Magnetic Resonance (MR) images can be considered as multispectral images so that MR imaging can be processed by
multispectral imaging techniques such as maximum likelihood classification. Unfortunately, most multispectral imaging
techniques are not particularly designed for target detection. On the other hand, hyperspectral imaging is primarily
developed to address subpixel detection, mixed pixel classification for which multispectral imaging is generally not
effective. This paper takes advantages of hyperspectral imaging techniques to develop target detection algorithms to find
lesions in MR brain images. Since MR images are collected by only three image sequences, T1, T2 and PD, if a
hyperspectral imaging technique is used to process MR images it suffers from the issue of insufficient dimensionality.
To address this issue, two approaches to nonlinear dimensionality expansion are proposed, nonlinear correlation
expansion and nonlinear band ratio expansion. Once dimensionality is expanded hyperspectral imaging algorithms are
readily applied. The hyperspectral detection algorithm to be investigated for lesion detection in MR brain is the well-known
subpixel target detection algorithm, called Constrained Energy Minimization (CEM). In order to demonstrate the
effectiveness of proposed CEM in lesion detection, synthetic images provided by BrainWeb are used for experiments.
Anomaly detection is one of most fundamental tasks in hyperspectral data exploitation. Since anomalies are generally unknown and unexpected, their detection must be carried out without prior knowledge. Most importantly, when anomalies are weak and moving as time goes on real time processing of anomaly detection becomes immense in detecting these anomalous targets. Due to hyperspectral sensor design two major formats are used for data acquisition in real-time. One is real-time sample processing which collects data in two different fashions, Band-interleaved by Pixel/Sample (BIP/BIS) sample-by-sample and Band-Interleaved-by-Line (BIL) line-by-line. Another is real-time band processing which follows the Band Sequential (BSQ) format to collect the data. Recently, anomaly detection using both BIP/BIS and BSQ has been reported in the literature. Since a hyperspectral imaging sensor generally collects the data in a push-broom manner, the BIL format is preferred to BIP/BIS. But it is interesting to note that AD using BIL has not been explored and investigated in the past mainly because it was expected that AD using BIL may perform similarly to AD using BIP/BIS. This paper shows otherwise due to the fact that the covariance/correlation matrix used by an anomaly detector has significant impact on the detectability of anomalies. It has been shown that anomaly detection is heavily determined by the ratio of anomalies to be detected to the image size that forms the covariance/correlation matrix. So, when AD using BIL is implemented, the information of covariance/correlation matrix provided BIL is different from that provided by BIP/BIS. As a result, it is anticipated that AD using BIL may result different performance from AD using BIP/BIS.
Convexity is a major concept used to design and develop endmember finding algorithms (EFAs). For abundance unconstrained techniques, Pixel Purity Index (PPI) and Automatic Target Generation Process (ATGP) which use Orthogonal Projection (OP) as a criterion, are commonly used method. For abundance partially constrained techniques, Convex Cone Analysis is generally preferred which makes use of convex cones to impose Abundance Non-negativity Constraint (ANC). For abundance fully constrained N-FINDR and Simplex Growing Algorithm (SGA) are most popular methods which use simplex volume as a criterion to impose ANC and Abundance Sum-to-one Constraint (ASC). This paper analyze an issue encountered in volume calculation with a hyperplane introduced to illustrate an idea of bounded convex cone. Geometric Convex Cone Volume Analysis (GCCVA) projects the boundary vectors of a convex cone orthogonally on a hyperplane to reduce the effect of background signatures and a geometric volume approach is applied to address the issue arose from calculating volume and further improve the performance of convex cone-based EFAs.
This paper develops a completely new technology,) from a hyperspectral imaging perspective, called Hyperspectral Vital Sign Signal Analysis (HyVSSA. A hyperspectral image is generally acquired by hundreds of contiguous spectral bands, each of which is an optical sensor specified by a particular wavelength. In medical application, we can consider a patient with different vital sign signals as a pixel vector in hyperspectral image and each vital sign signal as a particular band. In light of this interpretation, a revolutionary concept is developed, which translates medical data to hyperspectral data in such a way that hyperspectral technology can be readily applied to medical data analysis. One of most useful techniques in hyperspectral data processing is, Anomaly Detection (AD) which in this medical application is used to predict outcomes such as transfusion, length of stay (LOS) and mortality using various vital signs. This study compared transfusion prediction performance of Anomaly Detection (AD) and Logistic Regression (LR).
OSP has been used widely in detection and abundance estimation for about twenty years. But it can’t
apply nonnegative and sum-to-one constraints when being used as an abundance estimator. Fully
constrained least square algorithm does this well, but its time cost increases greatly as the number of
endmembers grows. There are some tries for unmixing spectral under fully constraints from different
aspects recently. Here in this paper, a new fully constrained unmixing algorithm is prompted based on
orthogonal projection process, where a nearest projected point is defined onto the simplex constructed
by endmembers. It is much easier, and it is faster than FCLS with the mostly same unmixing results. It
is also compared with other two constrained unmixing algorithms, which shows its effectiveness too.
Pixel Purity Index (PPI) is a very popular endmember finding algorithm due to its availability in ENVI software. According to the band sequential (BSQ) format of data acquisition this paper introduces a new concept of executing PPI band-by-band in a progressive manner. It is called progressive band processing of PPI (PBP-PPI) which allows users to process PPI band by band without waiting for full bands of data information acquired. To accomplish this goal PPI must be capable of calculating and updating PPI counts of data samples band by band. Furthermore, progressive-band-processing progressive PPI (PBP-P-PPI) and progressive-band-processing causal PPI (PBP-C-PPI) are proposed to address the issues that the number of skewers is undefined and only partial pixels are available correspondingly. Many benefits can be gained from PBP-PPI, for example, providing progressive profiles of PPI counts of data samples as more bands are included for data processing, finding crucial bands according to progressive changes in PPI counts.
Progressive band processing (PBP) processes data band by band according to the Band SeQuential (BSQ) format acquired by a hyperspectral imaging sensor. It can be implemented in real time in the sense that data processing can be performed whenever bands are available without waiting for data completely collected. This is particularly important for satellite communication when data download is limited by bandwidth and transmission. This paper presents a new concept of processing a well-known technique, Orthogonal Subspace Projection (OSP) band by band, to be called PBPOSP. Several benefits can be gained by PBP-OSP. One is band processing capability which allows different receiving ends to process data whenever bands are available. Second, it enables users to identify significant bands during data processing. Third, unlike band selection which requires knowing the number of bands needed to be selected or band prioritization PBP-OSP can process arbitrary bands in real time with no need of such prior knowledge. Most importantly, PBP can locate and identify which bands are significant for data processing in a progressive manner. Such progressive profile resulting from PBP-OSP is the best advantage that PBP-OSP can offer and cannot be accomplished by any other OSP-like operators.
Using maximal simplex volume as an optimal criterion for finding endmembers is a common approach and has been widely studied in the literature. Interestingly, very little work has been reported on how simplex volume is calculated. It turns out that the issue of calculating simplex volume is much more complicated and involved than what we may think. This paper investigates this issue from two different aspects, geometric structure and eigen-analysis. The geometric structure is derived from its simplex structure whose volume can be calculated by multiplying its base with its height. On the other hand, eigen-analysis takes advantage of the Cayley-Menger determinant to calculate the simplex volume. The major issue of this approach is that when the matrix is ill-rank where determinant is desired. To deal with this problem two methods are generally considered. One is to perform data dimensionality reduction to make the matrix to be of full rank. The drawback of this method is that the original volume has been shrunk and the found volume of a dimensionality-reduced simplex is not the real original simplex volume. Another is to use singular value decomposition (SVD) to find singular values for calculating simplex volume. The dilemma of this method is its instability in numerical calculations. This paper explores all of these three methods in simplex volume calculation. Experimental results show that geometric structure-based method yields the most reliable simplex volume.