Iván Olmos

Segmentation and Classification of Bone Marrow Cells Images Using Contextual Information for Medical Diagnosis of Acute Leukemias

Morphological identification of acute leukemia is a powerful tool used by hematologists to determine the family of such a disease. In some cases, experienced physicians are even able to determine the leukemia subtype of the sample. However, the identification process may have error rates up to 40% (when classifying acute leukemia subtypes) depending on the physician’s experience and the sample quality. This problem raises the need to create automatic tools that provide hematologists with a second opinion during the classification process. Our research presents a contextual analysis methodology for the detection of acute leukemia subtypes from bone marrow cells images. We propose a cells separation algorithm to break up overlapped regions. In this phase, we achieved an average accuracy of 95% in the evaluation of the segmentation process. In a second phase, we extract descriptive features to the nucleus and cytoplasm obtained in the segmentation phase in order to classify leukemia families and subtypes. We finally created a decision algorithm that provides an automatic diagnosis for a patient. In our experiments, we achieved an overall accuracy of 92% in the supervised classification of acute leukemia families, 84% for the lymphoblastic subtypes, and 92% for the myeloblastic subtypes. Finally, we achieved accuracies of 95% in the diagnosis of leukemia families and 90% in the diagnosis of leukemia subtypes.

Leukemia identification from bone marrow cells images using a machine vision and data mining strategy

The morphological analysis of medical images to support medical diagnosis is an important research area. This is the case of leukemia identification from bone marrow smears in which cells morphology is studied in order to classify the disease into its main family and subtype, so that a proper treatment can be indicated to the patient. In this paper we present a method to identify leukemia from bone marrow cells images using a combined machine vision and data mining strategy. Our process starts with a segmentation method to obtain leukemia cells and extract from them descriptive characteristics (geometrical, texture, statistical) and eigenvalues. We use these attributes to feed machine learning algorithms that learn to classify acute leukemia families and subtypes according to the FAB system. We show how the combination of descriptive features and eigenvalues helps to improve classification accuracy. Our method achieved accuracy above 95.5% to distinguish between the acute myeloblastic and lymphoblastic leukemia families and accuracy of 90% (and above) among five leukemia subtypes (after the acute leukemia families classification).

An Automaton for Motifs Recognition in DNA Sequences

In this paper we present a new algorithm to find inexact motifs (which are transformed into a set of exact subsequences) from a DNA sequence. Our algorithm builds an automaton that searches for the set of exact subsequences in the DNA database (that can be very long). It starts with a preprocessing phase in which it builds the finite automaton, in this phase it also considers the case in which two different subsequences share a substring (in other words, the subsequences might overlap), this is implemented in a similar way as the KMP algorithm. During the searching phase, the algorithm recognizes all instances in the set of input subsequences that appear in the DNA sequence. The automaton is able to perform the search phase in linear time with respect to the dimension of the input sequence. Experimental results show that the proposed algorithm performs better than the Aho-Corasick algorithm, which has been proved to perform better than the naive approach, even more; it is considered to run in linear time.

Reductions between the Subgraph Isomorphism Problem and Hamiltonian and SAT Problems

Subgraph isomorphism (SI) detection is an important problem for several computer science subfields. In this paper we present a study of the subgraph isomorphism problem (SIP) and its relation with the Hamiltonian cycles and SAT problems. In particular, we describe how instances of those problems can be solved throughout SI detection (using problems reductions). In our experiments we use an algorithm developed by the authors, which is capable to find all valid mappings in a SI instance. We performed several experiments, including cases for which there exists a known solution in polynomial time. In our analysis, we show the advantage and disadvantage of using a SI representation to solve Hamiltonian cycles and SAT problems

Mining common patterns on graphs

Finding common patterns is an important problem for several computer science subfields such as Machine Learning (ML) and Data Mining (DM). When we use graph-based representations, we need the Subgraph Isomorphism (SI) operation for finding common patterns. In this research we present a new approach to find a SI using a list code based representation without candidate generation. We implement a step by step expansion model with a width-depth search. The proposed approach is suitable to work with labeled and unlabeled graphs, with directed and undirected edges. Our experiments show a promising method to be used with scalable graph matching.

Finding patterns in labeled graphs using spectrum feature vectors in a SOM network

Knowledge discovery in structured databases is very important nowadays. In the last years, graph-based data mining algorithms have used artificial neural networks as tools to support clustering. Several of these algorithms have obtained promising results, but they show expensive computational costs. In this work we introduce an algorithm for clustering graphs based on a SOM network, which is part of a process for discovering useful frequent patterns in large graph databases. Our algorithm is able to handle non-directed, cyclic graphs with labels in vertices and edges. An important characteristic is that it presents polynomial computational complexity, because it uses as input a feature vector built with the spectra of the Laplacian of an adjacent matrix. Such matrix contains codes representing the labels in the graph, which preserves the semantic information included in the graphs to be grouped. We tested our algorithm in a small set of graphs and in a large structured database, finding that it creates meaningful groups of graphs.

Motifs recognition in DNA sequences comparing the motif finding automaton algorithm against a traditional approach

This paper presents the comparison between algorithms used to find inexact motifs (transformed into a set of exact sub-sequences) in a DNA sequence. The MFA algorithm builds an automaton that searches for the set of exact sub-sequences by building a finite automaton in a similar way to the KMP algorithm. This algorithm is compared against the traditional automaton using the basic idea of the subset construction. This traditional algorithm is implemented using some characteristics to increase its performance which will end in an algorithm that can be proven to have the optimal number of states.