Classifier performance estimation under the constraint of a finite sample size: Resampling schemes applied to neural network classifiers

In a practical classifier design problem the sample size is limited, and the available finite sample needs to be used both to design a classifier and to predict the classifier’s performance for the true population. Since a larger sample is more representative of the population, it is advantageous to design the classifier with all the available cases, and to use a resampling technique for performance prediction. We conducted a Monte Carlo simulation study to compare the ability of different resampling techniques in predicting the performance of a neural network (NN) classifier designed with the available sample. We used the area under the receiver operating characteristic curve as the performance index for the NN classifier. We investigated resampling techniques based on the cross-validation, the leave-one-out method, and three different types of bootstrapping, namely, the ordinary, .632, and .632+ bootstrap. Our results indicated that, under the study conditions, there can be a large difference in the accuracy of the prediction obtained from different resampling methods, especially when the feature space dimensionality is relatively large and the sample size is small. Although this investigation is performed under some specific conditions, it reveals important trends for the problem of classifier performance prediction under the constraint of a limited data set.     

Doxorubicin attached to HPMA copolymer via amide bond modifies the glycosylation pattern of EL4 cells

To avoid the side effects of the anti-cancer drug doxorubicin (Dox), we conjugated this drug to a N-(2-hydroxypropyl)methacrylamide (HPMA) copolymer backbone. Dox was conjugated via an amide bond (Dox-HPMA^AM, PK1) or a hydrazone pH-sensitive bond (Dox-HPMA^HYD). In contrast to Dox and Dox-HPMA^HYD, Dox-HPMA^AM accumulates within the cell’s intracellular membranes, including those of the Golgi complex and endoplasmic reticulum, both involved in protein glycosylation. Flow cytometry was used to determine lectin binding and cell death, immunoblot to characterize the presence of CD7, CD43, CD44, and CD45, and high-performance anion exchange chromatography with pulsed amperometric detector analysis for characterization of plasma membrane saccharide composition. Incubation of EL4 cells with Dox-HPMA^AM conjugate, in contrast to Dox or Dox-HPMA^HYD, increased the amounts of membrane surface-associated glycoproteins, as well as saccharide moieties recognized by peanut agglutinin, Erythrina cristagalli, or galectin-1 lectins. Only Dox-HPMA^AM increased expression of the highly glycosylated membrane glycoprotein CD43, while expression of others (CD7, CD44, and CD45) was unaffected. The binding sites for galectin-1 are present on CD43 molecule. Furthermore, we present that EL4 treated with Dox-HPMA^AM possesses increased sensitivity to galectin-1-induced apoptosis. In this study, we demonstrate that Dox-HPMA^AM treatment changes glycosylation of the EL4 T cell lymphoma surface and sensitizes the cells to galectin-1-induced apoptosis.     

Simultaneous manifestation of chronic lymphocytic leukemia (CLL) and hairy cell leukemia (HCL)

We report a unique case of 83-year-old Caucasian male with the initial simultaneous manifestation of chronic lymphocytic leukemia (CLL) and hairy cell leukemia (HCL). The patient presented with absolute lymphocytosis in the blood, asymptomatic generalized lymphadenopathy, and mild splenomegaly. The diagnosis of CLL was suggested from the blood film, but subsequent flow cytometric (FC) analysis on peripheral blood mononuclear cells (PBMNC) revealed two distinct abnormal clones of mature B cells. A small subpopulation (7%) of lymphoid cells expressed CD20, CD11c, FMC-7, CD103, CD25, and kappa surface light chain, consistent with HCL. The larger subpopulation (75%) of lymphoid cells expressed CD19, CD20, CD23, CD5, and lambda light chain, consistent with CLL. The expression of different immunoglobulin light chains on the circulating CLL (lambda) and HCL (kappa) cells suggested two, independent, malignant B-cell clones. Interestingly, FC analysis of bone marrow (BM) cells done 6 months later revealed bright lambda light chain expression on the HCL cells. Despite administration of several different courses of chemotherapy, the HCL subpopulation was not eliminated from the BM but remained stable between 7% and 10% of total BM lymphoid cells. The CLL, responsible for most of clinical symptoms in our patient, responded to combination chemotherapy with fludarabine and cytoxan, and later to monotherapy with rituximab.     

Aggressive T-cell large granular lymphocyte leukemia: a case report and review of the literature

The majority of patients with T-cell large granular lymphocyte (LGL) leukemia will have an indolent clinical course. Herein, we report a case of an aggressive T-cell LGL leukemia in a previously healthy 42-year-old Caucasian male who presented with acute onset of B-symptoms, hepatosplenomegaly, lymphocytosis, moderate anemia, and thrombocytopenia. Immunophenotypically, the malignant cells co-expressed CD3(+)CD8(+)CD56(+) markers and the T-cell receptor beta (TCR beta) gene demonstrated clonal rearrangement. The patient was treated with an intensive chemotherapeutic regimen (hyper-CVAD) and he achieved a complete remission. A systematic review of all available English literature revealed 12 well-described cases of aggressive T-cell LGL leukemia suggesting that this variant is a new and distinct entity in the spectrum of LGL disorders.     

Malignant and benign breast masses on 3D US volumetric images: effect of computer-aided diagnosis on radiologist accuracy

PURPOSE: To retrospectively investigate the effect of using a custom-designed computer classifier on radiologists' sensitivity and specificity for discriminating malignant masses from benign masses on three-dimensional (3D) volumetric ultrasonographic (US) images, with histologic analysis serving as the reference standard. MATERIALS AND METHODS: Informed consent and institutional review board approval were obtained. Our data set contained 3D US volumetric images obtained in 101 women (average age, 51 years; age range, 25-86 years) with 101 biopsy-proved breast masses (45 benign, 56 malignant). A computer algorithm was designed to automatically delineate mass boundaries and extract features on the basis of segmented mass shapes and margins. A computer classifier was used to merge features into a malignancy score. Five experienced radiologists participated as readers. Each radiologist read cases first without computer-aided diagnosis (CAD) and immediately thereafter with CAD. Observers' malignancy rating data were analyzed with the receiver operating characteristic (ROC) curve. RESULTS: Without CAD, the five radiologists had an average area under the ROC curve (A(z)) of 0.83 (range, 0.81-0.87). With CAD, the average A(z) increased significantly (P = .006) to 0.90 (range, 0.86-0.93). When a 2% likelihood of malignancy was used as the threshold for biopsy recommendation, the average sensitivity of radiologists increased from 96% to 98% with CAD, while the average specificity for this data set decreased from 22% to 19%. If a biopsy recommendation threshold could be chosen such that sensitivity would be maintained at 96%, specificity would increase to 45% with CAD. CONCLUSION: Use of a computer algorithm may improve radiologists' accuracy in distinguishing malignant from benign breast masses on 3D US volumetric images.     

Fundamental differences in cell cycle deregulation in human papillomavirus-positive and human papillomavirus-negative head/neck and cervical cancers

Human papillomaviruses (HPV) are associated with nearly all cervical cancers, 20% to 30% of head and neck cancers (HNC), and other cancers. Because HNCs also arise in HPV-negative patients, this type of cancer provides unique opportunities to define similarities and differences of HPV-positive versus HPV-negative cancers arising in the same tissue. Here, we describe genome-wide expression profiling of 84 HNCs, cervical cancers, and site-matched normal epithelial samples in which we used laser capture microdissection to enrich samples for tumor-derived versus normal epithelial cells. This analysis revealed that HPV(+) HNCs and cervical cancers differed in their patterns of gene expression yet shared many changes compared with HPV(-) HNCs. Some of these shared changes were predicted, but many others were not. Notably, HPV(+) HNCs and cervical cancers were found to be up-regulated in their expression of a distinct and larger subset of cell cycle genes than that observed in HPV(-) HNC. Moreover, HPV(+) cancers overexpressed testis-specific genes that are normally expressed only in meiotic cells. Many, although not all, of the hallmark differences between HPV(+) HNC and HPV(-) HNC were a direct consequence of HPV and in particular the viral E6 and E7 oncogenes. This included a novel association of HPV oncogenes with testis-specific gene expression. These findings in primary human tumors provide novel biomarkers for early detection of HPV(+) and HPV(-) cancers, and emphasize the potential value of targeting E6 and E7 function, alone or combined with radiation and/or traditional chemotherapy, in the treatment of HPV(+) cancers.     

New insight into intrachromosomal deletions induced by chrysotile in the gpt delta transgenic mutation assay

BACKGROUND: Genotoxicity is often a prerequisite to the development of malignancy. Considerable evidence has shown that exposure to asbestos fibers results in the generation of chromosomal aberrations and multilocus mutations using various in vitro approaches. However, there is less evidence to demonstrate the contribution of deletions to the mutagenicity of asbestos fibers in vivo. OBJECTIVES: In the present study, we investigated the mutant fractions and the patterns induced by chrysotile fibers in gpt delta transgenic mouse primary embryo fibroblasts (MEFs) and compared the results obtained with hydrogen peroxide (H2O2) in an attempt to illustrate the role of oxyradicals in fiber mutagenesis. RESULTS: Chrysotile fibers induced a dose-dependent increase in mutation yield at the redBA/gam loci in transgenic MEF cells. The number of lambda mutants losing both redBA and gam loci induced by chrysotiles at a dose of 1 microg/cm(2) increased by > 5-fold relative to nontreated controls (p < 0.005). Mutation spectra analyses showed that the ratio of lambda mutants losing the redBA/gam region induced by chrysotiles was similar to those induced by equitoxic doses of H2O2. Moreover, treatment with catalase abrogated the accumulation of y-H2AX, a biomarker of DNA double-strand breaks, induced by chrysotile fibers. CONCLUSIONS: Our results provide novel information on the frequencies and types of mutations induced by asbestos fibers in the gpt delta transgenic mouse mutagenic assay, which shows great promise for evaluating fiber/particle mutagenicity in vivo.     

Lack of association between ENAM gene polymorphism and dental caries in primary and permanent teeth in Czech children

Clinical Oral Investigations - The enamelin gene (ENAM) polymorphism (rs12640848) was recently associated with dental caries in primary teeth in Polish children. The aims of the present study were...