Detecting and classifying lesions in ultrasound images
View Patent ↗A method for determining a candidate lesion region within an ultrasound image. The method includes the steps of: accessing a digital ultrasound image of anatomical tissue; segmenting spatially contiguous pixels in the ultrasound image into a plurality of regions in accordance with substantially similar intensity values and spatial smoothness constraints; and selecting, from the plurality of regions, one or more candidate lesion regions having an intensity value lower than a pre-determined intensity value. In one arrangement, the one or more candidate lesion region is classified into at least one of the following classes: benign, malignant, or unknown.
1. A method for detecting a lesion in an input digital ultrasound image of anatomical tissue, the method being carried out by a digital computer including a computer readable storage medium storing a computer program of the method, the method comprising steps of:
accessing the input digital ultrasound image of anatomical tissue;
segmenting spatially contiguous pixels in the input digital ultrasound image into a plurality of regions in accordance with substantially similar intensity patterns, the segmenting step including substeps of:
segmenting the input digital ultrasound image into regions by computing an intensity histogram,
partitioning the histogram into a plurality of clusters that correspond to distinctive prominent intensities in the input digital ultrasound image,
classifying each pixel of the input digital ultrasound image to the closest cluster as a function of the mean values of the intensity clusters;
placing a neighborhood window at each pixel in order to determine what neighborhood pixels are used to compute a local intensity histogram for the pixel, wherein an initial neighborhood window size is set at the size of the entire input digital ultrasound image,
recomputing local mean values of each intensity class based on the current segmentation,
reclassifying the pixels according to the recomputed local mean values of intensity classes,
repeating the recomputing and reclassifying substeps until a convergence is reached, and
repeating the placing, recomputing, reclassifying and repeating substeps for one or more reduced neighborhood window sizes, until segmentation of the input digital ultrasound image is obtained when a minimum window size is processed;
selecting, from the plurality of regions, one or more candidate lesion regions having an intensity value lower than a pre-determined intensity value; and
classifying the one or more candidate lesion regions into at least one of the following classes: benign, malignant, or unknown.
2. The method of claim 1 , wherein the step of classifying the one or more candidate lesion regions includes the steps of:
classifying the one or more candidate lesion regions into the benign or malignant class in accordance with at least one of the following features: a shape feature, an interior feature, or a boundary feature.
3. The method of claim 1 , further comprising the step of providing a confidence value for the lesion classification.
4. The method of claim 1 , further comprising the step of, prior to the step of segmenting, pre-processing the input digital ultrasound image.
5. The method of claim 4 , wherein the input digital ultrasound image comprises a lesion, and the input digital ultrasound image is pre-processed to enhance the contrast between the lesion and the normal anatomical tissue.
6. The method of claim 1 , wherein the selected candidate lesion regions have an average intensity value lower than a pre-determined intensity value.
7. A method of determining a candidate lesion region within an input digital ultrasound image of anatomical tissue the method being carried out by a digital computer including a computer readable storage medium storing a computer program of the method, the method comprising steps of:
accessing the input digital ultrasound image of anatomical tissue;
segmenting spatially contiguous pixels in the input digital ultrasound image into a plurality of regions in accordance with substantially similar intensity values and spatial smoothness constraints, the segmenting step including substeps of:
segmenting the input digital ultrasound image into regions by computing an intensity histogram,
partitioning the histogram into a plurality of clusters that correspond to distinctive prominent intensities in the input digital ultrasound image,
classifying each pixel of the input digital ultrasound image to the closest cluster as a function of the mean values of the intensity clusters;
placing a neighborhood window at each pixel in order to determine what neighborhood pixels are used to compute a local intensity histogram for the pixel, wherein an initial neighborhood window size is set at the size of the entire input digital ultrasound image,
recomputing local mean values of each intensity class based on the current segmentation,
reclassifying the pixels according to the recomputed local mean values of intensity classes,
repeating the recomputing and reclassifying substeps until a convergence is reached, and
repeating the placing, recomputing, reclassifying and repeating substeps for one or more reduced neighborhood window sizes, until segmentation of the input digital ultrasound image is obtained when a minimum neighborhood window size is processed; and
selecting, from the plurality of regions, one or more candidate lesion regions having an intensity value lower than a pre-determined intensity value.
8. The method of claim 7 , further comprising the step of, prior to segmenting, pre-processing the input digital ultrasound image such that a contrast between a lesion and normal anatomical tissue would be enhanced.
9. The method of claim 8 , where in the step of pre-processing comprises a gamma adjustment.
10. The method of claim 7 , further comprising the step of classifying the one or more candidate lesion region into at least one of the following classes: benign, malignant, or unknown.
11. The method of claim 10 , wherein the step of classifying the one or more candidate lesion regions includes the steps of:
classifying the one or more candidate lesion region into the benign or malignant class in accordance with at least one of the following features: a shape feature, an interior feature, or a boundary feature.
12. The method of claim 10 , wherein the step of classifying further comprises steps of:
extracting an outline of the selected candidate lesion region;
determining at least one feature from the group: a shape feature, an interior feature, or a boundary feature; and
classifying the selected candidate lesion region as either benign, malignant, or unknown or malignant in response to the computed at least one feature.
13. The method of claim 12 , wherein at least two features are determined for the selected candidate lesion region.
14. The method of claim 12 , wherein the shape feature is an aspect ratio of the selected candidate lesion region.
15. The method of claim 7 , further comprising the step of:
displaying the input digital ultrasound image on a display with an outline of the selected candidate lesion region.
16. A system for detecting a lesion in an input digital ultrasound image of anatomical tissue, the system including a digital computer comprising:
a computer readable storage medium having stored therein medical image data and digital data processor programs; and
a digital data processor coupled to the storage medium, whereby the processor is configured to:
access the input digital ultrasound image of anatomical tissue;
segment spatially contiguous pixels in the input digital ultrasound image into a plurality of regions in accordance with substantially similar intensity patterns; by
segmenting the input digital ultrasound image into regions by computing an intensity histogram,
partitioning the histogram into a plurality of clusters that correspond to distinctive prominent intensities in the input digital ultrasound image,
classifying each pixel of the digital image to the closest cluster as a function of the mean values of the intensity clusters,
placing a neighborhood window at each pixel in order to determine what neighborhood pixels are used to compute a local intensity histogram for the pixel, wherein an initial neighborhood window size is set at the size of the entire input digital ultrasound image,
recomputing local mean values of each intensity class based on the current segmentation,
reclassifying the pixels according to the recomputed local mean values of intensity classes,
repeating the recomputing and reclassifying substeps until a convergence is reached, and
repeating the placing, recomputing, reclassifying and repeating for one or more reduced neighborhood window sizes, until segmentation of the input digital ultrasound image is obtained when a minimum neighborhood window size is processed;
select from the plurality of regions, one or more candidate lesion regions having an intensity value lower than a pre-determined intensity value; and
classify the one or more candidate lesion regions into at least one of the following classes: benign, malignant, or unknown.
17. A method for detecting a lesion in an input digital ultrasound image of anatomical tissue, the method being carried out by a digital computer including a computer readable storage medium storing a computer program of the method, the method comprising steps of:
accessing the input digital ultrasound image of anatomical tissue;
segmenting spatially contiguous pixels in the input digital ultrasound image into a plurality of regions in accordance with substantially similar intensity patterns, the segmenting step including substeps of:
segmenting the input digital ultrasound image into regions by computing an intensity histogram,
partitioning the histogram into a plurality of clusters that correspond to distinctive prominent intensities in the input digital ultrasound image,
classifying each pixel of the input digital ultrasound image to the closest cluster as a function of the mean values of the intensity clusters;
placing a neighborhood window at each pixel in order to determine what neighborhood pixels are used to compute a local intensity histogram for the pixel, wherein an initial window size is set at the size of the entire input digital ultrasound image,
recomputing local mean values of each intensity class based on the current segmentation,
reclassifying the pixels according to the recomputed local mean values of intensity classes,
repeating the recomputing and reclassifying substeps until a convergence is reached, and
repeating the placing, recomputing, reclassifying and repeating substeps for one or more reduced neighborhood window sizes, until segmentation of the input digital ultrasound image is obtained when a minimum neighborhood window size is processed;
selecting, from the plurality of regions, one or more candidate lesion regions having an intensity value lower than a pre-determined intensity value;
eliminating from the selected candidate lesion regions any regions that touch image borders; and
classifying the one or more selected candidate lesion regions into at least one of the following classes: benign, malignant, or unknown.
18. A method of determining a candidate lesion region within an input digital ultrasound image, the method being carried out by a digital computer including a computer readable storage medium storing a computer program of the method, the method comprising steps of:
accessing the input digital ultrasound image of anatomical tissue;
segmenting spatially contiguous pixels in the input digital ultrasound image into a plurality of regions in accordance with substantially similar intensity values and spatial smoothness constraints, the segmenting step including substeps of:
segmenting the input digital ultrasound image into regions by computing an intensity histogram,
partitioning the histogram into a plurality of clusters that correspond to distinctive prominent intensities in the input digital ultrasound image,
classifying each pixel of the input digital ultrasound image to the closest cluster as a function of the mean values of the intensity clusters;
placing a neighborhood window at each pixel in order to determine what neighborhood pixels are used to compute a local intensity histogram for the pixel, wherein an initial neighborhood window size is set at the size of the entire input digital ultrasound image,
recomputing local mean values of each intensity class based on the current segmentation,
reclassifying the pixels according to the recomputed local mean values of intensity classes,
repeating the recomputing and reclassifying substeps until a convergence is reached, and
repeating the placing, recomputing, reclassifying and repeating substeps for one or more reduced neighborhood window sizes, until segmentation of the input digital ultrasound image is obtained when the minimum window size is processed;
selecting, from the plurality of regions, one or more candidate lesion regions having an intensity value lower than a pre-determined intensity value; and
eliminating from the selected candidate lesion regions any regions that touch image borders.