A decision-tree approach for the differential diagnosis of chronic lymphoid leukemias and peripheral B-cell lymphomas.

BACKGROUND AND OBJECTIVE: Here we propose a decision-tree approach for the differential diagnosis of distinct WHO categories B-cell chronic lymphoproliferative disorders using flow cytometry data. Flow cytometry is the preferred method for the immunophenotypic characterization of leukemia and lymphoma, being able to process and register multiparametric data about tens of thousands of cells per second. METHODS: The proposed decision-tree is composed by logistic function nodes that branch throughout the tree into sets of (possible) distinct leukemia/lymphoma diagnoses. To avoid overfitting, regularization via the Lasso algorithm was used. The code can be run online at https://codeocean.com/2018/03/08/a-decision-tree-approach-for-the-differential-diagnosis-of-chronic-lymphoid-leukemias-and-peripheral-b-cell-lymphomas/ or downloaded from https://github.com/lauramoraes/bioinformatics-sourcecode to be executed in Matlab. RESULTS: The proposed approach was validated in diagnostic peripheral blood and bone marrow samples from 283 mature lymphoid leukemias/lymphomas patients. The proposed approach achieved 95% correctness in the cross-validation test phase (100% in-sample), 61% giving a single diagnosis and 34% (possible) multiple disease diagnoses. Similar results were obtained in an out-of-sample validation dataset. The generated tree reached the final diagnoses after up to seven decision nodes. CONCLUSIONS: Here we propose a decision-tree approach for the differential diagnosis of mature lymphoid leukemias/lymphomas which proved to be accurate during out-of-sample validation. The full process is accomplished through seven binary transparent decision nodes.

From big flow cytometry datasets to smart diagnostic strategies: The EuroFlow approach.

The rise in the analytical speed of mutiparameter flow cytometers made possible by the introduction of digital instruments, has brought up the possibility to manage progressively higher number of parameters simultaneously on significantly greater numbers of individual cells. This has led to an exponential increase in the complexity and volume of flow cytometry data generated about cells present in individual samples evaluated in a single measurement. This increase demands for new developments in flow cytometry data analysis, graphical representation, and visualization and interpretation tools to address the new big data challenges, i.e. processing data files of ≥10-25 parameters per cell in samples with >5-10 million cells (= up to 250 million data points per cell sample) obtained in a few minutes. Here, we present a comprehensive review of some of the tools developed by the EuroFlow consortium for processing flow cytometric big data files in diagnostic laboratories, particularly focused on automated EuroFlow approaches for: i) identification of all cell populations coexisting in a sample (automated gating); ii) smart classification of aberrant cell populations in routine diagnostics; iii) automated reporting; together with iv) new tools developed to visualize n-dimensional data in 2-dimensional plots to support expert-guided automated data analysis. The concept of using reference data bases implemented into software programs, in combination with multivariate statistical analysis pioneered by EuroFlow, provides an innovative, highly efficient and fast approach for diagnostic screening, classification and monitoring of patients with distinct hematological and immune disorders, as well as other diseases.

Automated database-guided expert-supervised orientation for immunophenotypic diagnosis and classification of acute leukemia.

Precise classification of acute leukemia (AL) is crucial for adequate treatment. EuroFlow has previously designed an AL orientation tube (ALOT) to guide towards the relevant classification panel (T-cell acute lymphoblastic leukemia (T-ALL), B-cell precursor (BCP)-ALL and/or acute myeloid leukemia (AML)) and final diagnosis. Now we built a reference database with 656 typical AL samples (145 T-ALL, 377 BCP-ALL, 134 AML), processed and analyzed via standardized protocols. Using principal component analysis (PCA)-based plots and automated classification algorithms for direct comparison of single-cells from individual patients against the database, another 783 cases were subsequently evaluated. Depending on the database-guided results, patients were categorized as: (i) typical T, B or Myeloid without or; (ii) with a transitional component to another lineage; (iii) atypical; or (iv) mixed-lineage. Using this automated algorithm, in 781/783 cases (99.7%) the right panel was selected, and data comparable to the final WHO-diagnosis was already provided in >93% of cases (85% T-ALL, 97% BCP-ALL, 95% AML and 87% mixed-phenotype AL patients), even without data on the full-characterization panels. Our results show that database-guided analysis facilitates standardized interpretation of ALOT results and allows accurate selection of the relevant classification panels, hence providing a solid basis for designing future WHO AL classifications.

Impact of maternal education level on risk of low Apgar score.

OBJECTIVE: To investigate the association between 5-min Apgar score and socio-economic characteristics of pregnant women, particularly education level. STUDY DESIGN: Population-based cross-sectional study. METHODS: This study used hospital records of live term singleton births in Brazil from 2004 to 2009, obtained from the Ministry of Health National Information System. Crude and adjusted odds ratios (ORs) were used to estimate the risk of a low 5-min Apgar score (≤6) associated with maternal education level, maternal age, marital status, primiparity, number of prenatal visits and mode of delivery (vaginal/caesarean section). RESULTS: Nearly 12 million records were analysed. Births from mothers with 0, 1-3, 4-7 and 8-11 years of education resulted in crude ORs for low 5-min Apgar score of 3.1, 2.2, 1.8 and 1.3, respectively (reference: ≥12 years of education). The crude OR for mothers aged ≥41 years (reference 21-34 years) was 1.4, but no risk was detected for those with ≥12 years of education and those who gave birth by caesarean section (OR 1.0 [95% confidence interval 0.9-1.2]). Generally, the risk of a low 5-min Apgar score was found to increase as maternal age moved away from 21 to 34 years (OR 1.1-1.7), and for mothers with the same characteristics, the risk of a low 5-min Apgar score was found to decrease markedly as education level increased (adjusted OR decreased from 2.6 to 1.2). CONCLUSION: Maternal education level is clearly associated with the risk of a low 5-min Apgar score.

Automated pattern-guided principal component analysis vs expert-based immunophenotypic classification of B-cell chronic lymphoproliferative disorders: a step forward in the standardization of clinical immunophenotyping.

Immunophenotypic characterization of B-cell chronic lymphoproliferative disorders (B-CLPD) is becoming increasingly complex due to usage of progressively larger panels of reagents and a high number of World Health Organization (WHO) entities. Typically, data analysis is performed separately for each stained aliquot of a sample; subsequently, an expert interprets the overall immunophenotypic profile (IP) of neoplastic B-cells and assigns it to specific diagnostic categories. We constructed a principal component analysis (PCA)-based tool to guide immunophenotypic classification of B-CLPD. Three reference groups of immunophenotypic data files-B-cell chronic lymphocytic leukemias (B-CLL; n = 10), mantle cell (MCL; n = 10) and follicular lymphomas (FL; n = 10)--were built. Subsequently, each of the 175 cases studied was evaluated and assigned to either one of the three reference groups or to none of them (other B-CLPD). Most cases (89%) were correctly assigned to their corresponding WHO diagnostic group with overall positive and negative predictive values of 89 and 96%, respectively. The efficiency of the PCA-based approach was particularly high among typical B-CLL, MCL and FL vs other B-CLPD cases. In summary, PCA-guided immunophenotypic classification of B-CLPD is a promising tool for standardized interpretation of tumor IP, their classification into well-defined entities and comprehensive evaluation of antibody panels.

A new local-global approach for classification.

In this paper, we propose a new local-global pattern classification scheme that combines supervised and unsupervised approaches, taking advantage of both, local and global environments. We understand as global methods the ones concerned with the aim of constructing a model for the whole problem space using the totality of the available observations. Local methods focus into sub regions of the space, possibly using an appropriately selected subset of the sample. In the proposed method, the sample is first divided in local cells by using a Vector Quantization unsupervised algorithm, the LBG (Linde-Buzo-Gray). In a second stage, the generated assemblage of much easier problems is locally solved with a scheme inspired by Bayes' rule. Four classification methods were implemented for comparison purposes with the proposed scheme: Learning Vector Quantization (LVQ); Feedforward Neural Networks; Support Vector Machine (SVM) and k-Nearest Neighbors. These four methods and the proposed scheme were implemented in eleven datasets, two controlled experiments, plus nine public available datasets from the UCI repository. The proposed method has shown a quite competitive performance when compared to these classical and largely used classifiers. Our method is simple concerning understanding and implementation and is based on very intuitive concepts.

Generalized risk zone: selecting observations for classification.

In this paper, we extend the risk zone concept by creating the Generalized Risk Zone. The Generalized Risk Zone is a model-independent scheme to select key observations in a sample set. The observations belonging to the Generalized Risk Zone have shown comparable, in some experiments even better, classification performance when compared to the use of the whole sample. The main tool that allows this extension is the Cauchy-Schwartz divergence, used as a measure of dissimilarity between probability densities. To overcome the setback concerning pdf's estimation, we used the ideas provided by the Information Theoretic Learning, allowing the calculation to be performed on the available observations only. We used the proposed methodology with Learning Vector Quantization, feedforward Neural Networks, Support Vector Machines, and Nearest Neighbors.

A probabilistic approach for the evaluation of minimal residual disease by multiparameter flow cytometry in leukemic B-cell chronic lymphoproliferative disorders.

Multiparameter flow cytometry has become an essential tool for monitoring response to therapy in hematological malignancies, including B-cell chronic lymphoproliferative disorders (B-CLPD). However, depending on the expertise of the operator minimal residual disease (MRD) can be misidentified, given that data analysis is based on the definition of expert-based bidimensional plots, where an operator selects the subpopulations of interest. Here, we propose and evaluate a probabilistic approach based on pattern classification tools and the Bayes theorem, for automated analysis of flow cytometry data from a group of 50 B-CLPD versus normal peripheral blood B-cells under MRD conditions, with the aim of reducing operator-associated subjectivity. The proposed approach provided a tool for MRD detection in B-CLPD by flow cytometry with a sensitivity of < or =8 x 10(-5) (median of < or =2 x 10(-7)). Furthermore, in 86% of B-CLPD cases tested, no events corresponding to normal B-cells were wrongly identified as belonging to the neoplastic B-cell population at a level of < or =10(-7). Thus, this approach based on the search for minimal numbers of neoplastic B-cells similar to those detected at diagnosis could potentially be applied with both a high sensitivity and specificity to investigate for the presence of MRD in virtually all B-CLPD. Further studies evaluating its efficiency in larger series of patients, where reactive conditions and non-neoplastic disorders are also included, are required to confirm these results.

A new automated flow cytometry data analysis approach for the diagnostic screening of neoplastic B-cell disorders in peripheral blood samples with absolute lymphocytosis.

Currently, multiparameter flow cytometry immunophenotyping is the selected method for the differential diagnostic screening between reactive lymphocytosis and neoplastic B-cell chronic lymphoproliferative disorders (B-CLPD). Despite this, current multiparameter flow cytometry data analysis approaches still remain subjective due to the need of experienced personnel for both data analysis and interpretation of the results. In this study, we describe and validate a new automated method based on vector quantization algorithms to analyze multiparameter flow cytometry immunophenotyping data in a series of 307 peripheral blood (PB) samples. Our results show that the automated method of analysis proposed compares well with currently used manual approach and significantly improves semiautomated approaches and, that by using it, a highly efficient discrimination with 100% specificity and 100% sensitivity can be made between normal/reactive PB samples and cases with B-CLPD based on the total B-cell number and/or the sIgkappa+/sIglambda+ B-cell ratio. In addition, the method proved to be able to detect the presence of pathologic neoplastic B-cells even when these are present at low frequencies (<5% of all lymphocytes in the sample) and in poor-quality samples enriched in 'noise' events.

Modeling exchange rates: smooth transitions, neural networks, and linear models.

The goal of this paper is to test and model nonlinearities in several monthly exchange rates time series. We apply two different nonlinear alternatives, namely: the artificial neural-network time series model estimated with Bayesian regularization; and a flexible smooth transition specification, called the neuro-coefficient smooth transition autoregression. The linearity test rejects the null hypothesis of linearity in 10 out of 14 series. We compare, using different measures, the forecasting performance of the nonlinear specifications with the linear autoregression and the random walk models.

Multilayer neural networks and function reconstruction by using a priori knowledge.

This paper proposes a new methodology to approximate functions by incorporating a priori information. The relationship between the proposed scheme and multilayer neural networks is explored theoretically and numerically. This approach is particularly interesting for the very relevant class of limited spectrum functions. The number of free parameters is smaller if compared to Back-Propagation Algorithm opening the way for better generalization results.

A new optimization based approach to experimental combination chemotherapy.

A new approach towards the design of optimal multiple drug experimental cancer chemotherapy is presented. Once an adequate model is specified, an optimization procedure is used in order to achieve an optimal compromise between after treatment tumor size and toxic effects on healthy tissues. In our approach we consider a model including cancer cell population growth and pharmacokinetic dynamics. These elements of the model are essential in order to allow less empirical relationships between multiple drug delivery policies, and their effects on cancer and normal cells. The desired multiple drug dosage schedule is computed by minimizing a customizable cost function subject to dynamic constraints expressed by the model. However, this additional dynamic wealth increases the complexity of the problem which, in general, cannot be solved in a closed form. Therefore, we propose an iterative optimization algorithm of the projected gradient type where the Maximum Principle of Pontryagin is used to select the optimal control policy.