A fast recognition algorithm for liver tissue segmentation from CT scans

Purpose: The purpose of this work is to provide the surgeons an efficient tool for liver segmentation and visualization and help them by operation planning. Our semi-automatic approach is a fast way to segment liver from CT data and obtain its 3D reconstruction. The information, extracted by means of CT-data analysis can be used for precise estimation of such operational risks as vessel injuries, blood loss during the operation and lesion relapse as well as for instrument navigation within the surgical intervention. Methods: Our approach is a combination of fast and efficient segmentation techniques and minimal user interactions via target-oriented interface. Among the segmentation techniques region growing, histogram analysis and object selection rules are in use. Results: A set of 18 oncological patient datasets (2843 original CT-images with average dimensions 320×320 pixels) and reference segmentation from medical radiologists were used for the evaluation of algorithm performance. Three evaluation methods were applied for estimation of segmentation quality: average symmetric surface distance, Dice similarity coefficient as volume overlap measure and binary classification test. The mean average symmetric surface distance was 2.34 mm. The mean sensitivity and specificity were 0.95 and 0.98 respectively. Average volume overlap was 94%. Average processing time was 1.8 seconds per dataset (11.5 milliseconds per slice). Conclusions: All obtained results are comparable with best results of other works, excepting processing time, which was considerably reduced. This makes our algorithm usable in real time in clinical routine.