Nodule detection with false positive reduction

Patent metadata

Field	Value
Publication number	US-10346728-B2
Application number	US-201715794369-A
Country	US
Kind code	B2
Filing date	Oct 26, 2017
Priority date	Oct 26, 2017
Publication date	Jul 9, 2019
Grant date	Jul 9, 2019

Abstract

In some examples, a system may train a false positive reduction machine learning model (MLM) for nodule detection. The system may receive training data images including negative images and positive images, along with an indication of nodule locations in the positive images. The system may determine elliptical approximations for nodules in the positive images, and may determine respective binarized contours from the elliptical approximations. Further, the system may determine an elliptical approximation space for the binarized contours, and may determine a subspace angle between individual image samples in the positive images and the elliptical approximation space as at least one feature of the MLM. Subsequently, when applying the MLM during nodule detection, one or more images may be input to the MLM to determine whether an indication of a nodule is correct, and if so, a visualization of a location of the nodule may be provided.

First claim

What is claimed:

1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media maintaining executable instructions, which, when executed by the one or more processors, program the one or more processors to perform operations comprising:
training a false positive reduction machine learning model, wherein the false positive reduction machine learning model utilizes at least one feature based on determining a subspace angle between an elliptical approximation in a selected image and an elliptical approximation space determined for a plurality of image samples;
receiving a plurality of images for performing nodule identification;
inputting the plurality of images to a nodule detection machine learning model to perform preliminary detection of a nodule in one or more images of the plurality of images;
inputting the one or more images and nodule location information from the nodule detection machine learning model to the false positive reduction machine learning model, wherein the false positive reduction machine learning model outputs an indication of correctness of the preliminary detection of the nodule; and
based on the false positive reduction machine learning model indicating that the preliminary detection of the nodule is correct, sending enhanced image information to a visualization program, the enhanced image information including at least an indication of a location of the nodule in the one or more images.

2. The system as recited in claim 1 , wherein the operation of training the false positive reduction machine learning model further comprises:
receiving training data images and associated indications of nodule locations;
determining a first set of the training data images that include indications of nodule locations and a second set of the training data images that do not include an indication of nodule locations; and
determining a plurality of training data sets from the training data images, each training data set including a plurality of images from the first set and a plurality of images from the second set.

3. The system as recited in claim 2 , wherein the operation of determining the plurality of training data sets further comprises:
determining relative locations of respective images of the second set in an image volume;
performing Voronoi tessellation on the respective locations of the respective images determine a plurality of Voronoi polygons; and
determining a sampling frequency for performing informed sampling of the second set based on respective volumes of the plurality of Voronoi polygons to obtain a sampling of the training data images of the second set for the plurality of data sets.

4. The system as recited in claim 3 , wherein the operation of determining a sampling frequency for performing informed sampling of the second set based on respective volumes of the plurality of Voronoi polygons further comprises:
determining a histogram based on the respective volumes of the plurality of Voronoi polygons, the histogram including a plurality of bins corresponding to different ranges of the respective volumes; and
determining respective sampling frequencies for individual bins, wherein the respective sampling frequencies are inversely proportional to heights of the individual bins.

5. The system as recited in claim 2 , wherein the operation of determining the plurality of training data sets further comprises performing data augmentation using a selected image from the first set of training data images by at least one of:
translation of the selected image along a horizontal and/or vertical direction;
flipping the selected image along the horizontal and/or vertical axis;
zooming and/or cropping the selected image;
rotating the selected image to an angle; and/or
rescaling voxel intensity of the selected image with a factor.

6. The system as recited in claim 1 , wherein the operation of training the false positive reduction machine learning model further comprises:
determining elliptical approximations for nodules in images in the first set;
determining respective binarized contours from the elliptical approximations;
determining an elliptical approximation space for the binarized contours in the first set; and
determining a subspace angle between an elliptical approximation determined for a selected image of the first set and the elliptical approximation space.

7. The system as recited in claim 1 , wherein the false positive reduction machine learning model further includes at least one additional feature comprising:
a grayscale distribution feature;
a geometry feature;
a histogram of oriented gradients feature;
a Hessian matrix based feature; or
a local binary pattern feature.

8. The system as recited in claim 1 , the operations further comprising:
receiving, via a user interface presented on a user device, an indication of the plurality of images for performing the nodule identification; and
sending, to the user device, the enhanced image information to cause, at least in part, the user device to present the enhanced image information on the user interface to provide a visual indication of the location of the nodule.

9. The system as recited in claim 1 , wherein the false positive reduction machine learning model comprises a gradient boosting model, the operation of training the false positive reduction machine learning model further comprising tuning parameters of the gradient boosting model using a randomized search technique.

10. A method comprising:
receiving, by one or more processors, training data images and associated indications of nodule locations;
determining, by the one or more processors, a first set of the training data images that include indications of nodule locations and a second set of the training data images that do not include an indication of nodule locations;
determining, by the one or more processors, elliptical approximations for nodules in images in the first set;
determining, by the one or more processors, an elliptical approximation space for the elliptical approximations in the first set;
determining, by the one or more processors, a subspace angle between individual image samples in the first set and the elliptical approximation space as at least one feature of a machine learning model;
receiving, by the one or more processors, one or more images for performing nodule identification;
inputting, by the one or more processors, the one or more images to the machine learning model to determine whether an indication of a nodule is accurate; and
based on an output of the machine learning model indicating that the indication of the nodule is accurate, associating, by the one or more processors, enhanced image information with the at least one image for providing a visualization of a location of the nodule.

11. The method as recited in claim 10 , further comprising determining, by the one or more processors, a training data set including images from the first set and images from the second set by:
determining relative locations in an image volume of respective images in the second set of images;
performing Voronoi tessellation on the relative locations of the respective images to determine a plurality of Voronoi polygons; and
determining a sampling frequency for performing informed sampling of the second set based on respective volumes of the plurality of Voronoi polygons.

12. The method as recited in claim 11 , further comprising determining the sampling frequency for performing the informed sampling of the second set based on the respective volumes of the plurality of Voronoi polygons by:
determining a histogram based on the respect

Assignees

• Hitachi Ltd

Inventors

• Fu Maojing
• Tang Hsiu-Khuern
• Mehta Abhay

Classifications

• G06V30/1916 Primary

  Validation; Performance evaluation · CPC title

• G06V30/19173

  Classification techniques · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06F18/2411

  based on the proximity to a decision surface, e.g. support vector machines · CPC title