Development of automated image analysis tools for verification of radiotherapy field accuracy with an electronic portal imaging device

The successful management of cancer with radiation relies on the accurate deposition of a prescribed dose to a prescribed anatomical volume within the patient. Treatment set-up errors are inevitable because the alignment of field shaping devices with the patient must be repeated daily up to eighty times during the course of a fractionated radiotherapy treatment. With the invention of electronic portal imaging devices (EPIDs), patient's portal images can be visualized daily in real-time after only a small fraction of the radiation dose has been delivered to each treatment field. However, the accuracy of human visual evaluation of low-contrast portal images has been found to be inadequate. The goal of this research is to develop automated image analysis tools to detect both treatment field shape errors and patient anatomy placement errors with an EPID. A moments method has been developed to align treatment field images to compensate for lack of repositioning precision of the image detector. A figure of merit has also been established to verify the shape and rotation of the treatment fields. Following proper alignment of treatment field boundaries, a cross-correlation method has been developed to detect shifts of the patient's anatomy relative to the treatment field boundary. Phantom studies showed that the moments method aligned the radiation fields to within 0.5mm of translation and 0.5$\sp\circ$ of rotation and that the cross-correlation method aligned anatomical structures inside the radiation field to within 1 mm of translation and 1$\sp\circ$ of rotation. A new procedure of generating and using digitally reconstructed radiographs (DRRs) at megavoltage energies as reference images was also investigated. The procedure allowed a direct comparison between a designed treatment portal and the actual patient setup positions detected by an EPID. Phantom studies confirmed the feasibility of the methodology. Both the moments method and the cross-correlation technique were implemented within an experimental radiotherapy picture archival and communication system (RT-PACS) and were used clinically to evaluate the setup variability of two groups of cancer patients treated with and without an alpha-cradle immobilization aid. The tools developed in this project have proven to be very effective and have played an important role in detecting patient alignment errors and field-shape errors in treatment fields formed by a multileaf collimator (MLC). ^

Comparison of multiple comparison methods for identifying differential gene expression in simulated and real papillary thyroid cancer microarray data

The difficulty of detecting differential gene expression in microarray data has existed for many years. Several correction procedures try to avoid the family-wise error rate in multiple comparison process, including the Bonferroni and Sidak single-step p-value adjustments, Holm's step-down correction method, and Benjamini and Hochberg's false discovery rate (FDR) correction procedure. Each multiple comparison technique has its advantages and weaknesses. We studied each multiple comparison method through numerical studies (simulations) and applied the methods to the real exploratory DNA microarray data, which detect of molecular signatures in papillary thyroid cancer (PTC) patients. According to our results of simulation studies, Benjamini and Hochberg step-up FDR controlling procedure is the best process among these multiple comparison methods and we discovered 1277 potential biomarkers among 54675 probe sets after applying the Benjamini and Hochberg's method to PTC microarray data.^

A RANDOMIZED TRIAL OF BILEVEL POSITIVE AIRWAY PRESSURE DEVICE AND HIGH FLOW OXYGEN FOR PERSISTENT DYSPNEA IN

Background: Dyspnea is a common and distressing symptom among patients with advanced cancer. The role of bilevel positive airway pressure (BIPAP) and Vapotherm in the relief of dyspnea have not been well defined. We aimed to determine and to compare the efficacy of BIPAP and VapoTherm for cancer related dyspnea. Methods: In this randomized, open-label, crossover study, we randomly assigned advanced cancer patients with persistent dyspnea >=3/10 to either Vapotherm for 2 hours followed by BiPAP for 2 hours, or BiPAP followed by Vaptherm. A variable washout period was instituted between interventions. The primary end point was change in numeric rating scale before and after each intervention. We planned to enroll 50 patients in total. Results: Among the 803 patients screened over the last 8 months, 62 (26%) were eligible, and 16 (2%) were enrolled so far. Five patients completed the entire study successfully, 4 discontinued the study prematurely due to prolonged relief of dyspnea, and 7 dropped out for various reasons, including inability to tolerate BiPAP (N=3), anxiety (N=2), fatigue (N=1) and pain requiring opioids (N=1). The median baseline numeric rating score for dyspnea was 7/10 (interquartile range (IQR) 5-8), and the median baseline Borg score was 4/10 (3-7). Interim analysis revealed that BiPAP was associated with a median change in numeric rating score of -3 (N=10, IQR -6.3 to -1, p=0.007) and modified Borg score of -1 (N=10, IQR -3 to 0.3, p=0.058), while Vapotherm was associated with a median change in numeric rating score of -2 (N=9, IQR -3 to -1, p=0.011) and modified Borg score of -2.5 (N=8, IQR -5.5 to -0.1, p=0.051). Among the 5 individuals who completed the entire study, 2 preferred Vapotherm, 2 favored BiPAP, and 1 liked both. The respiratory rate decreased and the oxygen saturation improved with both interventions. No significant toxicities were observed. Conclusions: We were successfully able to enroll patients onto this clinic trial. Our preliminary results suggest that BiPAP and Vapotherm are highly efficacious in providing relief for patients with persistent refractory dyspnea. A direct comparison of the two interventions will be done upon study completion. Further research is necessary to confirm our findings.

Using citation data to improve retrieval from MEDLINE.

OBJECTIVE: To determine whether algorithms developed for the World Wide Web can be applied to the biomedical literature in order to identify articles that are important as well as relevant. DESIGN AND MEASUREMENTS A direct comparison of eight algorithms: simple PubMed queries, clinical queries (sensitive and specific versions), vector cosine comparison, citation count, journal impact factor, PageRank, and machine learning based on polynomial support vector machines. The objective was to prioritize important articles, defined as being included in a pre-existing bibliography of important literature in surgical oncology. RESULTS Citation-based algorithms were more effective than noncitation-based algorithms at identifying important articles. The most effective strategies were simple citation count and PageRank, which on average identified over six important articles in the first 100 results compared to 0.85 for the best noncitation-based algorithm (p < 0.001). The authors saw similar differences between citation-based and noncitation-based algorithms at 10, 20, 50, 200, 500, and 1,000 results (p < 0.001). Citation lag affects performance of PageRank more than simple citation count. However, in spite of citation lag, citation-based algorithms remain more effective than noncitation-based algorithms. CONCLUSION Algorithms that have proved successful on the World Wide Web can be applied to biomedical information retrieval. Citation-based algorithms can help identify important articles within large sets of relevant results. Further studies are needed to determine whether citation-based algorithms can effectively meet actual user information needs.

Tracking the Flow of Information through the Hippocampal Formation in the Rat

The hippocampus receives input from upper levels of the association cortex and is implicated in many mnemonic processes, but the exact mechanisms by which it codes and stores information is an unresolved topic. This work examines the flow of information through the hippocampal formation while attempting to determine the computations that each of the hippocampal subfields performs in learning and memory. The formation, storage, and recall of hippocampal-dependent memories theoretically utilize an autoassociative attractor network that functions by implementing two competitive, yet complementary, processes. Pattern separation, hypothesized to occur in the dentate gyrus (DG), refers to the ability to decrease the similarity among incoming information by producing output patterns that overlap less than the inputs. In contrast, pattern completion, hypothesized to occur in the CA3 region, refers to the ability to reproduce a previously stored output pattern from a partial or degraded input pattern. Prior to addressing the functional role of the DG and CA3 subfields, the spatial firing properties of neurons in the dentate gyrus were examined. The principal cell of the dentate gyrus, the granule cell, has spatially selective place fields; however, the behavioral correlates of another excitatory cell, the mossy cell of the dentate polymorphic layer, are unknown. This report shows that putative mossy cells have spatially selective firing that consists of multiple fields similar to previously reported properties of granule cells. Other cells recorded from the DG had single place fields. Compared to cells with multiple fields, cells with single fields fired at a lower rate during sleep, were less likely to burst, and were more likely to be recorded simultaneously with a large population of neurons that were active during sleep and silent during behavior. These data suggest that single-field and multiple-field cells constitute at least two distinct cell classes in the DG. Based on these characteristics, we propose that putative mossy cells tend to fire in multiple, distinct locations in an environment, whereas putative granule cells tend to fire in single locations, similar to place fields of the CA1 and CA3 regions. Experimental evidence supporting the theories of pattern separation and pattern completion comes from both behavioral and electrophysiological tests. These studies specifically focused on the function of each subregion and made implicit assumptions about how environmental manipulations changed the representations encoded by the hippocampal inputs. However, the cell populations that provided these inputs were in most cases not directly examined. We conducted a series of studies to investigate the neural activity in the entorhinal cortex, dentate gyrus, and CA3 in the same experimental conditions, which allowed a direct comparison between the input and output representations. The results show that the dentate gyrus representation changes between the familiar and cue altered environments more than its input representations, whereas the CA3 representation changes less than its input representations. These findings are consistent with longstanding computational models proposing that (1) CA3 is an associative memory system performing pattern completion in order to recall previous memories from partial inputs, and (2) the dentate gyrus performs pattern separation to help store different memories in ways that reduce interference when the memories are subsequently recalled.

Gene transfer by viral vectors

To initiate our clinical trial for chemotherapy protection, I established the retroviral vector system for human MDR1 cDNA gene transfer. The human MDR1 cDNA continued to be expressed in the transduced bone marrow cells after four cohorts of serial transplants, 17 months after the initial transduction and transplant. In addition, we used this retroviral vector pVMDR1 to transduce human bone marrow and peripheral blood CD34$\sp+$ cells on stromal monolayer in the presence of hematopoietic growth factors. These data suggest that the retroviral vector pVMDR1 could modify hematopoietic precursor cells with a capacity for long-term self renewal. Thus, it may be possible to use the MDR1 retroviruses to confer chemotherapeutic protection on human normal hematopoietic precursor cells of ovarian and breast cancer patients in whom high doses of MDR drugs may be required to control the diseases.^ Another promising vector system is recombinant adeno-associated virus (rAAV) vector. An impediment to use rAAV vectors is that production of rAAV vectors for clinical use is extremely cumbersome and labor intensive. First I set up the rAAV vector system in our laboratory and then, I focused on studies related to the production of rAAV vectors for clinical use. By using a self-inactivating retroviral vector carrying a selection marker under the control of the CMV immediate early promoter and an AAV genome with the deletion of both ITRs, I have developed either a transient or a stable method to produce rAAV vectors. These methods involve infection only and can generate high-titer rAAV vectors (up to 2 x 10$\sp5$ cfu/ml of CVL) with much less work.^ Although recombinant adenoviral vectors hardly infect early hematopoietic precursor cells lacking $\alpha\sb v\beta\sb5$ or $\alpha\sb v\beta\sb3$ integrin on their surface, but efficiently infect other cells, we can use these properties of adenoviral vectors for bone marrow purging as well as for development of new viral vectors such as pseudotyped retroviral vectors and rAAV vectors. Replacement of self-inactivating retroviral vectors by recombinant adenoviral vectors will facilitate the above strategies for production of new viral vectors. In order to accomplish these goals, I developed a new method which is much more efficient than the current methods to construct adenoviral vectors. This method involves a cosmid vector system which is utilized to construct the full-length recombinant adenoviral vectors in vitro.^ First, I developed an efficient and flexible method for in vitro construction of the full-length recombinant adenoviral vectors in the cosmid vector system by use of a three-DNA fragment ligation. Then, this system was improved by use of a two-DNA fragment ligation. The cloning capacity of recombinant adenoviral vectors constructed by this method to develop recombinant adenoviral vectors depends on the efficiency of transfection only. No homologous recombination is required for development of infectious adenoviral vectors. Thus, the efficiency of generating the recombinant adenoviral vectors by the cosmid method reported here was much higher than that by the in vitro direct ligation method or the in vivo homologous recombination method reported before. This method of the in vitro construction of recombinant adenoviral vectors in the cosmid vector system may facilitate the development of adenoviral vector for human gene therapy. (Abstract shortened by UMI.) ^

Bayesian generalized linear models for meta-analysis of diagnostic tests

With the recognition of the importance of evidence-based medicine, there is an emerging need for methods to systematically synthesize available data. Specifically, methods to provide accurate estimates of test characteristics for diagnostic tests are needed to help physicians make better clinical decisions. To provide more flexible approaches for meta-analysis of diagnostic tests, we developed three Bayesian generalized linear models. Two of these models, a bivariate normal and a binomial model, analyzed pairs of sensitivity and specificity values while incorporating the correlation between these two outcome variables. Noninformative independent uniform priors were used for the variance of sensitivity, specificity and correlation. We also applied an inverse Wishart prior to check the sensitivity of the results. The third model was a multinomial model where the test results were modeled as multinomial random variables. All three models can include specific imaging techniques as covariates in order to compare performance. Vague normal priors were assigned to the coefficients of the covariates. The computations were carried out using the 'Bayesian inference using Gibbs sampling' implementation of Markov chain Monte Carlo techniques. We investigated the properties of the three proposed models through extensive simulation studies. We also applied these models to a previously published meta-analysis dataset on cervical cancer as well as to an unpublished melanoma dataset. In general, our findings show that the point estimates of sensitivity and specificity were consistent among Bayesian and frequentist bivariate normal and binomial models. However, in the simulation studies, the estimates of the correlation coefficient from Bayesian bivariate models are not as good as those obtained from frequentist estimation regardless of which prior distribution was used for the covariance matrix. The Bayesian multinomial model consistently underestimated the sensitivity and specificity regardless of the sample size and correlation coefficient. In conclusion, the Bayesian bivariate binomial model provides the most flexible framework for future applications because of its following strengths: (1) it facilitates direct comparison between different tests; (2) it captures the variability in both sensitivity and specificity simultaneously as well as the intercorrelation between the two; and (3) it can be directly applied to sparse data without ad hoc correction. ^

The sequence comparison index: A novel method for comparing proteins and proteomes

Historically morphological features were used as the primary means to classify organisms. However, the age of molecular genetics has allowed us to approach this field from the perspective of the organism's genetic code. Early work used highly conserved sequences, such as ribosomal RNA. The increasing number of complete genomes in the public data repositories provides the opportunity to look not only at a single gene, but at organisms' entire parts list. ^ Here the Sequence Comparison Index (SCI) and the Organism Comparison Index (OCI), algorithms and methods to compare proteins and proteomes, are presented. The complete proteomes of 104 sequenced organisms were compared. Over 280 million full Smith-Waterman alignments were performed on sequence pairs which had a reasonable expectation of being related. From these alignments a whole proteome phylogenetic tree was constructed. This method was also used to compare the small subunit (SSU) rRNA from each organism and a tree constructed from these results. The SSU rRNA tree by the SCI/OCI method looks very much like accepted SSU rRNA trees from sources such as the Ribosomal Database Project, thus validating the method. The SCI/OCI proteome tree showed a number of small but significant differences when compared to the SSU rRNA tree and proteome trees constructed by other methods. Horizontal gene transfer does not appear to affect the SCI/OCI trees until the transferred genes make up a large portion of the proteome. ^ As part of this work, the Database of Related Local Alignments (DaRLA) was created and contains over 81 million rows of sequence alignment information. DaRLA, while primarily used to build the whole proteome trees, can also be applied shared gene content analysis, gene order analysis, and creating individual protein trees. ^ Finally, the standard BLAST method for analyzing shared gene content was compared to the SCI method using 4 spirochetes. The SCI system performed flawlessly, finding all proteins from one organism against itself and finding all the ribosomal proteins between organisms. The BLAST system missed some proteins from its respective organism and failed to detect small ribosomal proteins between organisms. ^