Method and system for re-screening nodules in radiological images using multi-resolution processing, neural network, and image processing

• US 6,125,194 A
• Filed: 02/04/1998
• Issued: 09/26/2000
• Est. Priority Date: 02/06/1996
• Status: Expired due to Term

• Current Assignee: Cad Investments, LLC, Riverain Technologies LLC
• Original Assignee: Caelum Research Corporation
• Inventors: Lin, Jyh-Shyan, Lure, Yuan-Ming F, Yeh, Hwa-Young M
• Primary Examiner(s): Johns, Andrew W.
• Assistant Examiner(s): NAKHJAVAN, SHERVIN K
• Application Number:
  US09/018,789
• Time in Patent Office: 965 Days
• Field of Search: 382/132, 382/156, 382/173, 382/257, 382/128, 382/171, 382/294, 382/130, 382/224, 378/37, 128/922
• US Class Current: 382/132
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06T 2207/30061 Lung
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G16H 30/40 for processing medical images, e.g. editing
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

An automated detection method and system improve the diagnostic procedures of radiological images containing abnormalities, such as lung cancer nodules. The detection method and system use a multi-resolution approach to enable the efficient detection of nodules of different sizes, and to further enable the use of a single nodule phantom for correlation and matching in order to detect all or most nodule sizes. The detection method and system use spherical parameters to characterize the nodules, thus enabling a more accurate detection of non-conspicuous nodules. A robust pixel threshold generation technique is applied in order to increase the sensitivity of the system. In addition, the detection method and system increase the sensitivity of true nodule detection by analyzing only the negative cases, and by recommending further re-assessment only of cases determined by the detection method and system to be positive. The detection method and system use multiple classifiers including back propagation neural network, data fusion, decision based pruned neural network, and convolution neural network architecture to generate the classification score for the classification of lung nodules. Such multiple neural network architectures enable the learning of subtle characteristics of nodules to differentiate the nodules from the corresponding anatomic background. A final decision making then selects a portion of films with highly suspicious nodules for further reviewing.

Claims

1. A system for re-screening an abnormality such as a nodule in a radiological image, the system receiving radiological images identified as negative by previous radiological diagnostic techniques, the system comprising a detection unit comprising:

an image enhancement unit that uses a multi-resolution matched filtering approach to enhance contrast between any abnormalities that may be present and the image background;

a quick selection unit that preliminarily selects a suspect abnormality and uses a pixel thresholding method; and

a classification unit that determines a presence and a location of the abnormality and a classification score, and which identifies a false abnormality; and

a decision making unit that selects a portion of images for further diagnostic review;

wherein, if a radiological image analyzed by said detection unit is determined to be positive, further radiological diagnosis is performed on it to confirm the existence of a true nodule, and if a radiological image analyzed by said detection unit is determined to be negative, no further radiological diagnosis is performed on it.

2. A system according to claim 1, wherein said multi-resolution matched filtering approach generates low-resolution sub-images with different resolutions from an original image with a plurality of abnormalities of different sizes.

3. A system according to claim 1, wherein said multi-resolution matched filtering approach utilizes a single nodule phantom as a reference.

4. A system according to claim 3, wherein said nodule phantom is correlated with the image to determine likely locations of abnormalities.

5. A system according to claim 1, further including an image decimation unit that generates a low-resolution sub-image, said low-resolution sub-image having a smaller number of pixels than an original input digital image, by discarding redundant image pixels, and said sub-image covering the same area as the original image.

6. A system according to claim 1, wherein said image enhancement unit includes a low-pass filtering unit that smoothes the image.

7. A system according to claim 1, wherein said quick selection unit includes at least one of the following as means for determining threshold values for said pixel thresholding method:

a signal-to-noise ratio (SNR) evaluation unit, a cumulative distribution function (CDF) generation unit, a fraction estimation unit, an abnormality size update unit, and a CDF threshold determining unit.

8. A system according to claim 7, wherein said quick selection unit includes said fraction estimation unit, said fraction estimation unit estimating one or more suspect nodule area (SNA) fractions by evaluation the expression,

9. A system according to claim 7, wherein said quick selection unit includes said CDF threshold determining unit, said CDF threshold determining unit estimating one or more CDF thresholds by evaluating the expression,

10. A system according to claim 7, wherein said quick selection unit uses at least one of the following parameters of an abnormality:

a desired SNA size, a desired SNA amount, and SNR of an image.

11. A system according to claim 1, wherein said classification unit receives a suspect abnormality area and determines said classification score for that area, said classification unit comprising:

a feature extraction unit;

a feature pattern classifier,

an image area classifier; and

a data fusion unit, said data fusion unit integrating detection results from the different classifiers by weighing them.

12. A system according to claim 1, wherein said decision making unit receives said classification score to determine only portions of cases for further review.

13. A system according to claim 1, wherein said decision making unit uses prevalence rate of abnormality, risk factor of abnormality for a certain population, and performance of detection system, including number of true positives and false positives, to determine a classification threshold.

14. A system according to claim 1, wherein said decision making unit comprises:

a true occurrence evaluation unit that evaluates the occurrence of abnormality;

a fraction determination unit that determines a fraction of cases in which abnormality occurs;

a classification threshold determination unit; and

a reviewing thresholding unit that determines whether or not a given image is to undergo further diagnostic review, based on nodule classification score and on the output of the classification threshold determination unit.

15. A method for re-screening an abnormality such as a nodule in a radiological image comprising:

receiving at least one radiological image, said at least one radiological image having previously undergone diagnostic review;

identifying said at least one radiological image according the following rules;

if said radiological image was identified as positive during said previous diagnostic review, placing the image in a group for further radiological diagnosis to confirm the existence of a true nodule;

or

if said radiological image was determined to be negative during said previous diagnostic review, placing the image in a group for analysis according to a detection method, said detection method comprising the steps of;

performing an image enhancement step to enhance the contrast between any abnormalities present and image background, said image enhancement step comprising the sub-step of;

applying multi-resolution matched filtering;

performing a quick selection step to preliminarily select a suspect abnormality, said quick selection step comprising;

pixel thresholding;

performing a classification step to determine a presence and a location of an abnormality and a classification score, and to identify a false abnormality; and

performing a decision making step to select a portion of images for further diagnostic review, said decision making step including the following sub-steps;

if said radiological image analyzed by said detection unit is determined to be positive, indicating that further diagnostic review should be performed on it to confirm the existence of a true nodule;

or

if said radiological image analyzed by said detection unit is determined to be negative, indicating that no further radiological diagnosis should be performed on it.

16. A method according to claim 15, wherein said step of applying multi-resolution matched filtering includes the step of generating low-resolution sub-images with different resolutions from an original image with a plurality of abnormalities of different sizes.

17. A method according to claim 15, wherein said step of applying multi-resolution matched filtering utilizes a single nodule phantom as a reference.

18. A method according to claim 17, wherein said step of applying multi-resolution matched filtering comprises the step of correlating said nodule phantom with the image to determine likely locations of abnormalities.

19. A method according to claim 15, further including the step of decimating said image to generate a low-resolution sub-image, said low-resolution sub-image having a smaller number of pixels than an original input digital image, and said sub-image covering the same area as the original image, said step of decimating including the step of discarding redundant image pixels.

20. A method according to claim 15, wherein said step of performing image enhancement includes the step of low-pass filtering to smooth the image.

21. A method according to claim 15, wherein said pixel thresholding step includes at least one of the following steps for determining threshold values:

evaluating signal-to-noise ratio (SNR);

generating a cumulative distribution function (CDF);

estimating a suspect nodule area (SNA) fraction;

updating abnormality size; and

determining a CDF threshold.

22. A method according to claim 21, wherein said pixel thresholding step includes said step of estimating an SNA fraction, said SNA fraction estimation step estimating one or more suspect nodule area (SNA) fractions by evaluating the expression,

23. A method according to claim 21, wherein said pixel thresholding step includes said step of determining a CDF threshold, said CDF threshold determining step estimating one or more CDF thresholds by evaluating the expression,

24. A method according to claim 21, wherein said pixel threshold determination uses at least one of the following parameters of an abnormality:

a desired SNA size, a desired SNA amount, and SNR of an image.

25. A method according to claim 15, wherein said classification receives a suspect abnormality area and determines said classification score for that area, said classification step comprising the steps of:

performing feature extraction;

performing feature pattern classification;

performing image area classification; and

integrating detection results from the different classification steps by weighing them.

26. A method according to claim 15, wherein said decision making step receives and classification score and determines only portions of cases for further review.

27. A method according to claim 15, wherein said decision making step includes the step of using prevalence rate of abnormality, risk factor of abnormality for a certain population, and performance of detection system, including number of true positives and false positives, to determine a classification threshold.

28. A method according to claim 15, wherein said decision making step comprises the steps of:

evaluating the true occurrence of abnormality;

determining a fraction of cases in which abnormality occurs;

determining a classification threshold; and

applying said classification threshold to determine whether or not a given image is to undergo further diagnostic review, based on nodule classification score.