Combination machine learning algorithms for computer-aided detection, review and diagnosis

• US 8,296,247 B2
• Filed: 03/22/2008
• Issued: 10/23/2012
• Est. Priority Date: 03/23/2007
• Status: Active Grant

• Current Assignee: Three Palm Software
• Original Assignee: Three Palm Software
• Inventors: Zhang, Heidi Daoxian, Heffernan, Patrick Bernard
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): BUSS, BENJAMIN J
• Application Number:
  US12/053,600
• Publication Number:
  US 20090171871A1
• Time in Patent Office: 1,676 Days
• Field of Search: 706/10, 706/12, 706/20, 706/48, 706/924
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

A method of reviewing medical images and clinical data to generate a diagnosis or treatment decision is provided. The method includes receiving, at a computer-aided detection (CAD) system, the medical images and clinical data, processing the medical images and clinical data; to generate initial finding candidates and clustering the initial finding candidates into a plurality of groups. The method further includes classifying the initial finding candidates using machine learning algorithms integrated into the CAD system into one or more categories one or more categories of the initial finding candidates using type 2 fuzz logic, and determining detection and assessment statistics based on at least the assessed categories and classified findings using Bayesian probability analysis. The method also includes modifying the classified findings and assessed categories based on additional interactive input, and generating the diagnosis or treatment decision based on the determined detection, assessment statistics, and the additional interactive input.

Claims

1. A method of reviewing medical images and clinical data to generate a diagnosis or treatment decision, comprising:

receiving, at a computer-aided detection (CAD) system, the medical images and clinical data;

processing, by the CAD system, the medical images and clinical data;

to generate initial finding candidates;

clustering, by the CAD system, the initial finding candidates into a plurality of groups;

classifying, by the CAD system, the initial finding candidates into one or more categories using machine learning algorithms integrated into the CAD system and combined classifiers;

assessing, by the CAD system, one or more categories of the initial finding candidates using type 2 fuzzy logic;

determining, by the CAD system, detection and assessment statistics based on at least the assessed categories and classified findings using Bayesian probability analysis;

modifying, by the CAD system, the classified findings and assessed categories based on additional interactive input; and

generating the diagnosis or treatment decision based on the determined detection, assessment statistics, and the additional interactive input.

2. The method of claim 1, wherein processing the medical images and clinical data comprises:

segmenting a region of interest from a background;

normalizing the region of interest;

filtering the normalized region of interest; and

generating the initial finding candidates from the normalized region of interest.

3. The method of claim 1, wherein clustering the initial finding candidates into a plurality of groups comprises:

determining a number of clusters stored in a database;

measuring a distance between the clusters based on modality;

clustering the initial finding candidates using a hierarchical or partitional algorithm; and

saving a centroid from each initial finding candidate cluster.

4. The method of claim 3, wherein clustering the initial finding candidates into a plurality of groups further comprises:

loading saved centroids;

calculating distances between at least one candidate of the initial finding candidates and each loaded centroid; and

classifying the at least one candidate of the initial finding candidates based on the calculated distance.

5. The method of claim 1, wherein classifying the initial finding candidates comprises:

extracting selected features from the initial finding candidates;

performing, by the CAD system, a first classification using a first neural network trained by a first algorithm;

performing, by the CAD system, a second classification using a kernel-based method to map feature vectors of the extracted features into higher-dimensional space;

performing, by the CAD system, a third classification using a second neural network trained by a second algorithm;

combining the first, second, and third classifications; and

determining, by the CAD system, a minimum for each classification, a mean for each classification, and a maximum for each classification.

6. The method of claim 1, wherein assessing one or more categories of the initial finding candidates using type 2 fuzzy logic comprises:

inputting linguistic variables into a fuzzifier to generate fuzzified results;

inputting the fuzzified results and additional rules into an interface;

reducing the fuzzified results by type;

defuzzifying the reduced results; and

translating the defuzzified results into formal language.

7. The method of claim 1, wherein determining detection and assessment statistics based on at least the assessed categories and classified findings using Bayesian probability analysis comprises:

estimating parameters of a probability density function in a form of Rayleigh distribution using an expectation-maximization algorithm;

obtain prior probabilities; and

calculate posteriori probabilities of the assessed categories and classified results using the estimated parameters and prior probabilities.

8. The method of claim 1, wherein the additional interactive input is provided by a human observer.