. A Convolutional Neural Network (CNN)-based method for vein detection in HS images of human hands is proposed. It applies PCA, FPCA, WaLuMI and ICC_k-means and ICC_spectral dimensional ity reduction techniques to extract HS image features. A 3D-CNN model was trained on dimensionality reduced HSI data to accurately identify vein pixel locations. Experimental results demonstrate that the ICC_k-means method outperforms PCA and FPCA.