Computer-assisted method and apparatus for the detection of lung nodules

• US 5,987,094 A
• Filed: 10/30/1997
• Issued: 11/16/1999
• Est. Priority Date: 10/30/1996
• Status: Expired due to Term

• Current Assignee: University of South Florida (State University System of Florida)
• Original Assignee: University of South Florida (State University System of Florida)
• Inventors: Qian, Wei, Mao, Fei, Clarke, Laurence P.
• Primary Examiner(s): Bruce, David Vernon
• Application Number:
  US08/960,815
• Time in Patent Office: 747 Days
• Field of Search: 378/62, 378/98, 382/132, 382/128
• US Class Current: 378/62
• CPC Class Codes: G01N 2223/419 computed tomograph
• CPC Class Codes: G01N 2223/612 biological material
• CPC Class Codes: G01N 23/046 using tomography, e.g. computed tomography [CT]
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

A computer-assisted diagnostic (CAD) method and apparatus are described for the enhancement, detection, and classification of suspicious regions in digital x-ray images, with particular emphasis on lung nodule detection using chest x-ray images. An objective is to improve the sensitivity and specificity of detection of suspicious areas such as nodules, while maintaining a low false positive detection rate. A modular CAD technique has been developed to be potentially automatic and to be used as a second-opinion method for lung nodule detection. The method consists of using a plurality of CAD modules to preprocess and enhance image features, including image preprocessing, selective enhancement, segmentation, and feature extraction using a multiresolution/multiorientation wavelet transform and a computationally efficient filter, a 1.5 D circular pattern filter.

Claims

1. A computer-assisted method for detecting a lung nodule comprising the steps of:

suppressing noise in a digital chest image;

enhancing and segmenting anatomical structures in the noise-suppressed image;

enhancing a potentially suspicious area in the noise-suppressed image; and

further enhancing and segmenting the potentially suspicious area with the use of a filter having sensitivity to a shape feature of a lung nodule and having relative insensitivity to an overlapping object.

2. The method recited in claim 1, wherein the suspicious area further enhancing and segmenting step comprises applying a 1.5-dimensional filter having sensitivity to a circular pattern.

3. The method recited in claim 1, wherein the suspicious area segmenting step further comprises performing adaptive clustering to segment out a boundary of the suspicious area.

4. The lung-nodule detection method recited in claim 1, wherein the noise-suppressing step comprises adaptively filtering noise.

5. The lung-nodule detection method recited in claim 4, wherein the adaptive noise-filtering step comprises applying an adaptive nonlinear multistage nonlinear filter.

6. The lung-nodule detection method recited in claim 5, wherein the adaptive nonlinear multistage nonlinear filter is adapted to adaptively select a filter on a pixel-by-pixel basis.

7. The lung-nodule detection method recited in claim 6, wherein the filter-selecting step comprises selecting among a plurality of filters comprising a linear operator and an order statistic operator.

8. The lung-nodule detection method recited in claim 1, wherein the anatomical structure enhancing step comprises applying a wavelet transform to the noise-suppressed image.

9. The lung-nodule detection method recited in claim 8, wherein the wavelet transform applying step comprises applying one or more of a tree-structured wavelet transform and a directional wavelet transform.

10. The lung-nodule detection method recited in claim 1, wherein the anatomical structure enhancing step comprises applying a wavelet transform to the enhanced image.

11. The lung-nodule detection method recited in claim 10, wherein the wavelet transform applying step comprises applying a tree-structured wavelet transform and the anatomical structures segmenting step comprises applying adaptive clustering.

12. The lung-nodule detection method recited in claim 1, wherein the suspicious area enhancing step comprises subtracting enhanced anatomical structures from the noise-suppressed image.

13. The lung-nodule detection method recited in claim 12, wherein the suspicious area enhancing step further comprises applying a wavelet transform to the noise-suppressed image.

14. The lung-nodule detection method recited in claim 13, wherein the wavelet transform applying step comprises applying a wavelet-based multiresolution/multiorientation transform comprising a bank of wavelet filters implemented on two-stage adaptive combiners.

15. The lung-nodule detection method recited in claim 1, further comprising the steps, following the enhancing and segmenting steps, of:

performing feature extraction on the segmented anatomical structures and potentially suspicious areas;

differentiating a suspicious area from the potentially suspicious areas based upon the feature-extraction step; and

classifying the suspicious area as a true positive or a false positive.

16. The lung-nodule detection method recited in claim 15, wherein the feature extraction step comprises extracting a region containing a potentially suspicious area and performing a feature calculation in a grey-level domain, a direction texture domain, a morphological domain, and a connectivity domain.

17. The lung-nodule detection method recited in claim 15, wherein the feature extraction step further comprises calculating positional and connectivity measurements.

18. The lung-nodule detection method recited in claim 17, wherein the suspicious area selecting and classifying steps comprise applying a neural network to the extracted region.

19. The lung-nodule detection method recited in claim 17, wherein the neural network applying step comprises applying a mixed-feature neural network using backpropagation with Kalman filtering.

20. A system for detecting a lung nodule contained in a digital chest image comprising:

means for suppressing noise in a digital chest image;

means for enhancing and segmenting anatomical structures in the noise-suppressed image;

means for enhancing potentially suspicious areas in the noise-suppressed image; and

filter means for segmenting the potentially suspicious area, the filter means having sensitivity to a shape feature of a lung nodule and having relative insensitivity to an overlapping object.

21. The system recited in claim 20, further comprising:

means for performing feature extraction on the raw and segmented anatomical structures and potentially suspicious areas;

means for classifying a suspicious area as a true positive or a false positive from the potentially suspicious areas based upon input of computed features from the feature-extraction step.

22. A system for detecting a lung nodule contained in a digital chest image comprising:

a filter for suppressing noise in a digital chest image;

a wavelet transform adapted to enhance anatomical structures in the noise-suppressed image;

a subtraction algorithm for removing the enhanced anatomical structures from the noise-suppressed image;

a wavelet transform adapted to enhance a potentially suspicious areas in the noise-suppressed image; and

a 1.5-dimensional circular pattern filter for additionally enhancing a suspicious area from the enhanced, anatomic-structure-subtracted image.

23. The system recited in claim 22, further comprising:

means for performing a feature extraction in a grey-level domain, a direction texture domain, and a morphological domain, for performing feature extraction on the segmented anatomical structures and potentially suspicious areas; and

a neural network for classifying the suspicious area.

24. A system for detecting a lung nodule contained in a digital chest image comprising:

means for accessing digital chest image data;

a processor in electronic communication with the data accessing means, the processor having resident therein a plurality of computer-assisted diagnostic modules for processing the digital chest image data comprising;

a filter for suppressing noise in a digital chest image;

a wavelet transform adapted to enhance anatomical structures in the noise-suppressed image;

a wavelet transform adapted to enhance a potentially suspicious area in the noise-suppressed image; and

a circular pattern filter for selectively further enhancing the suspicious area.

25. The system recited in claim 24, wherein the modules further comprise software means for performing a feature calculation in a grey-level domain, a direction texture domain, a morphological domain, and a location/connectivity domain, for performing feature extraction on the enhanced and segmented anatomical structures and potentially suspicious area;

and

a neural network for classifying the suspicious area as a potential lung nodule or a false positive based upon input of computed features from the feature-extraction module.

26. A modular software system for processing digital chest image data to detect a lung nodule, the system comprising:

a filter module for suppressing noise in digital chest image data;

a wavelet transform module adapted to enhance anatomical structures in the noise-suppressed image;

a wavelet transform module adapted to enhance a potentially suspicious areas in the noise-suppressed image; and

a 1.5-dimensional circular pattern filter module for further enhancing the enhanced suspicious area.

27. The system recited in claim 26, further comprising:

a feature calculation module for performing feature extraction on the raw image and the segmented anatomical structures and potentially suspicious areas in a grey-level domain, a direction texture domain, a morphological domain, and a connectivity domain; and

a neural network module for selecting a suspicious area from the potentially suspicious areas based upon the feature-extraction step and for classifying the suspicious area.

28. A data file containing a selected and classified suspicious area in a chest image, the data file having been created with the use of the software system of claim 26.

29. A method for performing image processing to detect a lung nodule in a digital chest image at a first site, the image having been collected at a second site remote from the first site:

electronically transmitting the image from the second site to the first site;

preprocessing the image to segment out a potentially suspicious area by;

suppressing noise in the image;

enhancing and segmenting anatomical structures in the noise-suppressed image;

enhancing a potentially suspicious area in the noise-suppressed image; and

further enhancing the potentially suspicious area with the use of a filter having sensitivity to a shape feature of a lung nodule and having relative insensitivity to an overlapping object;

segmenting the further enhanced potentially suspicious area with the use of adaptive clustering; and

electronically transmitting the processed image from the first site to the second site.

30. A method for performing image processing to determine for a presence of a lung nodule in a digital chest image at a first site, the image having been collected at a second site remote from the first site:

electronically transmitting the image from the second site to the first site;

processing the image to determine whether a potentially suspicious area is present by segmenting out a potentially suspicious area by;

suppressing noise in the image;

enhancing and segmenting anatomical structures in the noise-suppressed image;

determining whether a potentially suspicious area is present;

if a potentially suspicious area is present, performing an image processing method on the area comprising the steps of;

enhancing the area in the noise-suppressed image; and

further enhancing the potentially suspicious area with the use of a filter having sensitivity to a shape feature of a lung nodule and having relative insensitivity to an overlapping object; and

segmenting the further enhanced potentially suspicious area with the use of adaptive clustering; and

if no potentially suspicious area is present, determining that no lung nodules are present; and

electronically transmitting the processed image and a determination as to the presence or absence of a lung nodule from the first site to the second site.