Patent Application
20080137921
The present invention relates to the field of digital imaging, and more particularly to analyzing a lesion in a digital image.
Image segmentation is a branch of digital image processing that performs the task of categorizing, or classifying, the picture elements of a digital image into one or more class types. For medical imaging applications, it is common that image segmentation is performed on the voxel (volume element) of a 3-dimensional image data set with the classification types related to anatomical structure. In thoracic medical images, it is convenient to segment the image voxels into classes such as bone, lung parenchyma, soft tissue, bronchial vessels, blood vessels, etc. There are many reasons to perform such a task, such as surgical planning, treatment progress, and patient diagnosis.
Of particular interest is the image segmentation approach generally known as region growing. Starting with a seed point, i.e., a voxel position that is known to be part of a particular class type, a region of contiguous voxels is grown, or developed, about the seed point. The region growing process progresses until a stopping condition is satisfied, e.g., no more contiguous voxels are found, or a predetermined number of voxels have been visited, etc.
The research paper “Advanced navigation tools for virtual bronchoscopy” Image Processing: Algorithms and Systems III Proceedings of the SPIE, Volume 5298, pp. 147-158 (2004), by Perchet et al., describes a region growing image segmentation for tracking the anatomical airway structures in thoracic CT (computed tomography) medical images. As part of the segmentation processing, the bronchi structures are modeled by a graph, or tree based representation. The region growing method described by Perchet et al., includes a branching decision process when the algorithm encounters a bifurcation of the bronchi structure into multiple smaller bronchi structures. The original bronchi structure, or parent, splits into two or more child bronchi structures. The analysis to determine the splitting nature of the bronchi anatomy is performed on the “front propagation,” i.e., an advancing surface corresponding to the newly segmented voxels. Thus, during region growing process the parent front propagation surface propagates into two or more child surface fronts.
Pulmonary lesions typically grow within the thoracic pleural cavity. Often cancerous regions, or lesions, in the lung can be identified in CT volume images. However, segmenting the voxels in a thoracic CT image is a difficult task since the voxel values associated with the abnormal lesion tissue are generally in the same numerical range, i.e., Hounsfield units, as many normal anatomical structures, e.g. muscle, heart, and vascular tissue. A shortcoming of many approaches for segmenting pulmonary lesions is the difficulty of differentiating between the lesion and other normal anatomy structures. Often pulmonary lesions in CT images are segmented by a combination of voxel value thresholding and morphological filtering and operations as in the paper “Three-dimensional segmentation and growth-rate estimation of small pulmonary nodules in helical CT images” by Kostis et al., IEEE Trans Med Imaging. 2003 October; 22(10):1259-74. The morphological filtering approach for segmenting pulmonary lesions described by Kostis et al. has difficulty distinguishing between the abnormal lesion tissue and the tissue associated with normal pulmonary structures, such as the cavity that separates the lungs. This cavity contains the heart, large blood vessels, trachea, thymus, and connective tissues.
It is an object of the present invention to effectively segment an anatomical structure, such as a pulmonary lesion, from the background tissue in a volumetric medical image.
This object is achieved by a method of analyzing a lesion in a medical digital image using at least one point contained within a lesion to be analyzed comprising:
a) propagating a wave-front surface from the point(s) for a plurality of steps;
b) partitioning the wave-front surface into a plurality of wave-front parts wherein each wave-front part is associated with a different portion of the wave-front surface corresponding to a previous propagation step; and
c) analyzing at least one feature associated with each wave-front part to classify anatomical structures associated with the lesion and normal anatomy within the medical digital image.
The present invention has an advantage of distinguishing a variety of different anatomical structures within the context of a region growing image segmentation algorithm. In particular, the present invention can distinguish between the structures of pulmonary lesions, pulmonary lesion speculations, blood vessels, and normal solid tissues such as the chest wall or heart.
a. is diagram showing the wave-front parts generated by the region growing processing;
b. is diagram showing the wave-front parts generated by the region growing processing;
a depicts a circumscribed lesion;
b depicts a circumscribed lesion with spiculations;
c depicts a juxta-vascular lesion;
d depicts a juxta-vascular lesion growing around vessels;
e depicts a sessile juxta-pleural lesion;
f depicts a pedunculated juxta-pleural lesion; and
Many medical imaging applications are implemented via a picture archiving and communications systems (PACS). These systems provide a way for displaying digital images acquired by a wide variety of medical imaging modalities such as, but not limited to, projection radiography (x-ray images), computed tomography (CT images), ultrasound (US images), and magnetic resonance (MR images). Each of the above mentioned medical imaging modalities contain slightly different diagnostic information. In particular, CT and MR images when viewed and studied by radiologist, can reveal much detail about a patient's 3-dimensional internal anatomy. Computer algorithm technology can also be applied to medical images to enhance the rendering of the diagnostic information, to detect an abnormal condition, i.e., computer aided detection (CAD), and to make measurements relating to the patient's condition, i.e., computer aided measurement (CAM).
The present invention represents an algorithmic computer method for segmenting a portion of a medical image with anatomical relevance. In particular, the primary motivation for the development of the technology described herein is the segmentation of abnormal pulmonary lesion tissue from normal tissue. An intended use for the herein described technology is as follows. A radiologist reviews a thoracic CT exam on a medical PACS and indicates to the CAM segmentation software the position of a suspected pulmonary lesion. Alternatively, the position of a suspected lesion can be supplied from an automatic nodule detection algorithmic method. The voxel position indicated represents a seed point assumed to be part of the pulmonary lesion. The CAM segmentation software then identifies voxels surrounding and contiguous with the seed point that are part of the pulmonary lesion. Once the region associated with the pulmonary lesion has been segmented a corresponding volumetric size can be calculated. The technology advancement of the present invention relates to the particular method of performing the image segmentation task.
In the following description, a preferred embodiment of the present invention will be described as a software program. Those skilled in the art will readily recognize that the equivalent of such software can also be constructed in hardware. Since image processing and manipulation algorithms and systems are well known, the present description will be directed in particular to algorithms and systems forming part of, or cooperating more directly with, the method in accordance with the present invention. Other aspects of such algorithms and systems, and hardware and software for producing and otherwise processing the image signals involved therewith, not specifically shown or described herein can be selected form such systems, algorithms, components, and elements known in the art. Given the description as set forth in the following specification, all software implementation thereof is conventional and within the ordinary skill in such arts.
A system suitable for practicing the present invention is illustrated in
A patient is positioned on a movable table within the capture device assembly such that the patient can be moved relative to the signal source of the capture device. The capture device receives the signals that have passed through the patient in raw form, processes these signals, and produces a slice digital image. Each slice digital image represents a cross-sectional, or slice, through the patient anatomy. A slice image series can be generated by repositioning the patient relative to the source signal and acquiring other slice digital images relating to different cross-sections through the patient. A slice image series, assembled in proper order, can represent the full 3-dimensional volume of an imaged patient.
Also connected to a communications network 110 is a digital radiographic (DR) capture device 10b capable of producing digital x-ray images. As such, the images produced by a DR capture device typically are one or more 2-dimensional digital images each representing a different exposure and or imaging path through the patient. For example, the DR capture device 10b can be used to acquire multiple projection radiographic digital images with its x-ray source located in different positions relative to the patient. The resulting DR radiographic digital images can be processed to produce a set of tomosynthesis slice digital images that represent a 3-dimensional digital image of the patient.
The slice digital images (not shown) produced by the capture device 10a are transmitted via the communications network 110 to an image archive computer 140 where, along with patient history information, they become part of an electronic patient history record. The main function performed by the image archive computer 140 is the facilitation of transfer or exchange of image information rather than the processing of information. The image archive computer 140 serves as a large storage repository of digital images and other, but unspecified, medical information. The arrows in the diagram of
The slice images are later queried on a diagnostic workstation computer 120, sometimes referred to as a PACS (Picture Archive and Communication System), for viewing and examination by a radiologist or similarly trained medical professional. The diagnostic workstation computer 120 can have multiple electronic displays connected for viewing medical images. Two such electronic display devices 42a and 42b are shown in
A PACS (Picture Archive and Communication System) can be defined as a system that acquires, transmits, stores, retrieves, and displays digital images and related patient information from a variety of imaging sources and communicates the information over a network. By this definition, the diagnostic workstation computer 120 shown in
For the purposes of the discussion of the present invention the collection of inter-connected computers including the communications network will be also be referred to as a DICOM network since DICOM (Digital Imaging and Communications in Medicine PS 3.3-1999, National Electrical Manufacturers Association, 1999) formatted digital images are currently is the industry standard and the most prevalent file encoding used for medical digital images. Typically a 3-dimensional volume image is constructed from set of 2-dimensional slice digital images wherein each slice digital image is encoded as an individual DICOM file.
Often an individual digital image data element, i.e., single value representing a measured image signal intensity, is referred to as a for 3-dimensional images and a pixel for 2-dimensional images. The term voxel is commonly used to characterize a volume-element whereas the term pixel is commonly used to characterize a picture-element. The technology embodied within the present invention can be applied to 2-dimensional and 3-dimensional images. As such, for the purposes of the description herein, the terms voxel and pixel should be considered interchangeable, i.e., describing an image elemental datum capable of having a range of numerical values. Voxel and pixels can be said to have the attributes both of location and value.
With the medical imaging system setup as described above and shown in
The term lesion can be defined in medical dictionaries as “any pathological or traumatic discontinuity of tissue or loss of function of a part”. For the purposes of the discussion presented herein, focal (localized) lesions are of greatest interest, i.e., abnormalities that can be visually attributed to a localized domain. Some technical presentations use the term lung nodule or pulmonary nodule interchangeably with the term pulmonary lesion. Although the term nodule can be defined in medical dictionaries as “a small lump, swelling or collection of tissue” the discussion here will treat the term nodule as a subset of the lesion. As the following discussion of technology will illuminate, the present invention is designed to work with lesions of general nature as well as smaller nodular lesions.
An overview of the flow of image information into, out of, and within software application program 140 is depicted by
As an option, an image renderer 240 receives the volume image 202 and lesion segmentation map 204 and generates the rendered segmentation map 209. A radiologist or other medical professional can display the rendered segmentation map 209 on one of the display devices 42a or 42b. In so doing, the medical professional can visualize the geometry of the suspected lesion in relation to patient's normal anatomy. Both the rendered segmentation map 209 and volume image 202 can be visualized in a 2-dimensional presentation (slice presentation) or in a 3-dimensional presentation (volume rendering presentation) such as the method described in U.S. Pat. No. 5,187,660.
Pulmonary lesion size is considered a valuable diagnostic quantity. In particular, nodule size correlates with the probability that a nodule represents cancerous tissue. The volume estimator 250 (shown in
Many examples of CT volume images have anisotropically sampled voxel data, i.e., the voxel spacing in the x, y, and z-directions is not equal. Typically, the x and y-direction voxel samples are equally spaced and the z-direction voxel spacing is somewhat greater The set of source digital images 201 can be interpolated in the z-direction such that the voxel image data is isotropically resampled. The software application program 150 can then be operated on the resampled source digital images yielding a reported volume based on isotropic data.
Although the technology of the present invention was designed, developed and tested with a user providing the seed point 205a for the starting point of the image segmentation processing, those skilled in the art will appreciate that there are other methods of identifying seed points that can be used to good effect. In particular, any computer aided detection (CAD) algorithm that is designed to automatically identify the location of suspected nodules, such as the method described in U.S. Pat. No. 5,638,458, can be used to provide the seed point 205a.
The details of the lesion segmentor 230 shown in
The details of the initial object segmentor 310 shown in
In region growing algorithms, the processing can start with a point or collection of points. In either case, new voxels are added to the previously segmented voxels (classified), either one-at-a-time or a set of voxels at a time. (The present invention can be used with either type of region growing method). Within the propagate the wave-front step 312, (especially for the one-at-a-time region growing method) preferably many iterations of adding new voxels to the current segmentation map are performed. The number of voxels added to the segmentation map during a propagation step is controlled by D1; a parameter in the image processing parameters 206. The voxels added during the propagation step are associated with the current wave-front. In a preferred implementation, D1 represents an increment in the path-length traveled from the current wave-front to the voxels currently under consideration.
After the propagate the wave-front step 312 the set of voxels connected to the previous segmented voxels constitute a wave-front surface from which further region growing can be initiated. In an alternative implementation of the invention, the wave-fronts contain the segmented voxels that constitute the boundary between segmented and non-segmented voxels. In a classify the wave-front surface into active and inactive parts step 313, the set of voxels that constitute the current wave-fronts are classified into active and inactive categories. Inactive voxels are voxels in which no further growth of the segmentation can occur according to the segmentation criteria. Voxels that can initiate further growth, i.e., voxels that can be incorporated into the segmentation map, are considered “active”.
The active voxels represent the wave-front surface of the region growing process. However, the set of active voxels do not necessarily constitute a contiguous region, i.e., they are not all connected to each other. In a classify the wave-front parts into separate parts step 314, the set of active voxels and the set of inactive voxels representing the overall wave-front surface are further classified, or partitioned, into separate parts, a plurality of wave-front parts, wherein the voxels of each wave-front part are connected. Each separated wave-front part is given a unique label. Starting with an arbitrary active(inactive) voxel v1, a connected component analysis is performed on the set of active(inactive) voxels. All voxels within the set of active voxels adjacent to v1 or adjacent to voxels adjacent to v1 (and therefore connected) are identified as part of the same wave-front part #1. After identifying all the connected voxels to v1, a v2 voxel is arbitrarily chosen from the set of remaining active(inactive) voxels. In similar fashion, all of the voxels connected to v2 are identified as belonging to a new wave-front part #2. This process is repeated until all voxels in the set of active(inactive) voxels are identified and associated with a particular wave-front part.
When the classify the wave-front parts into separate paths step 314 is performed, each identified wave-font part is associated with an identified wave-front part from the previous loop. Initially, the overall wave-front surface starts out as a single wave-front part, e.g. #1. At some later iteration pass, the overall wave-front surface is no longer a set of connected voxels and thus there become a plurality new wave-front parts, e.g. #2 and #3. Therefore, the total collection of wave-front parts can be organized in a tree structure wherein each individual wave-front part has a parent wave-front part and many have one or more child wave-front parts. Each segmented voxel in the current segmentation map is associated with a labeled wave-front. Voxels that were added during the propagate the wave-front step 312 that were not part of a labeled wave-front part are associated with the parent wave-front part. Alternatively these voxels can be associated with the child front of the parent. These voxels correspond to the volume that is between two successive wave-front parts. A wave-front path is the list of distinct wave-fronts parts traversed starting from the seed point 205a in which successive wave-fronts parts are linked together by a parent child relationship and which is terminated by a wave-front part (terminal wave-front) who does not posses a child wave-front part. There is one unique wave-front path between the seed point 205a and each terminal wave-front part. The collection of the these paths constitutes the wave-front paths 208. Thus, each wave-front part of a wave-front surface is associated with a portion of the wave-front surface corresponding to a previous propagation step.
The diagram shown in
The wave-front parts 384, 385, 392, 393, and 394 represent a single wave-front path. Similarly, wave-front parts 384, 385, 388, 395, and 396 represent a different wave-front path.
Region growing methods, in general, start with a seed point (single point) or a set of points, and add new points to the classified region that meet a growth criteria as well as the condition of begin adjacent. Connectivity can be defined in many different ways. The technology of the present invention can be used with a variety of different adjacency criteria depending on the needs of the application. The simplest connectivity criteria includes the voxels, about a voxel vi,jk, that are one index different in one of the three cardinal directions. Let the indices i, j, and k refer to the x, y, and z-directions respectively. With this simple definition of connectivity, the following six voxels are considered connected vi+1,jk, vi−1,jk, vi,j+1k, vi,j−1k, vi,jk+1, and, vi,jk−1. For a 2-dimensional application there would only be the four connected voxels vi+1,j, vi−1,j, vi,j+1 and, vi,j−1 with this simple criterion. Other connectivity criteria can include the eight corner voxels about the a voxel vi,jk, e.g. vi+1,j+1k+1, vi+1,j+1k−1, vi+1,j−1k+1, vi−1,j+1k+1, vi+1,j−1k−1, vi−1,j+1k−1, vi−1,j−1k+1, and vi−1,j−1k−1. Still further, there can be as many as 26 connected voxels to voxel vi,jk if all of the voxels within one index of i, j, k are considered. Thus, the present invention can be used with many definitions of connectivity. However, with fewer connected voxels considered, some diagonally connected structures can be omitted in the region growing process. In general, there is a trade-off between computational efficiency and region growing integrity.
On the first pass through the loop of steps 312, 313, and 314, a continue propagation? step 316 (
In an alternative embodiment, an analyze each active wave-front path step 315 (
Pulmonary lesions can be classified into different types of lesions by their juxta-position to normal anatomy. The lungs can a variety of normal tissue structures that are discernable when investigating a CT or MR volume image. For example, the normal lung parenchyma tissue appears as dark (voxel values typically in the range of −700 to −1000 Hounsfield units (HU)) featureless regions. Blood vessels, arteries and veins, appear as meandering tubes with voxel values varying in value depending on the diameter of the vessel. Due to partial volume effects, thinner diameter vessels, on CT exams, will have lower values typically in the −700 to −500 range. Larger diameter bronchia appear as hollow tubes with the thinnest bronchia appearing much like blood vessels. The lung parenchyma is border by adjacent structures such as the chest wall, heart, and diaphragm.
Some pulmonary lesions grow in the midst of the lung parenchyma without perceivably intersecting any of the above mentioned normal anatomy structures. This type of pulmonary lesion can be termed “circumscribed” since the lesion is essentially self-contained. Although a circumscribed lesion does not intersect with a normal anatomy structure, it still can be irregular in shape and contain spiculated tentacles, i.e., thin tapering lesion structures. Spiculated structures typically taper to a point, however, some taper and then connect to secondary lesion dislocated from the primary. Juxta-vascular lesions within the lung parenchyma next to and sometimes encompassing a vessel. Juxta-pleura lesions grow within the lung parenchyma next to the pulmonary parietal pleural membrane, i.e., the pulmonary parietal pleural membrane that encompasses the lungs. Since the pulmonary parietal pleural membrane itself is not always discernable, these lesions appear to grow next to the normal structure on the other side of the pulmonary parietal pleural membrane such as the heart, diaphragm, or chest wall. Juxta-pleural lesions can be further typed as either sessile or pedunculated. Pedunculated juxta-pleural have a pedicle, or tentacle-like structure that connects the main part of the lesion to the pulmonary parietal pleural membrane. Sessile juxta-pleural lesions have no discernable pedicle and appear to grow as a protuberance from the pulmonary parietal pleural membrane.
The above mentioned lesion types have similar voxel value ranges for the voxels that are part of the lesion. Typically, the lesion tissue has similar properties to fat and soft tissue (solid with no voids) with corresponding voxels values on CT exams ranging from −500 to +100 HU. Lesions can also exhibit calcified sub-structures, and therefore, can have CT voxel values that exceed +100 HU. In addition to the above mentioned lesion types, there is a non-solid lesion type that exhibits quite different properties owing to a much reduced opacity. For the non-solid lesion type, the voxel values are relatively close to that of the surrounding lung parenchyma at about −750 to −600 HU. In fact, the non-solid lesions are sometime difficult for observers to visualize.
In the diagrams
For illustrative purposes, the
Referring to
Features for each wave-front part k and the volume associated with each wave-front part can be computed, such as, the number of voxels Nk, the center of gravity (centroid) Ck, shape and orientation, direction between successive wave-fronts parts, curvature, etc. The centroid of a wave-front part is defined as
where Xi is the coordinate of the ith voxel in wave front part k. The direction between successive wave front parts is defined as
D
k
=C
k+1
−C
k
and the curvature is defined as
κk=|Dk−Dk−1|.
The shape and orientation of the kth wave-front part can be determined by the eigenvalues and eigenvectors of the covariance matrix of the voxels spatial coordinates X that makeup the kth wave-front part. The eigenvectors yield the orientation or principal axes of the front and eigenvectors yield information related to the shape of the front. For example, if eigenvalues λ are ordered in magnitude such that λ1≦λ2≦λ3, then a front associated with a vessel is indicated by λ1 being small (ideally zero), and λ2 and λ3 are of larger and equal magnitude. The respective eigenvectors u1 indicates the direction along the vessel and u2 and u3 form the normal (orthogonal) plane.
The computed features associated with each wave-front part can then be used to analyze the anatomical structures associated with the lesion and normal anatomy within the medical digital image.
It should be noted that the calculation of surface normal vectors can be computationally intensive and somewhat inexact. This is mostly due to noise in the capture system and course spatial sampling relative to the anatomical structures of interest. For isotropically and anisotropically sampled voxel data, the number of voxels contained in a wave-front part can be considered a surrogate for surface area. For many cases, any of measure of wave-front part surface area plotted for successive wave propagation steps reveals much about the underlying anatomical structure.
Idealized plots for four different anatomical structures are shown in
Another anatomical structure that can be differentiated by analyzing the properties of the wave-front is a spiculated tentacle. As shown in the graph depicted in
Conversely, vessels tend to have corresponding progressions of surface area that are relatively constant as a function of path-length. Additionally, larger vessels have correspondingly greater number of voxels and larger surface area values. The surface area of the wave-front parts that correspond to vessels essentially represents the cross-sectional area of the vessel since the wave-front surface progression tends to follow the axis of vessels. Therefore, medium to larger size vessels can be differentiated from other anatomical structure types by virtue of having approximately constant surface area and surface area values above approximately 20 mm2, or about 5 mm in diameter. (Similarly, the square root of the number of voxels or the square root of the surface area can be used as an effective structure diameter when analyzing the wave-front parts). By observation, most spiculated tentacles have smaller corresponding surface area values associated with their respective wave-front parts.
Lesions, especially cancerous lesions, can grow along vessels and form spiculated tentacles that connect to another lesion. For these cases, it can be difficult to ascertain which lesion is the primary and secondary or if both lesions should be considered a single lesion. When a lesion is connected to another structure, e.g. another lesion, the chest wall or heart, the progression of surface area values, or number of voxels, can initially diminish as a function of path-length and then increase. The increase in the surface area value corresponds to the wave-front surface passing the intersection point with a different anatomical structure. If the wave-front parts of a wave-front path progress along a vessel, the corresponding surface area values can stay relatively constant (when in the vessel) and then increase when a larger anatomical structure is intersected as shown in the graph of
The wave-front paths, or sections of a given wave-front path, are classified by the manifestation of predetermined relationships between successive wave-front parts of a given wave-front path. For several of the case types described above, not all of the wave-front parts of a wave-front path correspond to a single anatomical structure. Therefore, each of the wave-front parts of a wave-front path must be given a structural classification. For example, the wave-front parts of wave-front paths identified as spiculated tentacles, can all be classified as spiculated tentacle type.
Vessel-like structures can be identified by examining a plot of the surface area, or number of voxels, in a wave-front part as a function of successive wave-front parts of a wave-front path (see
Spiculated tentacle-like structures can be identified in a similar manner characterized by a diminishing progression of wave-front part surface area over a relatively short length with termination. The wave-front parts that occur before (i.e., closer to the seed point 205a) identified vessel-like structures are classified as lesion type. The wave-front parts that occur after (i.e., further from the seed point 205a) identified vessel-like structures are unclassified and can be removed from the segmentation map.
Lesions that are attached to large normal structures, such as the pleural wall, heart, or mediastinum, are identified by looking for wave-front paths that have a progression of wave-front part surface area that increases at a large rate for a relatively long path length (see
Alternatively, for the pedunculated juxta-pleural lesion type wave-front paths, the end of the pedicle structure must be defined. The preferred method uses the minimum surface area value Sm as the pinching point. The wave-front parts that have corresponding surface area values greater further away from to the seed point 205a than Sm are unclassified and removed from the segmentation map. Alternatively, the pinching point can be determined by calculating the corresponding wave-front part that has a surface area value greater than Sm+τ2. (another small tolerance) and is further away from the seed point 205 then the wave-front part corresponding to Sm. Good values for the small threshold values τ1 and τ2 are in the range from 1 to 2 mm2. However, these values can be set to a particular radiologist's preference.
Human anatomy exhibits great variation. Therefore, the above-described distinguishing characteristic progressions of surface area values should be considered guidelines rather than definitive, i.e., there will always be some anatomical structure classification errors. However, each of the above mentioned types of anatomical structural intersections do have distinguishable patterns. Thus, the following lesion types can be differentiated: circumscribed, juxta-vascular, juxta-pleural, pedunculated juxta-pleural, sessile juxta-pleural, and circumscribed spiculated. Additionally, a particular wave-front path can be distinguished as a lesion (distance squared progression about the seed point), vessel (relatively constant surface area progression), spiculated tentacle (diminishing progression of surface area values and relatively short length with termination), and large structure (distance squared progression of surface area).
Referring to
Alternatively, the segmentation map pruner 330 can employ a morphological filter to perform some of the pruning. For example, a morphological filtration closing operation, i.e., an erosion followed by a dilation operation, can be performed on only the voxels of the initial segmentation map 203 that area associated with wave-front parts classified as vessels. The same morphological filtering can be performed selectively for the case of pedunculated juxta-pleural classified wave-front parts.
The present invention can be used with different image processing region growing method. The preferred embodiment employs a version of the Fast-Marching method described by J. A. Sethian in the publication “Level Set Methods and Fast Marching Methods”, Cambridge University Press, 1999. The fast-marching approach is a one-voxel-at-a-time approach wave-front propagation technique. Starting with a seed point (single point), a set of point such as a line, or a surface of points, connected voxels are either added to the segmentation map or classified as not-part of the segmented object based on a cost-function criterion. Many different cost-function criteria can be used in the context of a fast-marching approach. For example, the cost-function can be based on the magnitude of the gradient of voxel values. Another cost-function can be based on the curvature of surface normal vectors. Additionally, any combination of cost-functions can serve as cost-function. The preferred embodiment uses a cost-function based on the voxel value and a constant threshold. The cost function returns 1 for voxels whose value is above the threshold and returns infinity for voxels whose value is below or equal to the threshold. By choosing a cost function that returns infinity for a voxel value below the threshold inactivates or freezes the growing front at that voxel. This type of cost function yields a geodesic distance map that associates each nodule voxel with its geodesic distance from the seed point. For example, a threshold value of −350 HU works well for segmenting solid-type pulmonary lesions. However, the cost-function threshold value (included in the image processing parameters 206 of
Those skilled in the art will recognize that there are many different algorithms that can be used to calculate a geodesic distance map or a cost weight variant of geodesic distance map that can be used in the current invention.
Those skilled in the art will recognize that the level set method approach described by J. A. Sethian can also be used with the present invention. The level set method a many-at-once wave-front surface progression method that can incorporate more mathematically complex cost-functions (with a corresponding increase in required computational resources).
Those skilled in the art will recognize that the simple case of a voxel threshold value based cost-function can be implemented with a modified connected component algorithm. That is, a connected component algorithm as described above, can be governed by connectivity and comparison to the voxel threshold value.
Although the technology described herein was developed and tested with volumetric patient images with the intent of segmented pulmonary lesions, those skilled in the art will appreciate that the general processes described herein can be used to segment anatomical features other than pulmonary lesions. For example, liver lesions have many geometric similarities to pulmonary lesions. However, it should be recognized that although the principles of the present invention can be applied to other domains, the particular logic and control parameters would most likely be application specific.
Those skilled in the art will recognize that although the present invention has been described in the context of segmenting and analyzing pulmonary lesions, the technology can be used to segment and analyze lesions in other organs of the body as well. For example, hepatic lesions grow in and around blood vessels and against other normal anatomical structures adjacent to the liver.
Although the present invention was primarily motivated by a medical digital imaging application, those skilled in the art will recognize that the technology described herein can be practiced with non-medical digital imaging applications. A non-medical application for which the present invention is well suited is non-destructive testing using x-ray sources. As long as the imaged objects have differentiated geometric shapes, the wave-front analysis approach described herein is applicable. For this generalized application, the differentiation is not between normal an abnormal tissue, it is between objects of dissimilar shape. For example, tube-like structures are akin to the vessel-like structures described above. Broad objects with low curvature, i.e., relatively flat, are akin to the chest wall as described above.
Although the present invention was primarily motivated by a 3-dimensional medical digital imaging application, those skilled in the art will recognize that the technology described herein is applicable to 2-dimensional image analysis applications as well. Most region growing techniques, and the Fast Marching technique in particular, have a corresponding 2-dimensional formulation. For example, the above 3-dimensional technology can also be applied to a single CT slice, i.e., a 2-dimensional entity. For 2-dimensional imaging applications, a wave-front surface is really a curve, i.e., a connected set of points that lie in a plane. Thus for the purposes of the discussion herein, the term “surface” also include “curves” for 2-dimensional applications.
The image processing operations depicted in the description above can be implemented and stored as a computer program stored in a computer readable storage medium. The computer readable storage medium can include, for example: magnetic storage media such as a magnetic disk (such as a floppy disk) or magnetic tape; optical storage media such as an optical disc, optical tape, or machine readable bar code; solid state electronic storage devices such as random access memory (RAM), or read only memory (ROM). The associated computer program implementation of the present invention can also be stored on any other physical device or medium employed to store a computer program. It is also possible to embed a software implementation of the present invention on an electronic circuit.
The invention has been described in detail with particular reference to certain preferred embodiments thereof, but it will be understood that variations and modifications can be effected within the spirit and scope of the invention.
