Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/0002—Inspection of images, e.g. flaw detection
  • G06T7/0012—Biomedical image inspection
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
  • G06V30/10—Character recognition
  • G06V30/24—Character recognition characterised by the processing or recognition method
  • G06V30/242—Division of the character sequences into groups prior to recognition; Selection of dictionaries
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/10—Image acquisition modality
  • G06T2207/10072—Tomographic images
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20004—Adaptive image processing
  • G06T2207/20012—Locally adaptive
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/30—Subject of image; Context of image processing
  • G06T2207/30004—Biomedical image processing
  • G06T2207/30061—Lung
  • G06T2207/30064—Lung nodule

Definitions

• the present invention relates to the field of machine classification, and, more particularly, to incorporating spatial knowledge for classification.
• a classifier is used to classify (i.e., separates) objects into two or more classes.
• An example of a classifier is as follows. Assume we have a set, A, of objects comprising two groups (i.e., classes) of the objects that we will call A+ and A-.
• object refers to one or more elements in a population.
• the classifier, A is a function, F, that takes every element in A and returns a label "+ "or "-", depending on what group the element is. That is, the classifier may be a FUNCTION F(A) -- ⁇ -1, 1 ⁇ , where -1 is a numerical value representing A- and +1 is a numerical value representing A+.
• the classifiers A+ and A- may represent two separate populations.
• A+ may represent structures in the lung (e.g., vessels, bronchi) and A- may represent nodules.
• F training data
• classifications of new and unseen data can be predicted using the function, F.
• a classifier can be trained in 10,000 known objects for which we have readings from doctors. This is commonly referred to as a "ground truth.” Based on the training from the ground truth, the classifier can be used to automatically diagnose new and unseen cases.
• a conventional classifier classifies the objects into classes based on an assumption that objects of the same class have comparable feature values, that is, belong to the same distribution in the feature space.
• feature refers to one or more attributes that describe an object belonging to a particular class.
• a nodule can be described by a vector containing a number of attributes, such as size, diameter, sphericity, etc.
• the vector may contain attribute values, which are termed herein as "feature values.”
• a function K (i.e., the kernel) takes the original data and maps it to a higher dimensional feature space (i.e., a feature space with more features) where the task of finding a classification function is easier to achieve.
• a higher dimensional feature space i.e., a feature space with more features
• the task of finding a classification function is easier to achieve.
• the original data is a single case that analyzes a certain number of features.
• a vector for the original data is a 10 dimensional feature space.
• a 1,000 dimensional feature space is created, mapped from the 10 dimensional feature space.
• the term “degrees of freedom” refers to the number of values in the final calculation of a statistic that are free to vary. Another way to produce the more complex classifier is to combine a number of simple classifiers, each trained using different sets of features, in series or in parallel. The results are merged to form an ensemble of classifiers. These and other comparable approaches have the disadvantage of requiring additional training examples to compensate for the large number of degrees of freedom needed to obtain the same generalization performance.
• generalization performance refers to the performance of a classifier on new and unseen data. A reduction of generalization performance may occur for many reasons.
• a method of classifying candidates in an image comprises classifying a first candidate of the image using a first classifier, wherein the first classifier is trained using training data specific to the spatial location of the first candidate; and classifying a second candidate of the image using a second classifier, wherein the second classifier is trained using training data specific to the spatial location of the second candidate; wherein the spatial location of the first candidate is mutually exclusive from the spatial location of the second candidate.
• a method of automatically detecting a nodule on an image comprises obtaining an image; identifying a first spatial location and a second spatial location on the image, wherein the first spatial location is mutually exclusive from the second spatial location; identifying a first candidate in the first spatial location and a second candidate in the second spatial location; classifying whether the first candidate is a nodule using a first classifier, wherein the first classifier is trained using training data specific to the first spatial location; and classifying whether the second candidate is a nodule using a second classifier, wherein the second classifier is trained using training data specific to the second spatial location; wherein the spatial location of the first candidate is mutually exclusive from the spatial location of the second candidate.
• a machine-readable medium having instructions stored thereon for execution by a processor to perform method of classifying candidates in an image comprises classifying a first candidate of the image using a first classifier, wherein the first classifier is trained using training data specific to the spatial location of the first candidate; and classifying a second candidate of the image using a second classifier, wherein the second classifier is trained using training data specific to the spatial location of the second candidate; wherein the spatial location of the first candidate is mutually exclusive from the spatial location of the second candidate.
• a machine- readable medium having instructions stored thereon for execution by a processor to perform method of automatically detecting a nodule on an image is provided.
• the method comprises obtaining an image; identifying a first spatial location and a second spatial location on the image, wherein the first spatial location is mutually exclusive from the second spatial location; identifying a first candidate in the first spatial location and a second candidate in the second spatial location; classifying whether the first candidate is a nodule using a first classifier, wherein the first classifier is trained using training data specific to the first spatial location; and classifying whether the second candidate is a nodule using a second classifier, wherein the second classifier is trained using training data specific to the second spatial location; wherein the spatial location of the first candidate is mutually exclusive from the spatial location of the second candidate.
• Figure 1 depicts a method of classifying objects based on the spatial locations of the objects is illustrated, in accordance with one embodiment of the present invention
• Figures 2A and 2B depict exemplary CT lung scans after classification of nodules, in accordance with one embodiment of the present invention.
• DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS Illustrative embodiments of the invention are described below. In the interest of clarity, not all features of an actual implementation are described in this specification.
• At least a portion of the present invention is preferably implemented as an application comprising program instructions that are tangibly embodied on one or more program storage devices (e.g., hard disk, magnetic floppy disk, RAM, ROM, CD ROM, etc.) and executable by any device or machine comprising suitable architecture, such as a general purpose digital computer having a processor, memory, and input/output interfaces.
• program storage devices e.g., hard disk, magnetic floppy disk, RAM, ROM, CD ROM, etc.
• suitable architecture such as a general purpose digital computer having a processor, memory, and input/output interfaces.
• kernel-based classifier refers to a classifier where a mapping function (i.e., the kernel) has been used to map the original training data to a higher dimensional space where the classification task may be easier.
• a mapping function i.e., the kernel
• the method 100 obtains (at 105) an image.
• the image may be obtained from a physical scan of the human body, such as from a computed tomography ("CT"), magnetic resonance imaging (“MRI”) or ultrasound device.
• CT computed tomography
• MRI magnetic resonance imaging
• ultrasound device In another embodiment, the image may be obtained from a satellite imaging system.
• the method 100 identifies (at 110) spatial locations for elements on the image.
• potential locations may include a wall-attached location representing elements attached to the lung wall, a bronchi classifier representing elements attached to the bronchi, and an inside-lung classifier representing elements inside the lung.
• Other images of the human body can be used, such as the heart.
• a satellite image may be classified into a type of landmass (e.g., mountain, desert, rainforest).
• the spatial locations may be identified through, for example, a graphical user-interface.
• the spatial locations may also be identified automatically using, for example, previous knowledge of various locations on an image. It should be appreciated that various other automated methods (e.g., pattern matching) for identifying spatial locations may be used, as contemplated by those skilled in the art.
• Classifiers are created (at 115) for each of the spatial locations. Using the lung scan example above, potential classifiers may include a wall-attached classifier representing elements attached to the lung wall, a bronchi classifier representing elements attached to the bronchi, and an inside-lung classifier representing elements inside the lung. The classifiers may be trained using previously-known data. For example, using the lung scan example above, the classifiers may be trained using training data of elements specified for the particular spatial location. Training data may comprise known and previously- verified data.
• doctor-verified cases determining whether a nodule exists can be used as training data.
• the various cases may be separated into spatial locations corresponding to the different classifiers for training purposes.
• a wall-attached classifier would be trained using only training data of elements attached to the lung wall.
• the advantage of such a method is that the classifier takes advantage of any novelties (known and unknown) of elements in a particular location.
• classifying elements in the bronchi may be different from classifying elements inside the lung.
• the method 100 identifies (at 120) candidates.
• the term "candidates" refers to elements (i.e:, structures) of interest in the image.
• the classifiers are unique to the spatial locations, the classification of candidates will be more precise than not taking into account the spatial location. Further, using such "local classifiers" also result in memory, storage and performance improvements over complex, universal classifiers. Candidates may be identified manually or automatically using any of a variety of methods and apparatus known to those skilled in the art. For example, computer-aided automatic identification of suspicious regions of interest is well-known in the field of cancer screening.
• the method 100 classifies (at 125) the candidates into labels using the appropriate classifier associated with the spatial location where the candidate is located. For example, in the lung scan example above, if a candidate is located inside the lung, then the candidate is classified using the inside-lung classifier.
• the candidate may be labeled "nodule" representing that a structure (i.e., the element) in the image is a nodule and "no-nodule” representing that the structure is not a nodule.
• a candidate may be classified with more than two labels.
• particular labels can be physically marked on the image for easy identification. For example, in the lung scan example above, a candidate identified as a nodule may be marked so that a human observer (e.g., a doctor) can easily identify the pathology. Different marking types and shapes may be used to differentiate between certain classifiers, as is contemplated by those skilled in the art.
• FIG. 2A and 2B show exemplary CT lung scans that are marked after classification, in accordance with one embodiment of the present invention.
• the square markings refer to wall-attached nodules
• the circle markings refer to inside-lung nodules.

Landscapes

• Engineering & Computer Science (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Theoretical Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Physics & Mathematics (AREA)
• Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
• Quality & Reliability (AREA)
• Radiology & Medical Imaging (AREA)
• Health & Medical Sciences (AREA)
• Medical Informatics (AREA)
• General Health & Medical Sciences (AREA)
• Multimedia (AREA)
• Image Analysis (AREA)
• Apparatus For Radiation Diagnosis (AREA)
• Measuring And Recording Apparatus For Diagnosis (AREA)
• Magnetic Resonance Imaging Apparatus (AREA)
• Ultra Sonic Daignosis Equipment (AREA)
• Image Processing (AREA)
• Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Priority Applications (2)

Applications Claiming Priority (4)

Publications (2)

Family

ID=34198001

Family Applications (1)

Country Status (4)

Families Citing this family (5)

Family Cites Families (24)

• 2004
  • 2004-08-10 US US10/915,076 patent/US7634120B2/en active Active
  • 2004-08-13 WO PCT/US2004/026425 patent/WO2005017815A2/en not_active Ceased
  • 2004-08-13 JP JP2006523424A patent/JP5105876B2/ja not_active Expired - Fee Related
  • 2004-08-13 DE DE112004001468.9T patent/DE112004001468B4/de not_active Expired - Fee Related

Also Published As

Similar Documents

Legal Events