CLASSIFICATION AND SEGMENTATION OF GROUND GLASS OPACITY (GGO) NODULES
Publication Date: 2015-Jul-23
The IP.com Prior Art Database
This disclosure proposes a technique for calculating an adaptive threshold for classification and segmentation of ground glass opacity (GGO) nodules. The technique uses median statistics to classify and segment the GGO nodules. The technique uses median and median absolute deviation (MAD) which is computed locally around the nodules to classify type of nodule as solid / part-solid / nonsolid nodule and further to segment the nodules. The technique also ensures adaptive computation of the statistics for accurate estimation.
The present invention relates generally to diagnostic imaging, and more particularly to a technique for classifying and segmenting ground glass opacity (GGO) nodules using median statistics.
In radiographic examinations, ground glass opacity (GGO) nodules generally refer to radiographic appearances of misty lung opacities not associated with an obscuration of underlying vessels. The GGO nodules are typically found in two forms, pure and mixed. Pure GGO nodules do not include solid components, whereas mixed GGO nodules include some solid components.
Conventionally, Otsu method and region growing algorithms are used for segmenting the GGO nodules. However, such method and algorithms are not tuned to segment both part-solid and non-solid nodules and mostly address segmentation of pure ground glass nodules or pure non-solid nodules. Further, presence of solid voxels and Hounsfield Unit (HU) of ground glass region very close to lung parenchyma makes the segmentation of the nodules more challenging.
Some conventional algorithms use average and standard deviation to compute thresholds and are therefore very sensitive to outliers. In case of part solid nodules, especially in contrast scans, solid voxels have high HU value. The high HU value skews the distribution and hence computed average and standard deviation is also biased towards such values. Therefore, the thresholds computed using such biased statistics turns out to be very high for ground glass region. Some other conventional algorithms utilize intensity based thresholding means, but fail to segment non-solid nodules.
It would be desirable to have an improved technique for classifying and segmenting the GGO nodules.
BRIEF DESCRIPTION OF DRAWINGS
Figure 1 depicts a workflow for analyzing a case with a lung nodule assessment application.
Figure 2 depicts a computed tomography lung image showing the presence of ground glass opacity (GGO) nodule.
Figure 3 depicts a histogram of the lung volume shown in Figure 2.
Figure 4 depicts a flow diagram to compute thresholds used in nodule segmentation.
Figure 5 depicts the user selected nodule and the sub-volume computed around a seed point.
This disclosure proposes a technique for classifying and segmenting ground glass opacity (GGO) nodules using median statistics. Figure 1 depicts a workflow for analyzing a case with a lung nodule assessment application. The technique loads and displays a volume on a viewer enabling a user to select a point on a nodule. A first estimate of nodule consistency is obtained. Subsequently, when the user switches to analysis mode, the technique segments the nodule with two contours based on the nodule consistency. The nodule is segmented into two contours for part solid and single contour for non-solid module. A consistency determination algorithm is described in detail below.