Systems and methods for automatic detection and quantification of pathology using dynamic feature classification
Methods, devices, and systems are provided for quantifying an extent of various pathology patterns in scanned subject images. The detection and quantification of pathology is performed automatically and unsupervised via a trained system. The methods, devices, and systems described herein generate unique dictionaries of elements based on actual image data scans to automatically identify pathology of new image data scans of subjects. The automatic detection and quantification system can detect a number of pathologies including a usual interstitial pneumonia pattern on computed tomography images, which is subject to high inter-observer variation, in the diagnosis of idiopathic pulmonary fibrosis.
1. An automatic detection and quantification system, comprising:
a server, comprising:
a processor; and
a computer readable medium, coupled to the processor and comprising instructions stored thereon that cause the processor to:
receive an input image data scan of a subject;
extract features from the input image data scan;
determine, using an elements dictionary and a classification model, one or more normal and/or abnormal pathology areas of the input image data scan based on the extracted features, wherein each pathology area comprises a cluster of pixels;
generate a marked image data scan by marking each of the determined one or more pathology areas on the input image data scan with a visual identifier;
compute one or more quantitative scores for the input image data scan based on a comparison of a number of regions classified as various categories of normal and/or abnormal with a number of regions tested; and
store the marked image data scan and the one or more quantitative scores in a memory.
2. The automatic detection and quantification system of claim 1 , wherein the elements dictionary is generated based on a plurality of sample image data scans received from a plurality of subjects.
3. The automatic detection and quantification system of claim 2 , wherein the plurality of sample image data scans comprises healthy image data scans and pathology image data scans.
4. The automatic detection and quantification system of claim 3 , wherein the classification model is generated in an unsupervised fashion using the elements dictionary and a clustering process.
5. The automatic detection and quantification system of claim 4 , wherein the classification model is generated using training examples comprising delineated image data scans outlining regions of interest having pathology.
6. The automatic detection and quantification system of claim 5 , further comprising:
an image scan system configured to provide cross-sectional image data scans of the subject, wherein the server is communicatively coupled to the image scan system.
7. The automatic detection and quantification system of claim 6 , wherein the marked image data scan is transmitted to a communication device across a communication network.
8. The automatic detection and quantification system of claim 7 , wherein the marked image data scan is configured to be rendered by a display device of the communication device.
9. The automatic detection and quantification system of claim 8 , wherein the visual identifier describes a pathology type and quantification value associated with the one or more pathology areas.
10. The automatic detection and quantification system of claim 1 , wherein multiple elements dictionaries are generated based on a plurality of sample image data scans received from a plurality of subjects.
11. The automatic detection and quantification system of claim 10 , wherein the multiple elements dictionaries comprise hierarchical dictionaries for features at multiple scales.
12. The automatic detection and quantification system of claim 10 , wherein the multiple elements dictionaries comprise separate dictionaries for normal features and abnormal features.
13. A computer program product, comprising:
a non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code comprising:
instructions to receive, by an image scanner, an input image data scan of a subject;
instructions to extract, by a processor, features from the input image data scan;
instructions to determine, by the processor, using an elements dictionary and a classification model, one or more normal and/or abnormal pathology areas of the input image data scan based on the extracted features, wherein each pathology area comprises a cluster of pixels;
instructions to generate, by the processor, a marked image data scan by marking each of the determined one or more pathology areas on the input image data scan with a visual identifier;
instructions to compute, by the processor, one or more quantitative scores for the input image data scan based on a comparison of a number of regions classified as various categories of normal and/or abnormal with a number of regions tested; and
instructions to store, by the processor, the marked image data scan and the one or more quantitative scores in a memory.
14. The computer program product of claim 13 , wherein the elements dictionary is generated based on a plurality of sample image data scans received from a plurality of subjects.
15. The computer program product of claim 13 , wherein the plurality of sample image data scans comprises healthy image data scans and pathology image data scans.
16. The computer program product of claim 15 , wherein the classification model is generated using training examples and the elements dictionary.
17. The computer program product of claim 16 , wherein the training examples comprise delineated image data scans outlining regions of interest having pathology.
18. The computer program product of claim 17 , further comprising:
an image scan system configured to provide cross-sectional image data scans of the subject, wherein the server is communicatively coupled to the image scan system.