Hamza Osman Ilhan

The evaluation of detectors and descriptors on determination of semen cell

Detectors and descriptors refer to the key points of the images where informative features can be detected or extracted to use in machine learning for classification or clustering problems. Four algorithms as two descriptor (SURF, MRES) and two detectors (Harris, Shi Tomasi) are utilized for semen cell detection problem in this paper. Results emphasize the best algorithm for future studies to use in tracking or morphological analysis of semen. The evaluation of algorithms is carried out on manually labeled images over a pre-defined verification area. Not only accuracy is measured owing to data imbalance problem, but also f-measure scores are registered to indicate the methods success rates. As a summary of paper, SURF surpasses over other methods with 87.67% accuracy rate and 0.92 F-measure score owing to the scale invariant method.

A novel data acquisition and analyzing approach to spermiogram tests

Abstract Spermiogram tests are currently performed in two ways; computer assisted and visual assessment. Computerized techniques are costly and parameter dependent. Therefore, it is not preferred in many laboratories. Analysis based on visual assessment technique is subjective. In this study, we proposed a new computerized approach in which data acquisition is performed as in visual assessment, but analysis is done by computer. This approach provides more generalized results, requires less parameters due to the usage of standard counting chambers as in visual assessment technique, and cheaper than the other computerized techniques. Proposed approach includes two modules; video stabilization and motile sperm detection module. Stabilization is a requirement because it is impossible to fix proposed approach in ocular part completely. Otherwise, vibration affects the detection process of motile sperms. In this respect, feature-matching video stabilization idea is firstly utilized within the video-microscopy concept. Several feature extraction techniques were tested to sustain more stable videos in stabilization module. Motile sperm detection algorithm is then adapted to evaluate the efficiency of video stabilization and analyzing part. In case of vibrated frame sequence, the detection of total motile sperms concludes immediate peak values while it is around average values in stable frames. Samples were also evaluated by an expert with the visual assessment technique. Comparative tables and figures emphasize that the proposed approach can be employed in laboratories due to the high correlation with the visual assessment results.

Detection and estimation of down syndrome genes by machine learning techniques

Down syndrome is accepted as the common birth defect in population and diagnosed as more physical development with less cognitive activity than an average human. Early diagnosis of disease play important role for the patient future life. Computer aided systems, in terms of artificial intelligence, results more accurate and consistent diagnosis in the detection and estimation of down syndrome genes compare to doctor decisions. In this study, detection and estimation of down syndrome disease is maintained by analyzing the protein levels in genes. In this sense, a Decision Support System based on machine learning techniques are proposed to estimate the down syndrome automatically. Additionally, another technique named as Principal Component Analyses are performed to eliminate multi proteins in genes into fewer number to achieve the same success with less information.

Sleep stage classification by ensemble learning methods with active sample selection techniques

In medical science, sleep stages are the main criteria to define the disorders and have crucial role on diagnostic. In this sense, accurate sleep stage classification plays important role due to provide better report on medications and diagnoses. In this study, EEG signals are classified by a rule based machine learning algorithm; Decision Tree with the ensemble and classical machine learning idea. Additionally, two of active sample selection technique using the idea of strictly separated discrimination and margin distances are applied on learning processes to obtain more accurate results with less samples. This paper proves that ensemble learning algorithms with one of the implemented active sample selection technique gives more successful result on the determination of stages.

The mesothelioma disease diagnosis with artificial intelligence methods

Asbestos is a carcinogenic substance, and threatens human health. Malignant Mesothelioma disease is one of the most dangerous kind of cancer caused by asbestos mineral. The most common symptom of the disease, progressive shortness of breath and constant pain. Early treatment and diagnosis are necessary. Otherwise, the disease can lead people to die in a short period of time. In this paper, different types of artificial intelligence methods are compared for effective Malignant Mesotheliomas diseases classification. Support Vector Machine, Neural Network and Decision Tree methods are selected in terms of regular machine learning concept. Additionally, Bagging and Adaboost re-sampling within ensemble learning terminology is also adapted. Totally 324 Malignant Mesothelioma data which consists of 34 features is used in this study. K-fold cross-validation technique is performed to compute the performance of the algorithms with different K values. 100% classification accuracies are obtained from three tested methods; Support Vector Machine, Decision Tree and Bagging. Additionally, the process time of methods are measured in case of using method in lots of data. In this sense, methods are evaluated based on accuracy and time complexity. The results of this paper are also compared with previous studies using same Malignant Mesotheliomas dataset.