REDUCING FALSE POSITIVES FOR AUTOMATIC COMPUTERIZED DETECTION OF OBJECTS

This application claims the benefit of Provisional Application No. 60/716,843 filed on 9/14/2005 in the United States Patent and Trademark Office, the contents of which are herein incorporated by reference in its entirety.

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present disclosure relates to image processing, and more particularly to a system and method for detecting an object in an image while eliminating false positive detections.

2. Description of Related Art

The pulmonary embolism (PE) is defined as a thrombus, which is generally recognized as dark regions within enhanced pulmonary arteries in computed tomography angiography (CTA) images. For PE detection, it is difficult to distinguish PE from various PE look-alikes (false positives) including flow voids in veins and arteries, lymphoid tissues, streak artifacts near superior vena cava, partial volume artifacts at vascular bifurcations, etc.

Therefore, a need exists for a system and method for object detection including a feature based approach for automatically removing the false positives, while preserving the true positives.

SUMMARY OF TB[E INVENTION

According to an embodiment of the present disclosure, a computer-implemented method for identifying an object of interest includes providing input data including an image and a candidate for the object of interest in the image, extracting a boundary of the candidate, and extracting a segment of a region of interest containing the candidate. The method further includes determining a plurality of features of an extracted segment of the region of interest containing the candidate, and outputting the object of interest, wherein the object of interest is characterized by the plurality of features, wherein the object of interest and the plurality of features are stored as computer-readable code.

The region of interest is a vessel, and the object of interest is a pulmonary embolism.

Extracting the segment of the region of interest includes seeding the segment extraction using the boundary and growing the boundary.

Extracting the segment of the region of interest comprises providing the region of interest having a fixed size.

Providing the input data includes providing the candidate as a plurality of voxels, providing a value for stopping a region growing as a maximum distance between a voxel and the boundary of the candidate, and providing a threshold for an intensity of the voxel.

According to an embodiment of the present disclosure, a computer-implemented method for identifying an object of interest includes providing input data including an image and a candidate for the object of interest in the image, extracting a boundary of the candidate according to a set of neighbors of each voxel of the candidate, extracting a segment of a region of interest containing the candidate based on the boundary by determining a minimum cumulative cost path map of the boundary, determining a plurality of features of an extracted segment of the region of interest containing the candidate, and outputting the object of interest, wherein the object of interest is characterized by the plurality of features, wherein the object of interest and the plurality of features are stored as computer-readable code.

Outputting the object of interest further includes comparing the plurality of features to a known set of features, and verifying the object of interest based on the comparison. Outputting the object of interest further includes classifying the candidate as a false positive or a true positive based on the plurality of features.

According to an embodiment of the present disclosure, a program storage device is provided readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for identifying an object of interest. The method steps includes providing input data including an image and a candidate for the object of interest in the image, extracting a boundary of the candidate, and extracting a segment of a region of interest containing the candidate. The method further includes determining a plurality of features of an extracted segment of the

region of interest containing the candidate, and outputting the object of interest, wherein the object of interest is characterized by the plurality of features,

wherein the object of interest and the plurality of features are stored as computer- readable code.

BRIEF DESCRIPTION OF THE DRAWINGS

Preferred embodiments of the present invention will be described below in more detail, with reference to the accompanying drawings:

Figures IA-C illustrate a difference between a vessel containing a PE and a vessel containing a PE false positive (FP) according to an embodiment of the present disclosure;

Figure 2 is a flow chart of a method for extracting an object of interest according to an embodiment of the present disclosure;

Figure 3 is a flow chart of a method for extracting an object of interest according to an embodiment of the present disclosure;

Figure 4 is an illustration of a method for extracting an object of interest according to an embodiment of the present disclosure;

Figure 5 is a drawing of an extracted object of interest according to an embodiment of the present disclosure; and

Figure 6 is a drawing of a system according to an embodiment of the present disclosure.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

According to an embodiment of the present disclosure, a feature-based system and method substantially eliminates false positives in automatic Pulmonary Embolism detection, including vessel segment extraction and feature calculation.

Figures IA-C show CT images for the PE detection. Figure IA shows a true PE in an artery, Figure IB shows an PE FP in a vein, and Figure 1C shows a PE FP in an artery. The texture and intensity of the vessel surrounding the true PE (Figure IA) are different from those of the vessel surrounding the FPs. A system and method according to an embodiment of the present invention exploits vessel characteristics to differentiate

the true PEs from the false positives. Referring to Figure 2, the approach is a vessel- based approach, wherein, given one PE candidate 201, a segment of vessel containing the candidate is extracted 202. Given the extracted candidate, features based on the extracted vessel segment are extracted 203, so that these extracted

features may be utilized by other modules, for example, a classifier, to identify the candidate as a true PE or a non-PE object.

Referring to Figure 3 and the vessel extraction 203; each potential PE is given by a set of voxels 301, also referred to as candidate, component or candidate component. Block 301 includes two parts: candidate boundary extraction 302 and vessel segment extraction 303. The boundary points of the PE candidate component are set as seed points in the vessel segmentation 303. There are a various approaches that may be used for extracting the boundary, for example, edge detection, derivatives, or dilate the PE candidate and then subtract the PE candidate from the dilated PE candidate. In detail, for example, the boundary is obtained by checking the neighbors of each voxel of the PE candidate component. If all of its neighbors belong to the component, then the voxel is not on the boundary. If any of its neighbors does not belong to the component, then the voxel is on the boundary and is recorded.

After boundary extraction 302, a neighboring segment of the vessel is extracted that contains the candidate component. There are various approaches that may be applied. For example, vessel segment extraction 303, which is based on using a graph- searching method to create a minimum cumulative cost path map. Here the cost is the Euclidean distance from the boundary of the candidate component.

The graph search vessel extraction 302 method may be written as follows: Input (see Figure 2, block 201): c [Voxels in one PE candidate component]

MaxDistance [Value for stopping the region growing (the maximum distance between a vessel voxel and the boundary of the candidate component] Threshlntensity [Threshold for the intensity of the vessel]

Data Structures:

s [Seed voxels (boundary points of the candidate component)] (see Figure 3, block 303)

Q [Priority queue of active voxels sorted by total cost]

N(q) [Neighborhood set of voxel q (contains 26 neighbors of the voxel)]

C(q) [Total cost function from seed points to q]

E{q) [Boolean function indicating if q has been processed]

Outputs (see Figure 2, block 204): p [Pointers from each voxel indicating the minimum cost path] t [List of voxels on extracted vessel segment]

Exemplary Algorithm:

5 <— boundary(c) [Get the boundary voxels of the component (seed voxels)]

E(c \ s) <r- TRUE [Mark all the voxels inside the component as processed

(do not process the voxels inside the component)]

C(s) <— 0,2 <— s [Initialize the priority queue with zero cost seed voxels] while Q ≠ φ do [While the queue is not empty:] q <— min(0 [Remove minimum cost voxel from the queue] if C(q) ≥MaxDistance break; [Stop growing if it reaches the maximum cost (maximum length of vessel)] end if q <£ s [If q is not on the boundary of component] t <— q; [Add q to the list of vessel points] end

E(Q) <- TRUE [Mark q as processed] for each voxel r £ N(q) if intensity (r) ≥Threshlntensity [Assume the intensity of vessel is not less than the threshold]

c current <- C(q) + d(q) [Calculate the current total cost to neighbor] if r e Q and Ccurrent < C(r) r <r- Q [Remove higher cost neighbors from queue] end if r * Q, [If neighbor is not in the queue]

C(r) ±- C _cιιrmιt [Assign neighbor's total cost] p(r) <- q [Set back pointer]

[Put neighbor into the queue] end end end

In experiments, based on the property of CT images, it has been assumed that all the vessels are brighter than a threshold (Threshlntensity) and other areas in the lung are darker than the threshold. As illustrated in Figure 4, the extracted region 401 is growing in layers 402 (equal distance to the boundary of the candidate component) and stops if the vessel length is larger than a given value (MaxDistance). Figure 5 shows a binary image in which the extracted vessel is shown in white. The black region inside the vessel is the PE component.

According to an embodiment of the present disclosure, in the vessel extraction process, a segment of the vessel is extracted alone with corresponding features and for example, stored as computer readable code, output to a display device, etc. However, the vessel segment can be as large as the whole vascular structure. The method for extracting the vessel presented here should be an illustrative embodiment. Other methods, for example, region growing, level set method among many others, can be used for extracting the vessel.

Referring to feature extraction 203, after vessel segmentation 202, texture features, histogram features, intensity features, difference features, and curvature features are extracted.

The texture features include texture features based on calculating the co-

occurrence matrix and texture features based on the discrete wavelet transform.

Histogram features include those having a most frequent gray value of the vessel.

The intensity features include the mean, median, standard deviation, minimum, maximum, skewness, and the curtosis of the gray values of the vessel segment.

Difference features include those having the difference between the mean value of the vessel and that of the candidate component, the difference between the most frequent gray value of the vessel and that of the candidate component.

Curvature features include the Gaussian curvature, mean curvature, the shape index, and the curvedness of the shell of the PE candidate

While the texture features, histogram features, intensity features, difference features, and curvature features are representative, any number of different features can be extracted from the ROI indicated by the candidate and the extracted vessel. The extracted features can be implemented to verify the object of interest based on a comparison of the plurality of features to a known or expected set of features of one or more objects of interest (see Figure 2, 205). The comparison may be used to eliminate FP candidates for objects of interest. These extracted features may be utilized by other modules, for example, a classifier, to identify the candidate as a true PE or a non-PE object, so as to eliminate the false positives and preserve the true positives.

It is to be understood that the present invention may be implemented in various forms of hardware, software, firmware, special purpose processors, or a combination thereof. In one embodiment, the present invention may be implemented in software as an application program tangibly embodied on a program storage device. The application program may be uploaded to, and executed by, a machine comprising any suitable architecture.

Referring to Figure 6, according to an embodiment of the present disclosure, a computer system 601 for object detection including a feature based approach for automatically removing the false positives can comprise, inter alia, a central processing unit (CPU) 602, a memory 603 and an input/output (JJO) interface 604. The computer system 601 is generally coupled through the I7O interface 604 to a display 605 and various input devices 606 such as a mouse and keyboard. The support circuits can include circuits such as cache, power supplies, clock circuits, and a communications

bus. The memory 603 can include random access memory (RAM), read only memory (ROM), disk drive, tape drive, etc., or a combination thereof. The present invention can be implemented as a routine 1007 that is stored in memory 603 and executed by the CPU 602 to process the signal from the signal source 608. As such, the computer system 601 is a general-purpose computer system that becomes a specific purpose computer system when executing the routine 607 of the present invention.

The computer platform 601 also includes an operating system and microinstruction code. The various processes and functions described herein may either be part of the microinstruction code or part of the application program (or a combination thereof), which is executed via the operating system. In addition, various other peripheral devices may be connected to the computer platform such as an additional data storage device and a printing device.

It is to be further understood that, because some of the constituent system components and method steps depicted in the accompanying figures may be implemented in software, the actual connections between the system components (or the process steps) may differ depending upon the manner in which the present invention is programmed. Given the teachings of the present disclosure provided herein, one of ordinary skill in the related art will be able to contemplate these and similar implementations or configurations.

Having described embodiments for a system and method for object detection including a feature based approach for automatically removing the false positives, it is noted that modifications and variations can be made by persons skilled in the art in light of the above teachings. It is therefore to be understood that changes may be made in embodiments of the present disclosure that are within the scope and spirit thereof.