Evolution of multiorgan segmentation techniques from traditional to deep learning in abdominal CT images – A systematic review

Abstract

Abdominal organ segmentation is the crucial research direction in computer assisted diagnostic systems. Segmentation of multiple organs in medical images is known as Multiorgan segmentation. It is a widespread subject of research in the realm of medical image analysis. The purpose of this study is to provide the comprehensive systematic literature review on segmentation of multiple organs in abdomen CT scans. This paper focuses on the progression of state-of-art methods from traditional techniques to deep learning models. Firstly, the methods are classified into three categories: atlas based, statistical shape models and deep learning models. Secondly, research is carried out to determine which organs require more attention. The liver, kidney, and spleen are the most often selected organs, whereas the esophagus, duodenum, and portal vein are rarely picked. When medical images are taken into account for research, datasets play a vital role. This paper sheds light on publicly available datasets along with their size, no of organ classes and, related challenges which make the current study more effective and useful for the researchers in the same field. Further, evaluation metrics along with their scope and characteristics are presented. We conclude with a discussion of challenges and future directions which will open pathways for researchers. Based on the surveyed research papers, Dense-Net came out as an optimal choice. Recently, the standard practice in multi organ segmentation is two step deep learning models in sequential manner, which can take leverage of two models.