Leopoldo Altamirano

Acute leukemia classification by ensemble particle swarm model selection

OBJECTIVE Acute leukemia is a malignant disease that affects a large proportion of the world population. Different types and subtypes of acute leukemia require different treatments. In order to assign the correct treatment, a physician must identify the leukemia type or subtype. Advanced and precise methods are available for identifying leukemia types, but they are very expensive and not available in most hospitals in developing countries. Thus, alternative methods have been proposed. An option explored in this paper is based on the morphological properties of bone marrow images, where features are extracted from medical images and standard machine learning techniques are used to build leukemia type classifiers. METHODS AND MATERIALS This paper studies the use of ensemble particle swarm model selection (EPSMS), which is an automated tool for the selection of classification models, in the context of acute leukemia classification. EPSMS is the application of particle swarm optimization to the exploration of the search space of ensembles that can be formed by heterogeneous classification models in a machine learning toolbox. EPSMS does not require prior domain knowledge and it is able to select highly accurate classification models without user intervention. Furthermore, specific models can be used for different classification tasks. RESULTS We report experimental results for acute leukemia classification with real data and show that EPSMS outperformed the best results obtained using manually designed classifiers with the same data. The highest performance using EPSMS was of 97.68% for two-type classification problems and of 94.21% for more than two types problems. To the best of our knowledge, these are the best results reported for this data set. Compared with previous studies, these improvements were consistent among different type/subtype classification tasks, different features extracted from images, and different feature extraction regions. The performance improvements were statistically significant. We improved previous results by an average of 6% and there are improvements of more than 20% with some settings. In addition to the performance improvements, we demonstrated that no manual effort was required during acute leukemia type/subtype classification. CONCLUSIONS Morphological classification of acute leukemia using EPSMS provides an alternative to expensive diagnostic methods in developing countries. EPSMS is a highly effective method for the automated construction of ensemble classifiers for acute leukemia classification, which requires no significant user intervention. EPSMS could also be used to address other medical classification tasks.

Teaching a robot to perform task through imitation and on-line feedback

Service robots are becoming increasingly available and it is expected that they will be part of many human activities in the near future. It is desirable for these robots to adapt themselves to the users needs, so non-expert users will have to teach them how to perform new tasks in natural ways. In this paper a new teaching by demonstration algorithm is described. It uses a Kinect® sensor to track the movements of a user, eliminating the need of special sensors or environment conditions, it represents the tasks with a relational representation to facilitate the correspondence problem between the user and robot arm and to learn how to perform tasks in a more general description, it uses reinforcement learning to improve over the initial sequences provided by the user, and it incorporates on-line feedback from the user during the learning process creating a novel dynamic reward shaping mechanism to converge faster to an optimal policy. We demonstrate the approach by learning simple manipulation tasks of a robot arm and show its superiority over more traditional reinforcement learning algorithms.

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.

FPGA-based Pipeline Architecture to Transform Cartesian Images into Foveal Images by Using a new Foveation Approach

In vision systems the image processing represents a bottle neck because the big amount of information that should be analyzed. Working with variant spaces over the visual field has been widely proposed as a way to reduce such information. Foveal vision is one of these proposals by providing a way to transform the visual field obtained with conventional cameras into a sampling with high resolution at the center and decreasing over the periphery such as in mammal vision systems. In this paper, an FPGA based architecture to transform conventional images into foveal images is presented. The hardware algorithm has been taken from new proposal to foveate images. Strategies as parallelism and pipeline are exploited to obtain a high performance and thus, with both of them, reduction in the visual field and the transformation in real time of the digital images into foveal images, a vision system can accelerate its performance and reaches real time restrictions

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).

Genetic selection of fuzzy model for acute leukemia classification

Leukemia is a disease characterized by an abnormal increase of white blood cells. This disease is divided into two types: lymphoblastic and myeloid, each of which is divided in subtypes. Differentiating the type and subtype of acute leukemia is important in order to determine the correct type of treatment to be assigned by the affected person. Diagnostic tests available today, such as those based on cell morphology, have a high error rate. Others, as those based on cytometry or microarray, are expensive. In order to avoid those drawbacks this paper proposes the automatic selection of a fuzzy model for accurate classification of types and subtypes of acute leukemia based on cell morphology. Our experimental results reach up to 93.52% in classification of acute leukemia types, 87.36% in lymphoblastic subtypes and 94.42% in myeloid subtypes. Our results show a significant improvement compared with classifiers which parameters were manually tuned using the same data set. Details of the proposed method, as well as experiments and results are shown.

Three hypothesis algorithm with occlusion reasoning for multiple people tracking

Abstract. This work proposes a detection-based tracking algorithm able to locate and keep the identity of multiple people, who may be occluded, in uncontrolled stationary environments. Our algorithm builds a tracking graph that models spatio-temporal relationships among attributes of interacting people to predict and resolve partial and total occlusions. When a total occlusion occurs, the algorithm generates various hypotheses about the location of the occluded person considering three cases: (a) the person keeps the same direction and speed, (b) the person follows the direction and speed of the occluder, and (c) the person remains motionless during occlusion. By analyzing the graph, our algorithm can detect trajectories produced by false alarms and estimate the location of missing or occluded people. Our algorithm performs acceptably under complex conditions, such as partial visibility of individuals getting inside or outside the scene, continuous interactions and occlusions among people, wrong or missing information on the detection of persons, as well as variation of the person’s appearance due to illumination changes and background-clutter distracters. Our algorithm was evaluated on test sequences in the field of intelligent surveillance achieving an overall precision of 93%. Results show that our tracking algorithm outperforms even trajectory-based state-of-the-art algorithms.

Transition regions detection from satellite images based on evolutionary region growing segmentation

In nature, there exist transition regions between homogeneous land-cover changes. These specific regions cause ambiguity and uncertainty to segmentation and classification algorithms when applied to remote sensed images. Our research, proposes a novel method called TReDet to identify transition regions in the way they are found in nature. They appear in a random way without following a defined geometric pattern. Experimental results with synthetic images show that TReDet clearly outperforms commercial software using some of the quantitative measures used to evaluate our method. Even when in some cases the difference in percentage is not so high for some of the measures used, TReDet obtains a more realistic segmentation of the transition regions. Furthermore, it does not follow parallel transition bands or any other geometric pattern avoiding the errors falling in the borderline of the transition region of other methods.

Decision level multiple cameras fusion using dezert-smarandache theory

This paper presents a model for multiple cameras fusion, which is based on Dezert-Smarandache theory of evidence. We have developed a fusion model which works at the decision fusion level to track objects on a ground plane using geographically distributed cameras. As we are fusing at decision level, track is done based on predefined zones. We present early results of our model tested on CGI animated simulations, applying a perspective-based basic belief assignment function. Our experiments suggest that the proposed technique yields a good improvement in tracking accuracy when spatial regions are used to track.

OCCLUSION MODEL FROM HUMAN INTERACTION ANALYSIS FOR TRACKING MULTIPLE PEOPLE

In this work we investigate the problem of tracking multiple interacting people under uncontrolled stationary environments for intelligent surveillance applications. This domain is very challenging since the clothing appearance changes of the people over time make difficult the temporal association of their identities. The problem is emphasized when individuals move close to each other, are occluded, or abruptly change their trajectories. We propose a tracking graph that models spatial and temporal relationships among people in order to predict and resolve partial and total occlusions. When a total occlusion event occurs, the model generates three possible hypotheses about the location of the occluded person according to a human interaction analysis. This model is able to detect false positives and false negatives in the detection measurements and it can also estimate the location of missing or occluded people. Our approach was evaluated on benchmark sequences and results show how it outperforms state-of-the-art algorithms even in the presence of long periods of occlusion.