Search

In this work, we developed an algorithm for the integrated analysis of multiple high-dimensional data matrices based on sparse rank-one matrix approximations. The algorithm approximates multiple data matrices with rank one outer products composed of sparse left singular-vectors that are unique to each matrix and a right singular-vector that is shared by all of the data matrices. The right-singular vector represents a signal we wish to detect in the row-space of each matrix. The non-zero components of the resulting left-singular vectors identify rows of each matrix that in aggregate provide a sparse linear representation of the shared right-singular vector. This sparse representation facilitates downstream interpretation and validation of the resulting model based on the rows selected from each matrix. False discovery rate is used to select an appropriate l1 penalty parameter that imposes sparsity on the left singular-vector but not the common right singular-vector of the joint approximation. Since a given multi-modal data set (MMDS) may contain multiple signals of interest the algorithm is iteratively applied to the residualized version of original data to sequentially capture and model each distinct signal in terms of rows from the different matrices. We show that the algorithm outperforms standard singular value decomposition over a wide range of simulation scenarios in terms of detection accuracy. Analysis of real data for ovarian and liver cancer resulted in compact gene expression signatures that were predictive of clinical outcomes and highly enriched for cancer related biology.

PET using fluorine-18 fluorocholine ((18)F-fluorocholine) may detect malignancies that involve altered choline metabolism. While (18)F-fluorocholine PET/CT has shown greater sensitivity for detecting hepatocellular carcinoma (HCC) than (18)F-fluoro-D-deoxyglucose (FDG) PET/CT, it is not known whether it can also detect intrahepatic cholangiocarcinoma (ICC), a less common form of primary liver cancer. Clinical, radiographic, and histopathologic data from 5 patients with ICC and 23 patients with HCC from a diagnostic trial of liver (18)F-fluorocholine PET/CT imaging were analyzed to preliminarily evaluate (18)F-fluorocholine PET/CT for ICC. Imaging was correlated with whole-genome expression profiling to identify molecular pathways associated with tumor phenotypes. On PET/CT, all ICC tumors demonstrated low (18)F-fluorocholine uptake with a significantly lower tumor to mean background uptake ratio than HCC tumors (0.69 vs. 1.64, p0.0001), but no corresponding significant difference in liver parenchyma uptake of (18)F-fluorocholine between ICC and HCC patients (8.0 vs. 7.7, p = 0.74). Two ICC patients demonstrated increased tumor metabolism on FDG PET/CT, while immunohistochemical analysis of ICC tumors revealed overexpression of glucose transporter 1 (GLUT-1) and hexokinase indicating a hyper-glycolytic phenotype. Gene expression analysis revealed down-regulation of farnesoid-X-receptor and other lipid pathways in ICC relative to HCC, and up-regulation of glycolytic pathways and GLUT-1 by HIF1α. These results imply limited utility of (18)F-fluorocholine in ICC, however, significant metabolic differences between ICC, HCC, and parenchymal liver tissue may still provide clues about the underlying liver pathology. Gene and protein expression analysis support hyperglycolysis as a more dominant metabolic trait of ICC.

Objective: The aim of this study was to initiate neural net construction for the detection of cervical intraepithelial neoplasia by fluorescence imaging. Study Design: Thirty-three women with abnormal Papanicolaou smears underwent fluorescence imaging during colposcopy. With the use of >4000 training pixels and >1000 test pixels, intrapatient nets were constructed from the spectral data of 17 women. An interpatient net that discriminated between cervical intraepithelial neoplasia 1 and normal tissue classes among patients was constructed with the use of >2300 training pixels and >2000 test pixels from 12 women. Average correct classification rates were determined. Sensitivities, specificities, and positive and negative predictive values for cervical intraepithelial neoplasia grade 1 and normal tissue classes were calculated. Extrapolated false-color cervical images were generated. Results: Average correct classification rates were 96.5% for the intrapatient nets and 97.5% for the interpatient net. The sensitivity, specificity, and positive and negative predictive values for cervical intraepithelial neoplasia grade 1 were 98.2%, 98.9%, 71.4%, and 99.9%, respectively. Conclusion: Initial results suggest that neural nets that are constructed from fluorescence imaging spectra may offer a potential method for the detection of cervical intraepithelial neoplasia. (Am J Obstet Gynecol 2002;187:398-402.)

Principal component analysis (PCA) in the wavelet domain provides powerful new features for the non-invasive detection of cervical intraepithelial neoplasia (CIN) using fluorescence imaging spectroscopy. These features are known as principal wavelet components (PWCs). The multiscale structure of the fluorescence spectrum for each pixel of the hyperspectral data cube is extracted using the continuous wavelet transform. PCA is then used to compress and denoise the wavelet representation for presentation to a feed- forward neural network for tissue classification. Using PWC features as inputs to a 5-class NN resulted in average correct classification rates of 95% over five cervical tissue classes corresponding to low-grade dysplasia, squamous, columnar, metaplasia plus a fifth class for other unspecified tissue types, blood and mucus. A 2-class NN was also trained to discriminate between CIN1 and normal tissue with sensitivity and specificity of 98% and 99%, respectively. All performance assessments were based on test data from a set of patients not seen during NN training. Trained neural classifiers were used to `compress' and transform 3D hyperspectral data cubes into 2D color-coded images that accurately mapped the spatial distribution of both normal and dysplastic tissue over the surface of the entire cervix.© (2001) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Principal component analysis (PCA) in the wavelet domain provides powerful features for underwater object recognition applications. The multiresolution analysis of the Morlet wavelet transform (MWT) is used to pre-process echo returns from targets ensonified by biologically motivated broadband signal. PCA is then used to compress and denoise the resulting time-scale signal representation for presentation to a hierarchical neural network for object classification. Wavelet/PCA features combined with multi-aspect data fusion and neural networks have resulted in impressive underwater object recognition performance using backscatter data generated by simulate dolphin echolocation clicks and bat- like linear frequency modulated upsweeps. For example, wavelet/PCA features extracted from LFM echo returns have resulted in correct classification rates of 98.6 percent over a six target suite, which includes two mine simulators and four clutter objects. For the same data, ROC analysis of the two-class mine-like versus non-mine-like problem resulted in a probability of detection of 0.981 and a probability of false alarm of 0.032 at the 'optimal' operating point. The wavelet/PCA feature extraction algorithm is currently being implemented in VLSI for use in small, unmanned underwater vehicles designed for mine- hunting operations in shallow water environments.© (1999) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Signal features based on multiresolution short-time Fourier transforms (STFT) and the Morlet wavelet transform (MWT) have been developed to classify echo returns from targets ensonified by simulated dolphin echolocation clicks. Spectrogram features are obtained at different scales of resolution using analysis windows of different sizes. A method of compressing the highly redundant time-scale representations provided by the MWT has been developed based on multiscale edge analysis (MSEA) of wavelet local maxima. Neural networks are used to evaluate the efficacy of the various feature sets for target recognition. Hierarchical neural networks are used to combine different feature sets for improved classification performance.