3D lymphoma segmentation in PET/CT images based on fully connected CRFs

Abstract

Positron Emission Tomography (PET) is widely used for lymphoma detection. It is often combined with the CT scan in order to provide anatomical information for helping lymphoma detection. Two common types of approaches can be distinguished for lymphoma detection and segmentation in PET. The first one is ROI dependent which needs a ROI defined by physicians who firstly detect where lymphomas are. The second one is based on machine learning methods which need a large learning database. However, such a large standard database is quite rare in medical field. Considering these problems, we propose a new approach which combines a multi-atlas segmentation of the CT with CRFs (Conditional Random Fields) segmentation method in PET. It consists of 3 steps. Firstly, an anatomical multi-atlas segmentation is applied on CT to locate and remove the organs having hyper metabolism in PET. Secondly, CRFs detect and segment the lymphoma regions in PET. The conditional probabilities used in CRFs are usually estimated by a learning step. In this work, we propose to estimate them in an unsupervised way. A list of the detected regions in 3D is visualized. The final step is to select real lymphomas by simply clicking on them. Our method is tested on ten patients. The rate of good detection is 100%. The average of Dice index over 10 patients for measuring the lymphoma is 80% compared to manual lymphoma segmentation. Comparing with other methods in terms of Dice index shows the best performance of our method.