Density nodule detection in 3-D digital images
View Patent ↗An algorithm is quickly scans a digital image volume to detect density nodules. A first stage is based on a transform to quickly highlight regions requiring further processing. The first stage operates with somewhat lower sensitivity than is possible with more detailed analyses, but operates to highlight regions for further analysis and processing. The transform dynamically adapts to various nodule sizes through the use of radial zones. A second stage uses a detailed gradient distribution analysis that only operates on voxels that pass a threshold of the first stage.
1. A method of detecting regions of interest in an image composed of volume elements (voxels) comprising the steps of:
initializing a multi-dimensional cumulative array:
determining a region of potential overlap for gradient vectors associated with edges in each of a plurality of image voxels, each said region being located in the cumulative array relative to a position that corresponds to that of an edge-containing voxel in the image;
accumulating in the cumulative array values associated with said regions of potential overlap; and
processing information contained in the cumulative array to identify the regions of interest.
2. The method of claim 1 wherein the step of determining a region of potential overlap is performed for substantially every image voxel having an edge.
3. The method of claim 1 wherein the step of determining a region of potential overlap comprises the steps of:
transforming gradient information for each of a plurality of image voxels to create transform outputs; and
projecting the transform outputs into regions of potential overlap, each of such regions of potential overlap representing a range of probabilities concerning the direction of a gradient vector derived from the gradient information.
4. The method of claim 3 wherein the step of accumulating values associated with said regions of potential overlap comprises the step of accumulating the projected transform outputs for each of the regions of potential overlap in a cumulative array having array locations associated with the regions of potential overlap.
5. The method of claim 4 wherein the cumulative array is incremented at array locations defined by a predetermined pattern extending from a voxel location and oriented in the direction of the gradient vector.
6. The method of claim 5 wherein said predetermined pattern subtends an angle at a voxel location and extends between a first radial distance and a second radial distance from said location.
7. The method of claim 6 wherein the first radial distance and the second radial distance are selected according to the expected size of the region of interest.
8. The method of claim 4 wherein the step of accumulating said regions of potential overlap comprises the step of incrementing by a first amount the array locations of the cumulative array associated with the regions of potential overlap for each of the image voxels.
9. The method of claim 3 wherein the gradient information is transformed by a Hough transform.
10. The method of claim 3 wherein the gradient information is obtained by a means of a Sobel operator.
11. The method of claim 1 wherein the cumulative array has a coordinate system identical to that of the image.