Paul P. K. Lee

Recent developments in tissue-type imaging (TTI) for planning and monitoring treatment of prostate cancer.

Because current methods of imaging prostate cancer are inadequate, biopsies cannot be effectively guided and treatment cannot be effectively planned and targeted. Therefore, our research is aimed at ultrasonically characterizing cancerous prostate tissue so that we can image it more effectively and thereby provide improved means of detecting, treating and monitoring prostate cancer. We base our characterization methods on spectrum analysis of radiofrequency (rf) echo signals combined with clinical variables such as prostate-specific antigen (PSA). Tissue typing using these parameters is performed by artificial neural networks. We employed and evaluated different approaches to data partitioning into training, validation, and test sets and different neural network configuration options. In this manner, we sought to determine what neural network configuration is optimal for these data and also to assess possible bias that might exist due to correlations among different data entries among the data for a given patient. The classification efficacy of each neural network configuration and data-partitioning method was measured using relative-operating-characteristic (ROC) methods. Neural network classification based on spectral parameters combined with clinical data generally produced ROC-curve areas of 0.80 compared to curve areas of 0.64 for conventional transrectal ultrasound imaging combined with clinical data. We then used the optimal neural network configuration to generate lookup tables that translate local spectral parameter values and global clinical-variable values into pixel values in tissue-type images (TTIs). TTIs continue to show cancerous regions successfully, and may prove to be particularly useful clinically in combination with other ultrasonic and nonultrasonic methods, e.g., magnetic-resonance spectroscopy.

Ultrasonic Detection of Myocardial Infarction by Amplitude Analysis

Myocardial infarctions were produced in dogs by occluding the left anterior descending artery; the dogs were killed at varying times, from 30 minutes to 8 days. Prior to sacrifice, Thioflavin S was injected intravenously. The excised heart was scanned by a B-scanner interfaced with a computer that permitted quantification of signal amplitude. The heart was sectioned, photographed, and studied pathologically. Infarcted myocardium showed high ultrasonic reflectivity with average amplitudes 1.4--2.6 times that of normal muscle. Perfusion-histomorphologic evidence of infarction correlated best in infarcts of 24 hours or less; older infarcts were generally underestimated. Tissue changes, sources of false-positive and false-negative findings, signal processing, and potential clinical utility are discussed.

Frequency‐dependent angle scattering of ultrasound by liver

Ultrasonic scattering by one specimen of normal pig liver and two specimens of abnormal human liver has been measured as a function of angle and frequency at each angular position of transmit and receive transducers which were rotated in equal and opposite directions about the axis of a cylindrically shaped tissue sample. Mean data values were determined by averaging points at like frequencies and angular positions in scans made at intervals along the axis of the specimen. Sections of the specimens were obtained throughout its length and stained to emphasize structures containing collagen and connective tissue shown by acoustic microscopy to produce sound speed variations which are known to cause scattering. Using the stained sections as diffraction screens, calculations of scattering were carried out via Fourier transforms on a scale comparable to that in the acoustic studies for comparison with the acoustic data. In all specimens studied, mean acoustic scattering exhibited a general decrease in level with increasing scattering angle. The rate of decrease in acoustic scattering with angle agreed qualitatively with the calculations of average angular falloff of scattering by the stained sections. Lower rates of decrease were observed in tissues with more closely spaced collagen containing structures. Although the data derived from the few substantially different specimens studied is not intended to be representative of all liver, the results show a qualitative correspondence between acoustic scattering and visual appearance which depends on tissue components and their arrangement.

Angle Scan and Frequency-Swept Ultrasonic Scattering Characterization of Tissue

Ultrasonic wave interference has been applied to characterize tissue by measuring scattered wave intensity as a function of frequency and angle. The measurements were made using a computer-based system for collection, processing, and display of data. Model studies consisted of data collection from deterministic targets of regular arrays of nylon filaments and random targets made from small spherical particles of cross-linked dextran in suspension. Angle and frequency scans have been made on post mortem human liver specimens. The model studies demonstrate that regular scatterer spacing can be inferred from measured diffraction data by Fourier inversion, and that scattering differences can be observed from particles of different sizes. Scattering from liver indicates the importance of off-axis scattering which may be related to scatterer size.

Swept‐frequency diffraction of ultrasound by cylinders and arrays

Diffraction of ultrasound in water by single cylinders and arrays of equally spaced cylinders is studied experimentally using a swept‐frequency technique and the results are compared with theoretical predictions. Scattering by single metallic cylinders is computed from a model which includes the effect of reverberations within the cylinder. A one‐dimensional, single‐scattering approximation is used for data reduction in the case of the regular array where Fourier inversion of the data is used to determine the scatterer spacing of the array. Swept‐frequency measurements were made in a fixed geometry using long pulses of a sinusoidal signal whose center frequency was essentially constant during the pulse interval but was varied over the total sequence of pulses. The observed scattering spectrum of a single metallic cylinder agrees closely with the calculated prediction. The Fourier transformation spectra of the array scattering data shows good agreement with the known scatterer spacing even when the data we...

Non-Uniformly Tiled CMOS Image Sensors for Efficient On-Chip Image Compression

We present a complementary metal-oxide semiconductor (CMOS) image sensor with non-uniform pixel placement that enables a highly efficient calculation of the discrete cosine transform (DCT), which is the most mathematically intensive step of an image compression algorithm. This technique is based on the arithmetic Fourier transform (AFT), which has been shown to be five times more computationally efficient than DCT derivation methods commonly used. In this paper, the focus is on the basic theory and algorithm as well as the sensitivity of the method to image sensor fixed pattern noise (FPN). The architecture and circuits have been implemented in a conventional CMOS process. The method has been demonstrated in the current prototype and results that enable an assessment of the sensitivity to FPN have been obtained.

Prostate-cancer imaging using machine-learning classifiers: Potential value for guiding biopsies, targeting therapy, and monitoring treatment

Prostate cancer (PCa) remains a major health concern in many countries. However, it cannot be imaged reliably by any commonly used imaging modality. Therefore, needle biopsies and treatments cannot be targeted to suspicious regions. Our objective is to develop and test an ultrasonic method based on spectrum analysis of radio-frequency (RF) ultrasound echo signals and on classification using current machine-learning tools for reliably imaging PCa and thereby guiding biopsies, targeting therapy, and eventually, monitoring treatment of PCa. RF data were acquired in the biopsy plane of 617 prostate biopsy cores obtained from 64 suspected prostate-cancer (PCa) patients. For each patient, clinical data such as PSA level also were recorded. A level of suspicion (LOS) was assigned based on the conventional image. Spectral computations were performed on acquired RF data in a region of interest that spatially matched the tissue-sampling location. Four non-linear classifiers were trained from these data using biopsy results as the gold standard: multi-layer-perceptron artificial neural networks (ANNs), logitboost algorithms (LBAs), support-vector machines (SVMs), and stacked, restricted Boltzmann machines (S-RBMs). Cross-validation methods were employed to obtain tissue-category scores. Areas under ROC curves (AUCs) were used to assess classifier performance in comparison with LOS-based performance. AUCs for the ANN, LBA, SVM, and RBM respectively were 0.84 ± 0.02, 0.87 ± 0.04, 0.89 ± 0.04, and 0.91 ± 0.04. In comparison, the LOS-based AUC was 0.64 ± 0.03. Tissue-type images (TTIs) based on these methods revealed cancerous foci that subsequently were identified histologically, but were undetected prior to prostatectomy pathology. The ultrasonic imaging methods described here show significant potential for achieving needed reliability. A clinically significant beneficial reduction in false-negative biopsy procedures would be possible if TTIs were used to guide biopsies. Benefits also would result from using TTIs to target focal treatment and reduce toxic side effects. Potentially, TTIs also could be used to assess tissue changes over time for active surveillance and therapy monitoring.

Assembling 3D histology volumes from sections of cancerous lymph nodes to match 3D high-frequency quantitative ultrasound images

High-Frequency Quantitative Ultrasound (HFQUS) imaging methods are under investigation to evaluate their ability to detect small metastases (< 2 mm) in lymph nodes freshly dissected from cancer patients. To assess the performance of these methods, 3D HFQUS must be compared to gold-standard histologic images. Histologic images have to be assembled to form volumetric histologic information. This study addresses this issue. The acquisition of high-frequency ultrasound (HFU) data with a 26-MHz center-frequency transducer and histologic preparation are described. Dissected nodes were longitudinally cut in half and pairs of histologic sections separated by 65 µm, for nodes < 5 mm, or 115 µm, for nodes > 5 mm, were photographed. Then a fully automatic method to assemble and orient a 3D histologic volume from a set of 2D images was developed and applied. Identification of the histology sections on each slide relies on a parametric shape modeling of the histologic sections with ellipses. Then a set of rigid transformations were estimated and applied to construct volumetric histologic data. The method was visually evaluated on a set of 50 lymph nodes and is valuable for comparing histologic data to HFQUS estimates in 3D.

Total ionizing dose gamma and proton radiation testing on a COTS interline CCD with microlens

Commercially fabricated interline CCDs were total ionizing dose (TID) tested with gamma (/spl gamma/) and proton (p/sup +/) radiation. Materials used for microlenses applied to the array surface were also irradiated on fused silica substrates. Device and material performance degradations are presented.

Classification of prostate tissue using neural networks

This paper presents the development of a method for distinguishing between cancerous and non-cancerous tissue of the prostate based on spectrum analysis of ultrasound radio-frequency echo-signals. The classification of prostate tissue is performed using the neural network technology. At this stage, the image segmentation and feature extraction processes are well established. Hence, we focus here on the method of selection, training and evaluation of a neural network classifier. Results from well-known statistical methods are provided for comparison. Our results demonstrate the viability both of our approach and of this technology when operating on real-world data.