Systems and Method for Computation and Visualization of Segmentation Uncertainty in Medical Images

Patent metadata

Field	Value
Publication number	US-2016267673-A1
Application number	US-201514642914-A
Country	US
Kind code	A1
Filing date	Mar 10, 2015
Priority date	Mar 10, 2015
Publication date	Sep 15, 2016
Grant date	—

Abstract

Systems and methods for computing uncertainty include generating a surface model of a target anatomical object from medical imaging data of a patient. Uncertainty is estimated at each of a plurality of vertices of the surface model. The uncertainty estimated at each of the plurality of vertices is visualized on the surface model.

First claim

1 . A method for computing uncertainty, comprising:
generating a surface model of a target anatomical object from medical imaging data of a patient; estimating uncertainty at each of a plurality of vertices of the surface model; and visualizing the uncertainty estimated at each of the plurality of vertices on the surface model.

2 . The method as recited in claim 1 , wherein estimating uncertainty at each of the plurality of vertices of the surface model further comprises:
estimating a range of acceptable points along a surface normal for each of the plurality of vertices; and fitting a respective uncertainty distribution to the range of acceptable points estimated for each of the plurality of vertices.

3 . The method as recited in claim 2 , wherein the range of acceptable points correspond to a probability distribution indicating a probability that each point in the range of acceptable points accurately identifies an image boundary.

4 . The method as recited in claim 2 , wherein estimating the range of acceptable points along the surface normal for each of the plurality of vertices further comprises:
detecting the range of acceptable points for each of the plurality of vertices using a trained classifier.

5 . The method as recited in claim 2 , wherein fitting the respective uncertainty distribution to the range of acceptable points estimated for each of the plurality of vertices comprises:
fitting a respective Gaussian distribution to the range of acceptable points estimated for each of the plurality of vertices.

6 . The method as recited in claim 5 , wherein a mean of the respective Gaussian distribution is at the vertex and a standard deviation defines a confidence interval for the vertex.

7 . The method as recited in claim 2 , further comprising:
refining the uncertainty by defining a minimum and a maximum for the range of acceptable points for at least one of the plurality of vertices.

8 . The method as recited in claim 1 , wherein generating the surface model of the target anatomical object further comprises:
segmenting the target anatomical object using marginal space learning based segmentation.

9 . The method as recited in claim 8 , wherein segmenting the target anatomical object using marginal space learning based segmentation further comprises:
detecting a bounding box encapsulating the target anatomical object in the medical imaging data using marginal space learning; detecting anatomical landmarks of the target anatomical object within the bounding box; and fitting a surface model of the target anatomical object to the detected anatomical landmarks.

10 . The method as recited in claim 1 , wherein visualizing the uncertainty estimated at each of the plurality of vertices on the surface model further comprises:
displaying a visualization of the surface model with a color representing a level of uncertainty for each of the plurality of vertices, wherein the level of uncertainty for each of the plurality of vertices is based on a standard deviation of a respective uncertainty distribution estimated for each of the plurality of vertices.

11 . The method as recited in claim 1 , further comprising:
calculating a measurement of the patient based on the surface model and the uncertainty.

12 . The method as recited in claim 11 , wherein calculating the measurement of the patient based on the surface model and the uncertainty comprises at least one of:
calculating a range of the measurement based on the surface model and the uncertainty; and calculating a mean value and a standard deviation value for the measurement with an associated confidence interval based on the surface model and the uncertainty.

13 . An apparatus for computing uncertainty, comprising:
means for generating a surface model of a target anatomical object from medical imaging data of a patient; means for estimating uncertainty at each of a plurality of vertices of the surface model; and means for visualizing the uncertainty estimated at each of the plurality of vertices on the surface model.

14 . The apparatus as recited in claim 13 , wherein the means for estimating uncertainty at each of the plurality of vertices of the surface model further comprises:
means for estimating a range of acceptable points along a surface normal for each of the plurality of vertices; and means for fitting a respective uncertainty distribution to the range of acceptable points estimated for each of the plurality of vertices.

15 . The apparatus as recited in claim 14 , wherein the range of acceptable points correspond to a probability distribution indicating a probability that each point in the range of acceptable points accurately identifies an image boundary.

16 . The apparatus as recited in claim 14 , wherein the means for estimating the range of acceptable points along the surface normal for each of the plurality of vertices further comprises at least one of:
means for detecting the range of acceptable points for each of the plurality of vertices using a trained classifier.

17 . The apparatus as recited in claim 14 , wherein the means for fitting the respective uncertainty distribution to the range of acceptable points estimated for each of the plurality of vertices comprises:
means for fitting a respective Gaussian distribution to the range of acceptable points estimated for each of the plurality of vertices.

18 . The apparatus as recited in claim 14 , further comprising:
means for refining the uncertainty by defining a minimum and a maximum for the range of acceptable points for at least one of the plurality of vertices.

19 . The apparatus as recited in claim 13 , wherein the means for generating the surface model of the target anatomical object further comprises:
means for segmenting the target anatomical object using marginal space learning based segmentation.

20 . The apparatus as recited in claim 19 , wherein the means for segmenting the target anatomical object using marginal space learning based segmentation further comprises:
means for detecting a bounding box encapsulating the target anatomical object in the medical imaging data using marginal space learning; means for detecting anatomical landmarks of the target anatomical object within the bounding box; and means for fitting a surface model of the target anatomical object to the detected anatomical landmarks.

21 . The apparatus as recited in claim 13 , wherein the means for visualizing the uncertainty estimated at each of the plurality of vertices on the surface model further comprises:
means for displaying a visualization of the surface model with a color representing a level of uncertainty for each of the plurality of vertices, wherein the level of uncertainty for each of the plurality of vertices is based on a standard deviation of a respective uncertainty distribution estimated for each of the plurality of vertices..

22 . The apparatus as recited in claim 13 , further comprising:
means for calculating a measurement of the patient based on the surface model and the uncertainty.

23 . The apparatus as recited in claim 22 , wherein the means for calculating the measurement of the patient based on the surface model and the uncertainty comprises at least one of:
means for calculating a range of the measurement based on the surface model and the uncertainty; and means for calculating a mean value and a standard deviation value for the measurement with an associat

Assignees

• Siemens Ag

Inventors

• Grbic Sasa
• Mansi Tommaso
• Voigt Ingmar
• Georgescu Bogdan
• Florin Charles Henri
• Comaniciu Dorin

Classifications

• G06F18/285 Primary

  Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system · CPC title

• G06T2207/10072

  Tomographic images · CPC title

• G06T2207/20192

  Edge enhancement; Edge preservation · CPC title

• G06T2210/12

  Bounding box · CPC title

• G06F19/321

  Physics · mapped topic