Automatic Detection and Segmentation of Kidneys in 3D CT Images Using Random Forests

Abstract

Abstract. Kidney segmentation in 3D CT images allows extracting useful information for nephrologists. For practical use in clinical routine, such an algorithm should be fast, automatic and robust to contrast-agent enhancement and fields of view. By combining and refining state-of-theart techniques (random forests and template deformation), we demonstrate the possibility of building an algorithm that meets these requirements. Kidneys are localized with random forests following a coarseto-fine strategy. Their initial positions detected with global contextual information are refined with a cascade of local regression forests. A classification forest is then used to obtain a probabilistic segmentation of both kidneys. The final segmentation is performed with an implicit template deformation algorithm driven by these kidney probability maps. Our method has been validated on a highly heterogeneous database of 233 CT scans from 89 patients. 80 % of the kidneys were accurately detected and segmented (Dice coefficient > 0.90) in a few seconds per volume.

Mediastinal lymph node detection and station mapping on chest CT using spatial priors and random forest

“…It has been successfully applied to various organ localization and segmentation tasks in different imaging modalities. [56][57][58][59] Finally, for each detected lymph node, with the segmentation of anatomic structures, its station is automatically determined according to the IASLC reference.…”

Semi-automatic Liver Tumor Segmentation in Dynamic Contrast-Enhanced CT Scans Using Random Forests and Supervoxels

“…In the context of tumor segmentation, methods relying on random decision forest such as [7,10] usually focus on one single image and therefore involve single-phase visual features θ(v ). During training, each tree takes the training voxel set S as input and optimizes its own internal nodes ({τ low , τ up ,θ(v )}) via information gain maximization [9] to obtain the most discriminative binary tests with respect to S .…”

“…2.1), voxels are acting without inter-dependencies which may result in a lack of spatial consistency regarding classification results. Features are sometimes explicitly related to spatial context [7,10] but spatial extent remains limited. A-posteriori regularization techniques such as conditional random field (CRF) [11] would more accurately introduce spatial constraints but it strongly increases computational complexity.…”

FPGA-Based Portable Ultrasound Scanning System with Automatic Kidney Detection

“…The organs in the images are detected by manually cropping along the contours [22], placing the landmarks on the contour [23] and marking the bounding boxes around the kidney [24]. Automatic detection of kidney in CT slices using random forest has been used in [25] for reconstructing the 3D structure of kidney. A two-stage detection algorithm is employed for automatic detection of lymph nodes in CT data; Haar features with the Adaboost cascade classifier are used in the first stage, and in the second stage, self-assigning features with the Adaboost cascade classifier are used [26].…”