Automatic segmentation of bone tumors in computed tomography images: Case study of calcaneus bone

Computer Aided Diagnosis (CAD) is a multi-disciplinary field of study in which brings the second opinion to physicians by processing medical images with image processing techniques. CAD includes various stages as image processing including computer algorithms, image analysis and data classification. Thus interpretation capability of physicians can be improved and it results by increasing accuracy of diagnosis and repeatability. Radiologic diagnosis and reporting depend on the physician's experience and misdiagnosis may occur due to non-objective and error-prone to evaluation of physicians. This presented thesis has been prepared for 3D modelling and determining tumor and tumor-like lesions on the calcaneus bone by CAD from computed tomography images. In this study, nine patients and 869 MDCT (Multi-detector Computed Tomography) retrospective foot images were used. The used parameters are (i) detector collimation of 64, (ii) scanning thickness of 0.5-3mm, (iii) pixel sizes of 512x512 in radiometric resolution of 16 bits' gray levels. Axial images have been obtained in DICOM (Digital Imaging and Communications in Medicine) format. For detection of tumor and similar lesions from MDCT images, 3D-Doctor software and Matlab (R2010) program were used and images were processed. The study consists of five main image processing steps. These are; (i) noise reduction, (ii) segmentation, (iii) applying of morphological operators, (iv) 3D modelling, (v) statistical analysis and interpretation. Matlab platform was used to apply Watershed and Canny Algorithms because of their capabilities compare to other methods. The 3D-Doctor software was preferred for interactive segmentation and 3D modeling steps. For evaluation of obtained results, four different performance/accomplishment criteria "accuracy, sensitivity, specificity, and F-measure" were used. Quantitative achievement of Segmentation methods and comparable analysis were obtained as Interactive Segmentation> Watershed Algorithm> Canny Algorithm respectively. The performance of the study was evaluated by ROC (Receiver Operating Characteristic Analysis). CAD systems may serve as an auxiliary member to increase the accuracy and speed of radiologists in detecting of tumors and tumor-like lesions in the calcaneus bone from MDCT images. It is thought that obtained results of the study can contribute to existing information with biometric and reconstruction technique. Keywords: Digital image processing, medical photogrammetry, CAD, segmentation, CT, calcaneus, tumor.