Biyomedikal görüntülerde hareketli nesnelerin analizi ve takibi için bir sistem gerçeklemesi

Spermiogram is the first step of the infertility diagnosis. Computer Aided Sperm Analysis (CASA) and Visual Assessment (VA) are two evaluation techniques employed in spermiogram analyses. The VA is carried out by manually observing the sperm on counting chambers. Hence, analysis and diagnosis strongly depend on the skills and experiences of the observers. On the other hand, the CASA is a more advanced technology due to the improved computerized techniques and minimization of human intervention. However, it is more expensive than VA since it is an integrated computer based system and requires exhaustive parameter setting process. In this thesis, we aim to develop a combinational approach using the smartphone and computer for the sperm concentration and motility analysis. Smartphone was utilized to obtain images similar to the VA technique. The acquired samples were analyzed by using computerized methods to eliminate the observer variability. In this thesis, a software named as Computerized Sperm Counting and Trajectory Analyzing Software (CSCTAS) for automatically counting and tracking the sperm over one of the commonly using counting chamber, is proposed. Proposed software consists of seven modules executed sequentials: (1) Data Acquisition and Organization, (2) Automatic Grid and Region of Interest (ROI) detection and extraction, (3) Video Stabilization, (4) Motile/Immotile Spermatozoon Detection, (5) Spermatozoa Counting, (6) Motile spermatozoa tracking, and (7) Trajectory Classification. Each module consists of various combination of image processing techniques. Firstly, data acquisition and organization were performed using a novel approach to provide inexpensive design contrary to traditional CASA systems. Secondly, Region of Interest (ROI) extraction was realized by a combinational approach of line detection and segmentation methods. Then, feature matching based video stabilization was introduced to eliminate the vibrations occurred during the data acquisition step. In this respect, different descriptors were tested. The fourthly, Background and foreground extraction techniques were employed in immotile and motile spermatozoon detection process, respectively. Additionally, active contour, dual thresholding and clustering were implemented to enhance the segmentation of immotile spermatozoon in this step as well. Thereafter, detected sperms were counted by pixel based blob analysis. Motile spermatozoa were tracked by the Mean Shift and the Kalman Filters for the motility analysis. Various motility features were extracted from the trajectories to classify them into four classes. As the final step, results were reported to the users. Each module of the proposed software was individually tested. Two approaches of the automatic ROI detection and extraction steps were tested and compared on 80 videos of 20 subjects. Video stabilization idea was evaluated on 42 videos of 14 subjects. Two techniques were performed for the sperm concentration analysis. The performance of the Fuzzy C-Means based segmentation was measured on 15 videos of 5 subjects. A more advanced technique, dual thresholding and active contour based segmentation, was evaluated on 32 videos of 8 subjects. Finally, 32 videos of 8 subjects were used for the verification of the tracking technique. As a result, totally 201 semen videos obtained at different times from 55 subjects were included for the determination of the proposed spermiogram analysis approach. In the clinical research, we initilaly compared the counting results of CASA system, VA, and proposed CSCTAS with the proper concentration calculation. Normally, experts separately and manually count the motile and immotile spermatozoon within 16 and 10 squares to generalize the result as million per ml in the VA technique, respectively. According to the concentration analysis, proposed CSCTAS resulted in similar outputs as VA. It has less variation for immotile spermatozoa counting and is more efficient than the VA for the determination of specific diseases such as Asthenospermia. It is known that conventional SQA-Vision CASA is useless in the case of less than 5 million sperm cells. Therefore, presented approach is more efficient in the infertility diagnosis. In the motility analysis, CSCTAS gives the similar outputs when compared to SQA-Vision CASA. The SQA-Vision is more reliable technique when compared to VA in motility analysis because it is impossible to track single spermatozoa by eye for a period of time within other spermatozoa. Therefore the similarities between SQA-Vision and proposed CSCTAS is more meaningfull than the comparision with VA technique for the motility analysis. According to the motility analysis results, CSCTAS is also efficient, cheaper and easier to use in labs for the motility analysis when compared to the conventional CASA systems. According to the obtained concentration and motility results, the proposed smartphone based sperm analysis can be adapted with the developed with the proposed CSCTAS in laboratories. Our proposed system stands out by its modularity, functionality, accuracy and low cost. Additionally, it eliminates the human factor in VA and CASA.