Ugur Halici

An automated differential blood count system

While the early diagnosis of hematopoietic system disorders is very important in hematology, it is a highly complex and time consuming task. The early diagnosis requires a lot of patients to be followed-up by experts which, in general is unfeasible because of the required number of experts. The differential blood counter (DBC) system that we have developed is an attempt to automate the task performed manually by experts in routine. In our system, the cells are segmented using active contour models (snakes and balloons), which are initialized using morphological operators. Shape based and texture based features are utilized for the classification task. Different classifiers such as k-nearest neighbors, learning vector quantization, multi-layer perceptron and support vector machine are employed.

Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer

Importance Application of deep learning algorithms to whole-slide pathology images can potentially improve diagnostic accuracy and efficiency. Objective Assess the performance of automated deep learning algorithms at detecting metastases in hematoxylin and eosin–stained tissue sections of lymph nodes of women with breast cancer and compare it with pathologists’ diagnoses in a diagnostic setting. Design, Setting, and Participants Researcher challenge competition (CAMELYON16) to develop automated solutions for detecting lymph node metastases (November 2015-November 2016). A training data set of whole-slide images from 2 centers in the Netherlands with (n = 110) and without (n = 160) nodal metastases verified by immunohistochemical staining were provided to challenge participants to build algorithms. Algorithm performance was evaluated in an independent test set of 129 whole-slide images (49 with and 80 without metastases). The same test set of corresponding glass slides was also evaluated by a panel of 11 pathologists with time constraint (WTC) from the Netherlands to ascertain likelihood of nodal metastases for each slide in a flexible 2-hour session, simulating routine pathology workflow, and by 1 pathologist without time constraint (WOTC). Exposures Deep learning algorithms submitted as part of a challenge competition or pathologist interpretation. Main Outcomes and Measures The presence of specific metastatic foci and the absence vs presence of lymph node metastasis in a slide or image using receiver operating characteristic curve analysis. The 11 pathologists participating in the simulation exercise rated their diagnostic confidence as definitely normal, probably normal, equivocal, probably tumor, or definitely tumor. Results The area under the receiver operating characteristic curve (AUC) for the algorithms ranged from 0.556 to 0.994. The top-performing algorithm achieved a lesion-level, true-positive fraction comparable with that of the pathologist WOTC (72.4% [95% CI, 64.3%-80.4%]) at a mean of 0.0125 false-positives per normal whole-slide image. For the whole-slide image classification task, the best algorithm (AUC, 0.994 [95% CI, 0.983-0.999]) performed significantly better than the pathologists WTC in a diagnostic simulation (mean AUC, 0.810 [range, 0.738-0.884]; P < .001). The top 5 algorithms had a mean AUC that was comparable with the pathologist interpreting the slides in the absence of time constraints (mean AUC, 0.960 [range, 0.923-0.994] for the top 5 algorithms vs 0.966 [95% CI, 0.927-0.998] for the pathologist WOTC). Conclusions and Relevance In the setting of a challenge competition, some deep learning algorithms achieved better diagnostic performance than a panel of 11 pathologists participating in a simulation exercise designed to mimic routine pathology workflow; algorithm performance was comparable with an expert pathologist interpreting whole-slide images without time constraints. Whether this approach has clinical utility will require evaluation in a clinical setting.

Feature extraction and classification of blood cells for an automated differential blood count system

The differential blood counter system we developed is an attempt to automate the task performed manually by experts in routine. Feature extraction and classification are two important components of our automated system. In this paper, classification of blood cells using various approaches including neural network based classifiers and support vector machine are presented together with the features used in the classification.

Reinforcement learning with internal expectation for the random neural network

Abstract The reinforcement learning scheme proposed in Halici (1977) (Halici, U., 1997. Journal of Biosystems 40 (1/2), 83–91) for the random neural network (Gelenbe, E., 1989b. Neural Computation 1 (4), 502–510) is based on reward and performs well for stationary environments. However, when the environment is not stationary it suffers from getting stuck to the previously learned action and extinction is not possible. In this paper, the reinforcement learning scheme is extended by introducing a weight update rule which takes into consideration the internal expectation of reinforcement. With the proposed scheme, the system behaves as in learning with reward when the reward for the learned action is not below the internal expectation, otherwise it behaves as in learning with punishment so that other possibilities can be explored. Such a scheme has made extinction possible while resulting in a good convergence to the most rewarding action.

A novel deep learning approach for classification of EEG motor imagery signals

OBJECTIVE Signal classification is an important issue in brain computer interface (BCI) systems. Deep learning approaches have been used successfully in many recent studies to learn features and classify different types of data. However, the number of studies that employ these approaches on BCI applications is very limited. In this study we aim to use deep learning methods to improve classification performance of EEG motor imagery signals. APPROACH In this study we investigate convolutional neural networks (CNN) and stacked autoencoders (SAE) to classify EEG Motor Imagery signals. A new form of input is introduced to combine time, frequency and location information extracted from EEG signal and it is used in CNN having one 1D convolutional and one max-pooling layers. We also proposed a new deep network by combining CNN and SAE. In this network, the features that are extracted in CNN are classified through the deep network SAE. MAIN RESULTS The classification performance obtained by the proposed method on BCI competition IV dataset 2b in terms of kappa value is 0.547. Our approach yields 9% improvement over the winner algorithm of the competition. SIGNIFICANCE Our results show that deep learning methods provide better classification performance compared to other state of art approaches. These methods can be applied successfully to BCI systems where the amount of data is large due to daily recording.

METU interoperable database system

METU INteroperable Database System (MIND) is a multidatabase system that aims at achieving interoperability among heterogeneous, federated DBMSs. MIND architecture if based on OMG distributed object management model. It is implemented on top of a CORBA compliant ORB, namely, ObjectBroker. MIND provides users a single ODMG-93 compliant common data model, and a single global query language based on SQL. This makes it possible to incorporate both relational and object oriented databases into the system. Currently Oracle 7, Sybase and METU OODBMS (MOOD) have been incorporated into MIND. The main components of MIND are a global query processor, a global transaction manager, a schema integrator, interfaces to supported database systems and a user graphical interface.In MIND all local databases are encapsulated in a generic database object with a well defined single interface. This approach hides the differences between local databases from the rest of the system. The integration of export schemas is currently performed manually by using an object definition language (ODL) which is based on OMGs interface definition language. The DBA builds the integrated schema as a view over export schemas. the functionalities of ODL allow selection and restructuring of schema elements from existing local schemas.MIND global query optimizer aims at maximizing the parallel execution of the intersite joins of the global subqueries. Through MIND global transaction manager, the serializable execution of the global transactions are provided.

Sleep spindles detection using short time Fourier transform and neural networks

Sleep spindles are 2 hallmark of the stage 2 sleep. Their distribution over the non-REM sleep is clinically important. In this paper, a method that detects the sleep spindles in sleep EEG is proposed. Short time Fourier transform is used for feature extraction. Both multilayer perceptron and Support Vector Machine are utilized in detection of the spindles in sleep EEG for comparison. The classification performance of MLP is found to be 88.7% and that of SVM as 95.4%. It should be noted that there might be differences also in visual scoring by experts, so the results obtained are quite satisfactory.

Texture-Based Airport Runway Detection

The automatic detection of airports is essential due to the strategic importance of these targets. In this letter, a runway detection method based on textural properties is proposed since they are the most descriptive element of an airport. Since the best discriminative features for airport runways cannot be trivially predicted, the Adaboost algorithm is employed as a feature selector over a large set of features. Moreover, the selected features with corresponding weights can provide information on the hidden characteristics of runways. Thus, the Adaboost-based selected feature subset can be used for both detecting runways and identifying their textural characteristics. Thus, a coarse representation of possible runway locations is obtained. The performance of the proposed approach was validated by experiments carried on a data set of large images consisting of heavily negative samples.

Formalization of Workflows and Correctness Issues in the Presence of Concurrency

In this paper, main components of a workflow system that are relevant to the correctness in the presence of concurrency are formalized based on set theory and graph theory. The formalization which constitutes the theoretical basis of the correctness criterion provided can be summarized as follows:-Activities of a workflow are represented through a notation based on set theory to make it possible to formalize the conceptual grouping of activities.-Control-flow is represented as a special graph based on this set definition, and it includes serial composition, parallel composition, conditional branching, and nesting of individual activities and conceptual activities themselves.-Data-flow is represented as a directed acyclic graph in conformance with the control-flow graph.The formalization of correctness of concurrently executing workflow instances is based on this framework by defining two categories of constraints on the workflow environment with which the workflow instances and their activities interact. These categories are:-Basic constraints that specify the correct states of a workflow environment.-Inter-activity constraints that define the semantic dependencies among activities such as an activity requiring the validity of a constraint that is set or verified by a preceding activity.Basic constraints graph and inter-activity constraints graph which are in conformance with the control-flow and data-flow graphs are then defined to represent these constraints. These graphs are used in formalizing the intervals among activities where an inter-activity constraint should be maintained and the intervals where a basic constraint remains invalid.A correctness criterion is defined for an interleaved execution of workflow instances using the constraints graphs. A concurrency control mechanism, namely Constraint Based Concurrency Control technique is developed based on the correctness criterion. The performance analysis shows the superiority of the proposed technique. Other possible approaches to the problem are also presented.

Design and Implementation of a Distributed Workflow Management System: METUFlow

Workflows are activities involving the coordinated execution of multiple tasks performed by different processing entities, mostly in distributed heterogeneous environments which are very common in enterprises of even moderate complexity. Centralized workflow systems fall short to meet the demands of such environments.