Automated Localization of Solid Organs in 3D CT Images: A Majority Voting Algorithm Based on Ensemble Learning

This paper describes a new approach to automatically find out the location of a target solid organ in 3D CT scans. Specifically, our goal is to detect a 3D rectangular box for the target organ in a way that this rectangle bounds the organ region tightly and accurately. The proposed approach combines the ensemble learning and the majority voting techniques to achieve a robust detection by using a small number of CT scans for training. A database including 3,329 torso CT scans is used in experiments. Among them, we manually label the heart and the left/right kidneys from nearly 100 3D CT scans as training samples, and use the proposed approach to localize those organs in the other CT scans. Experimental results show that detection rates are 99% for the heart, 85%-87% for the right and left kidney, with a computation time less than 15 seconds per CT scan on a general PC.