A spectral dissimilarity constrained nonnegative matrix factorization based cancer screening algorithm from hyperspectral fluorescence images
- Publication Type:
- Conference Proceeding
- Citation:
- ICCH 2012 Proceedings - International Conference on Computerized Healthcare, 2012, pp. 112 - 119
- Issue Date:
- 2012-01-01
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Bioluminescence from living body can help screen cancers without penetrating the inside of living body. Hyperspectral imaging technique is a novel way to obtain physical meaningful signatures, providing very fine spectral resolution, that can be very used in distinguishing different kinds of materials, and have been widely used in remote sensing field. Fluorescence imaging has proved effective in monitoring probable cancer cells. Recent work has made great progress on the hyperspectral fluorescence imaging techniques, which makes the elaborate spectral observation of cancer areas possible. So how to propose the proper hyperspectral image processing methods to handle the hyperspectral medical images is of practical importance. Cancer cells would be distinguishable with normal ones when the living body is injected with fluorescence, which helps organs inside the living body emit lights, and then the signals can be catched by the passive imaging sensor. Spectral unmixing technique in hyperspectral remote sensing has been introduced to detect the probable cancer areas. However, since the cancer areas are small and the normal areas and the cancer ares may not pure pixels so that the predefined endmembers would not available. In this case, the classic blind signals separation methods are applicable. Considering the spectral dissimilarity between cancer and normal cells, a novel spectral dissimilarity constrained based NMF method is proposed in this paper for cancer screening from fluorescence hyperspectral images. Experiments evaluate the performance of variable NMF based method and our proposed spectral dissimilarity based NMF methods: 1) The NMF methods do perform well in detect the cancer areas inside the living body; 2) The spectral dissimilarity constrained NMF present more accurate cancer areas; 3) The spectral dissimilarity constraint presents better performance in different SNR and different purities of the mixing endmembers. © 2012 IEEE.
Please use this identifier to cite or link to this item: