Adaptive local window-based methods for characterizing features of interest in digital images and systems for practicing same
AssigneeThe Board of Trustees of the Leland Stanford Junior University
InventorsAssaf Hoogi, Daniel L. Rubin
Life of this patent
Estimating the term…
Abstract
Provided are methods for characterizing a feature of interest in a digital image. In certain aspects, the methods use an adaptive local window and include obtaining an initial contour for a feature of interest, defining a region of interest around the contour, and segmenting the feature of interest by iteratively selecting a size of a local window surrounding each point on the contour. Non-transitory computer readable media and systems that find use in practicing the methods of the present disclosure are also provided.
Claims20
A computer-implemented method for characterizing a feature of interest in a digital image, using an adaptive local window, comprising the steps of:
- (a) obtaining an initial contour for a feature of interest, the initial contour formed of contour points;
- (b) defining a region of interest around the contour;
- and
- (c) segmenting the feature of interest by iteratively selecting a size of a local window surrounding each point on the contour, wherein the local window size is determined based on (i) homogeneity, (ii) contrast surrounding each contour point, (iii) approximate lesion size, and (iv) an energy function, wherein the energy decreases as the segmenting progresses.
A computer-implemented method for characterizing a feature of interest in a digital image, using a calculated adaptive level set method, comprising:
- (a) obtaining an initial contour for a feature of interest;
- and
- (b) producing an evolved contour towards the boundary of the feature of interest, wherein the producing comprises determining weighting parameters of a level set function using a convolutional neural network (CNN), wherein the weighting parameters are determined based on:
- output probabilities from the CNN that are probabilities of the initial contour being inside the feature of interest and far from its boundaries, outside the feature of interest and includes it and far from its boundaries, or close to a boundary of the feature of interest;
- wherein the weighting parameters control the direction and magnitude of contour evolution,
- to characterize the feature of interest.
Dependencies are inferred from the claim text.