DESC:FORM 2
THE PATENTS ACT, 1970
[39 of 1970]
&
THE PATENTS RULES, 2003
COMPLETE SPECIFICATION
(Section 10; Rule 13)

A UNIFIED BAYESIAN MODEL FOR ELECTRONIC CLEANSING IN CT COLONOGRAPHY

SAMSUNG R&D INSTITUTE INDIA – BANGALORE Pvt. Ltd.
# 2870, ORION Building, Bagmane Constellation Business Park,
Outer Ring Road, Doddanakundi Circle,
Marathahalli Post,
Bangalore -560037, Karnataka, India
Indian Company

The following Specification particularly describes the invention and the method it is being performed

RELATED APPLICATION
The present invention claims benefit of the Indian Provisional Application No. 5209/CHE/2014 titled "A UNIFIED BAYESIAN MODEL FOR ELECTRONIC CLEANSING IN CT COLONOGRAPHY” by Samsung R&D Institute India – Bangalore Private Limited, filed on 17th October 2014, which is herein incorporated in its entirety by reference for all purposes.

FIELD OF THE INVENTION

The present invention generally relates to the field of Electronic Colon Cleansing (ECC) in CT Colonography (CTC) and particularly relates to a method of Bayesian formulation of the cleansing model.

BACKGROUND OF THE INVENTION

In order to obtain high quality CT images, the colon should be well-cleansed and well distended. The patient is asked to maintain a clear liquid diet starting 24hs or 48hs before the study and to ingest a cathartic or laxative the night before the examination. Further colon distention is carried out by placing a thin tube into the rectum and performing the colon insufflation with air.

However, this process is uncomfortable for the patient; a trade-off is fecal tagging, with a contrast agent, orally administered. Due to the gravity, stool appears at the bottom of the colon (depending on prone or supine acquisition), with a horizontal interface as shown in Figure 1(a).

The colon is automatically segmented via iterative blob detection of air filled blobs, along with a crude cleansing to remove tagged stool that are adjacent to and below air. The stool density is estimated from a simple KMeans classification; the class with the highest mean being stool. This is followed by extraction of its end points (rectum, cecum) by locating two points with the furthest geodesic distance, represented in Figure 1(b). This is followed by solving the shortest path from the cecum to the rectum with the cost function being the Euclidean distance, represented in Figure 1(c) to the colonic wall to provide with the centerline, represented in Figure 1(d).

The Endo-luminal visualization is produced by ray tracing that is designed to mimic isosurface rendering, with the goal of producing high quality isosurface. In CPU ray tracer, the analytic method of finding the isosurface is employed, which solves a cubic polynomial to find the isosurface within a voxel. In GPU ray tracer, this is approximated by texture interpolation with subdivision search to localize the isosurface to a precision of 0.01mm.

In general, the cleansing method must therefore generate a reliable boundary and regenerate the partial volume at the boundary, otherwise one obtains staircase artifacts of the form. In principle, good cleansing strategies may also deal with unevenly tagged residue (stool). Typically inhomogeneous tagging is less common, and occurs due to reduced or non-laxative CTC data, resulting from a mixture of semisolid fecal materials, air bubbles, fat, undigested foodstuffs, and unevenly distributed contrast agents. Poor cleansing results in several artifacts during Endo-luminal navigation as described herein.
• Pseudo-enhancement: Soft-tissue structure degradation, due to the adjacent high-radio density tagging. In practice, this results in colonic mucosa being dismissed as enhanced fecal material
• Pseudo polyps and false fistulas: caused by the similarity between the air-tagging boundary and the thin, soft tissue and the PV effect in CT. This phenomenon can virtually erode the mucosal structure and put a false fistula or hole in the colonic wall that doesn't exist.
• Staircase artifacts: Due to poor regeneration of the partial volume (PV) layer. The Endo-luminal visualization shows step artifacts.
• Stalactites: due to under cleansing or inhomogeneous tagging

Therefore, cleansing must be carefully performed, even at the cost of not removing entire layer of barium coating.
In view of the foregoing, there is a need for providing an effective Electronic Cleansing mechanism for CT colonoscopy.

The above mentioned shortcomings, disadvantages and problems are addressed herein and which will be understood by reading and studying the following specification.

SUMMARY OF THE INVENTION

The various embodiments of the present invention disclose a method for Electronic Colon Cleansing (ECC) in CT Colonography (CTC) with a unified Bayesian model.
A unifying framework is disclosed that combines prior information (derived from local structure), scan orientation information (prone/supine: is the stool likely to be above or below air filled pockets), inhomogeneity and enforcement of spatial consistency of the tissue classes into a single generative model.

According to an embodiment of the present invention, the method for Electronic Colon Cleansing (ECC) in CT Colonography (CTC) with a Unified Bayesian model comprises of acquiring a colonoscopy scan image data associated with at least one of a scan orientation, location of stool around air filled pockets, one or more tissue classes, inhomogeneity and enforcement of inconsistency of the one or more tissue classes, identifying missing data by using an Expectation Maximization (EM) method based on the acquired colonoscopy scan data and formulating the electronic cleansing in an expectation maximization framework. The EM is executed using a sparse variant instead of using whole dataset to generate the one or more tissue classes. The EM is used only in the one or more voxels in the vicinity of the boundaries and folds, while avoiding voxels which are at least one of tagged stool, soft tissues and air pockets.

According to an embodiment of the present invention, the EM method comprises of obtaining values associated with a discrete observation of continuous uncleansed CT intensities, obtaining a set of unobservable variables representing partial volume contributions, performing a hard segmentation on the acquired colonoscopy scan image for labeling of tissues so as to maximize posterior probability, calculating a voxel-wise probabilities for each of the tissue classes and identifying a spatial interaction in voxel neighborhood based on Markov random field model.

According to an embodiment of the present invention, the voxel-wise probabilities for each of the tissue classes is calculated based on a Gaussian model using mean and variance in the calculated tissue classes.

According to an embodiment of the present invention, the one or more tissue classes comprises air bubbles, soft tissue and stool.

According to an embodiment of the present invention, generation of the sparse variant comprises of computing crude segmentation of the available dataset based on the one or more tissue classes, calculating means for the segmentation associated with the one or more tissue classes and computing a class variance for the segmentation associated with the one or more tissue classes.

According to an embodiment of the present invention, the stool is identified based on the classification and a preset criterion.

According to an embodiment of the present invention, the preset criterion for recognizing stool comprises a PV voxel belonging to a class with the highest mean intensity of the voxels which are at least one of tagged stool, soft tissues and air pockets on the boundary that face upwards, at least 20% exposed to air, and at 3mm of PV tissue.

The foregoing has outlined, in general, the various aspects of the invention and is to serve as an aid to better understand the more complete detailed description which is to follow. In reference to such, there is to be a clear understanding that the present invention is not limited to the method or application of use described and illustrated herein. It is intended that any other advantages and objects of the present invention that become apparent or obvious from the detailed description or illustrations contained herein are within the scope of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

The other objects, features and advantages will occur to those skilled in the art from the following description of the preferred embodiment and the accompanying drawings in which:

Figures 1a-1d are snap shots where 1(a) represents an axial slice showing the air filled sacs of the colon with tagged stool, 1(b) represents initial segmentation of the colon along with crude cleansing and detection of the rectum and cecum, 1(c) represents Euclidean distance to the colonic wall, and 1(d) represents extracted centerline for endoluminal navigation, according to a prior art illustration.

Figure 2 is a snapshot representing the partial volume mixtures of the tissue classes, according to a prior art illustration.

Figures 3a-3d are snapshots representing EC in the vicinity of a polyp where 3(a) represents CTC before EC, 3(b) represents result of EC, 3(c) represents magnified image of 3(a), and 3(d) represents magnified image of 3(c), according to a prior art illustration.

Figures 4a-4d are snapshots where 4(a) represents the slice of a CT volume, 4(b) represents initial classification based on KMeans, 4(c) represents blobs of tagged stool in pink separated from the other enhanced regions, and 4(d) represents extracted of partial volume samples, according to an embodiment of the present invention.

Figures 5a-5b are snapshots where 5(a) represents regions with structure and 5(b) represents priors supplied to the Bayesian model, according to an embodiment of the present invention.

Figures 6a-6b are snapshots where 6(a) represents a CT colongraphy scan showing a sessile polyp submerged in stool, and 6(b) represents a threshold local structure in the vicinity of the polyp, according to an embodiment of the present invention. .

Figure 7 is a graphical representation illustrating a plot along the line for the same CT image, according to an embodiment of the present invention.

Figures 8a-8f are snapshots where 8(a) represents a close up of a slice of the CT image, 8(b) represents an EM without the use of any MRF, according to an embodiment of the present invention.

Figures 9a-9c are snapshots where 9(a) represents a slice of a volume, 9(b) represents a MRF smoothing with the additional gravity specific kernel, and 9(c) represents the whole slice, according to an embodiment of the present invention.

Figures 10a-10b are snapshots where 10(a) represents a zoom on the same slice of the cleansed volume, and 10(b) represents the same representative slice, according to an embodiment of the present invention.

Figures 11a-11c are snapshots where 11(a) represents an uncleansed portion, 11(b) represents a cleansed slice with residual stool at the T junction, and 11(c) represents an Endo-luminal view, according to an embodiment of the present invention.

Figures 12a-12c are snapshots where 12(a) represents a Binary mask M created by thresholding the vertical gradient Iz, 12(b) represents the resulting boundary calculated from Iz, and 12(c) represents the final boundary results after 24 iterations, according to an embodiment of the present invention.

Figures 13a-13b are snapshots illustrating a colon, where 13(a) represents an uncleansed view of the colon, and 13(b) represents a cleansed view, according to an aspect of the present invention.

Figure 14 is a flow chart illustrating a method of creating a unified Bayesian model for electronic Colon Cleansing (ECC) in CT Colonography, according to an embodiment of the present invention.

Although specific features of the present invention are shown in some drawings and not in others, this is done for convenience only as each feature may be combined with any or all of the other features in accordance with the present invention.

DETAILED DESCRIPTION OF THE INVENTION

The various embodiments of the present invention disclose a method for Electronic Colon Cleansing (ECC) in CT Colonography (CTC) with a Unified Bayesian model.

The present invention solves the problem of the Bayesian formulation of the stool removal/segmentation using an Expectation Maximization (EM) method with the modeling of the class intensities based on parametric (Gaussian) finite mixtures. The model described in the present invention is designed to address the three problems: (a) the partial volume effect that creates voxels with the attenuations similar to colon wall at the boundary of stool and air (b) heterogeneity in the tagged stool material that results in inhomogeneous tagging. (c) preservation of polyp like regions. The method according to the present invention is formulated within an EM framework which maximizes the conditional expectation of the underlying tissue process given the acquired data. Here sparse set of samples are used for computational efficiency and MRF prior capturing orientation information (prone/supine) to eliminate the PV effect in fecal tagged stool in addition to capturing spatial homogeneity.

The embodiments of the present invention, uses Electronic cleansing (EC) with regeneration of the partial volume interfaces as described in Figure 2 and conservative cleansing in areas likely to comprise polyps as described in Figure 3. The present invention formulates EC in the EM framework.

The embodiments herein describes an overview of the Expectation Maximization (EM) method as follows: Given ‘n’ tissue classes, for instance, 3 tissue classes each one for air, soft tissue and stool, the Expectation-Maximization (EM) framework is natural, given the “missing data” aspect of this problem. The objective functions optimized by the EM framework is stated in the following Bayes equation:

Here
- ‘y’ represents a discrete observation of the continuous uncleansed CT intensities;
- ‘x’ comprises a set of unobservable variables, representing the partial volume contributions. In other words, with K tissue types, tissue k at a voxel i contributes xik to the observation yi such that .
The hard segmentation, assigns to each site in I one of K labels from the finite set L = {l1,..,lK}. The Bayesian (hard segmentation) solution is the tissue labeling which maximizes the posterior probability:

The likelihood, i.e. the voxel-wise probabilities, for any of the K tissue classes, at a voxel i is denoted as pk(yi|lk). This is modeled as a Gaussian, G, for mathematical simplicity, and can be written as where the parameters µk and sk respectively represent the mean and variance of the kth tissue. Marginalizing over all tissue classes and voxels, write the Likelihood term as:
.?k being the mixing proportions.

The likelihood model described above is entirely dependent on the observed intensity. Priors are often applied to enforce spatial homogeneity of the tissue classes, i.e. stool is generally adjacent to stool etc. A Markov Random Field (MRF) models spatial interactions in voxel neighborhoods. The prior term, p(x) is therefore written as below, where Z a normalization factor known as the partition function and U(x) the energy function. As is the case with many other segmentation algorithms of the same family, choose U(x) such that it is only composed of a sum over pairwise interactions between neighboring voxels for simplicity:

ß is a granularity term which weights the contribution of the MRF prior on the segmentation solution. Vij is the interaction between voxel i and another voxel j in its neighborhood. One way of defining Vij is the Kronecker delta, dij, based on the classical Ising potential (also known as a Potts model). Since neighborhoods are non-uniform and more than one neighborhood may be considered, these are often weighted by the distance. dij is the Euclidean distance between the (central) voxel i and the neighborhood voxel j, so that sites in the neighborhood closer to i are weighted more heavily than distant sites.

According to the present invention, with Bayesian formulation of the cleansing model, formulate EC in the EM framework. This is hard EM, but is also inherently simpler with fewer parameters.

According to an embodiment herein, sparse formulation and identifying stool is described as follows. Although EM could be run on the whole dataset, to generate the tissue classes, given its computational expense, use a sparse variant. In other words, not every voxel in the CT image is used as a sample. EM is used only in voxels in the vicinity of the boundaries and folds and avoid EM where it is sure that the voxel is tagged stool, soft tissue, or air. To generate these sparse (or boundary proximal) samples, a crude segmentation of the dataset is first computed. The initial means of the tissue classes are manually chosen.

Figure 4a represents the slice of a CT volume, according to an embodiment of the present invention. Figure 4(d) represents extracted of partial volume samples (brown, orange), according to an embodiment of the present invention.

Figure 4(b) illustrates that a crude classification of the dataset is obtained by KMeans, followed by the computation of class variances, resulting in a Gaussian mixture model. These class variances are estimated from a couple of central slices of the dataset.

Five classes are used to fully capture the PV voxels. The stool is identified on this classification. The criterion used to identify stool is that it should:
(a) belong to the class with the highest mean intensity
(b) Of those voxels that are on the boundary that face upwards, at least 20% must be exposed to air, through at most 3mm of PV tissue.

Figure 4(c) represents blobs of tagged stool in pink separated from the other enhanced regions, according to an embodiment of the present invention. Figure 4(c) appears to sufficiently identify all blobs of stool residue. Further the boundary voxels of the tagged stool (TM) class are identified by morphological methods. A 7x7x7 voxel thick transition zone on this boundary is added and a mask is constructed. Samples outside the mask are discarded. The total number of resulting samples in the result is 2% of the total number of voxels in the CT dataset [512x512x456].

Figure 4(d) represents extracted of partial volume samples, according to an embodiment of the present invention. The method of incorporation of local structure using prior data set is described as follows. Given that voxels are not wanted in the vicinity of polyp like structures to be affected, supply a prior which for the most part is 1/K (uniform), with the exception of a sparse set of voxels that have a local curvature indicative of a polyp. These priors are derived from the image by examining the mean curvature, which characterizes the local folding of the surface.

Figure 5(a) represents regions with structure, in which the first section indicate regions when submerged within the polyp and second section indicates regions when surrounded by air.

Figure 5(b) represents priors supplied to the Bayesian model, which contains regions submerged in tagged stool alone and threshold, according to an embodiment of the present invention. Figures 5(b) and Figure 6(b) show the regions with a prior that is weighted towards the soft tissue class. The color map is indicative of the bias towards soft tissue. This effectively reduces the likelihood of polyps from being misclassified and cleansed away. The other voxels are untouched (ie have a prior with a uniform 3 class distribution). It is worthwhile drawing an analogy with one of the prior arts, who use a “structure-analysis cleansing” which is also derived from the hessian matrix, a level set formulation is not used. In prior art structure-enhancement measure (formulated for enhancing of the soft-tissue structures) is used along with other measures (local roughness etc) into a speed function of a level set method for delineating the tagged fecal materials. This method has similar intent, but formulates this as priors in a Bayesian model.

These are added to the Bayesian model as prior probability. The prior probability that label lk corresponds to a particular voxel, regardless of its intensity, is the following spatially varying mixing proportion:

where tik is the prior probability value at voxel i, derived from the local structure in the image.

Figure 6(a) represents a CT colonography scan showing a sessile polyp submerged in stool and Figure 6(b) represents a threshold local structure in the vicinity of the polyp, according to an embodiment of the present invention.

According to an embodiment herein, MRF spatial priors incorporating direction for tissue interface correction is described as follows. The gravity causes stool to typically face downwards in supine acquisitions and upwards in prone acquisitions. This is clearly seen in the nearly flat appearance of the pockets of stool in Figure 1(a). The tissue at the interface between air and stool exhibits an intensity that is nearly the same as soft tissue. This is illustrated in Figure 7, which is a graphical representation illustrating a plot along the line for the same CT image with X axis units in mm, and Y axis represents HU densities.

Poor correction at this interface layer leads to pseudo polyps and false fistulas and a variety of ways have been used to mitigate this. One way to resolve pseudo polyps and false fistulas is to simply use MRF regularization. However, given the thickness (in voxels) of this transition zone, MRF regularization simply fails to do the job. The Figures 8(a)-8(e) illustrate this, where 8(a) represents a close up of a slice of the CT image, 8(b) represents an EM without the use of any MRF; EM with the use of a smoothing weight of 2, a regular MRF (potts model) with a neighborhood of radius 8(c) 1 (ie 3x3), 8(d) 2, 8(e) 3, and 8(f) 4.

MRF regularization is applied on the volume and a zoomed region of the representative slice is obtained. Only with a large radius and a coercive MRF smoothing factor as shown in Figure 8(f) do the voxels (misclassified as tissue) in the air-stool transition zone get regularized. In the process, the haustral folds and other colonic wall protrusions get eroded away.

According to an embodiment herein, the Gravity vector , is typically [0, 0, -1] for supine acquisitions and [0, 0, 1] for prone acquisitions. Consider the central voxel i, in the partial volume transition region that has been misclassified as soft-tissue. Presumably, the voxel i is in some sufficiently large neighborhood surrounded by air and stool. Gravity causes most of the air to be above and stool to lie below. The interaction term is written due to gravity , between a voxel i and another voxel j in its neighborhood formally as a combination of the interaction terms with air and stool in its neighborhood

Here is the distance from the ith to the jth voxel in mm; denotes the unit vector from the ith to the jth voxel. The dot product with the gravity vector, is a measure of how aligned the voxels are with gravity and the inverse weighting by the distance is applied as before so that sites in the neighborhood closer to i are weighted more heavily than distant sites. The resulting term, is a measure of how much air is above tissue in the neighborhood under consideration, while, is the same for stool below. To ensure that the terms are strictly positive, the results are clamped at 0 (i.e. intended to incorporate a penalization for soft tissue being below stool and not encourage its converse (soft tissue being above stool)).

Finally the measure for the soft tissue class is an AND aggregate of the two, and provides a measure of how much air lies above and stool lies below a soft tissue class. This is . The MRF propagation weights are transferred to the air and the stool classes, so as to encourage this sandwiched voxel, to re-organize itself as air or stool in the next iteration i.e. for the 3 classes/labels (air, tissue, stool), , where the weight lies between 0 and 1 and is simply computed as the relative distance to the centroid of the two classes (air and stool) at this iteration.

Intuitively, such a Markov random field encourages a soft tissue voxel sandwiched (along the gravity direction) between air and stool in a local neighborhood to reorganize itself as either air or stool depending on what its more likely to be.

Finally, this term is aggregated with the regular MRF terms, which are formulated in using a normalized exponent term, with the additional modifications to include the above priors. So the MRF prior for a given voxel, and a specific class labeling is formulized as:

where
k is the class for which the prior is being updated at the current iteration.
ß as before is the MRF smoothing factor that favors segmentations that are spatially extended
? is the MRF smoothing factor for the term described above.

The results of applying this are shown in Figure 9a-9c. As can be seen, the partial volume misclassification is rectified, while the other voxels are largely untouched, where 9(a) represents a slice of a volume, 9(b) represents a MRF smoothing with the additional gravity specific kernel, with a neighborhood of radius of 1x3x1. (3 along the Y direction), and 9(c) represents the whole slice, according to an embodiment of the present invention.

Figure 10(a) represents a zoom on the same slice of the cleansed volume and 10(b) represents the same representative slice, according to an embodiment of the present invention.

According to an embodiment herein, the stool removal process is described as follows. The segmented stool, from the partial volume regions, is then merged with that from the non-partial volume regions that lie in the core. The result is smoothed with a Gaussian with a sigma of the average pixel spacing in the dataset to approximate the effects of the Point Spread Function (PSF) of the scanner. This is subtracted from the uncleansed volume to produce the cleansed image as shown in Figure 10(b).

Several artifacts still remain in the cleansed volume. These can be categorized as falling in one of the following types as Meniscus and Shallow puddles.

According to an embodiment herein, meniscus for liquid in a container is the curve in the surface of a liquid and is produced because the liquid molecules have a different attraction to each other than to the container. The artifact caused by meniscus is referred to as residue that remains in tub after bath. These are comprised of meniscus regions (manifesting as jutting out fragments at the air-stool-soft tissue interface).

Figure 11 shows one such artifact. Figure 11(a) represents an uncleansed portion, Figure 11(b) represents a cleansed slice with residual stool at the T junction, and Figure 11(c) represents an Endo-luminal view, according to an embodiment herein. In case of the present invention, its presence is limited to regions where there is very little air above these partial volume voxels due to the fold being present right above the stool, therefore the MRF prior described earlier plays little role in causing these partial volume voxels to change their labeling to air or stool.

Figure 12(a) represents a Binary mask M created by thresholding the vertical gradient Iz, tagged material is in below the mask and air is on top of it, 12(b) represents the resulting boundary calculated from Iz using the binary mask and the exponential weight w in the first iteration, and 12(c) represents the final boundary results after 24 iterations, according to an embodiment of the present invention.

Figures 13a-13b are snapshots illustrating a colon, where 13(a) represents an uncleansed view of the colon, and 13(b) represents a cleansed view after the use of the priors, according to an aspect of the present invention.

Figure 14 is a flow chart illustrating a method of creating a unified Bayesian model for electronic Colon Cleansing (ECC) in CT Colonography, according to an embodiment of the present invention. At step 1402, a colonoscopy scan image data associated with at least one of a scan orientation, location of stool around air filled pockets, one or more tissue classes, inhomogeneity and enforcement of inconsistency of the one or more tissue classes is acquired. At step 1404, identify identifying missing data by using an Expectation Maximization (EM) method based on the acquired colonoscopy scan data. Further at step 1406, formulate the electronic cleansing in an expectation maximization framework. The EM is executed using a sparse variant instead of using whole dataset to generate the one or more tissue classes. The EM is used only in the one or more voxels in the vicinity of the boundaries and folds, while avoiding voxels which are at least one of tagged stool, soft tissues and air pockets.

According to an embodiment herein, a new ECC method was presented. The model and prior are within the EM framework which maximizes the conditional expectation of the underlying tissue process given the acquired data; using a sparse set of samples and an MRF prior capturing orientation information to discount the PV effect in fecal tagged stool with poor cathartic preparation.

In the following detailed description of the embodiments of the invention, reference is made to the accompanying drawings that form a part hereof, and in which are shown by way of illustration specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present invention. The following detailed description is, therefore, not to be taken in a limiting sense, and the scope of the present invention is defined only by the appended claims.

,CLAIMS:
CLAIMS

We Claim:

1. A method for Electronic Colon Cleansing (ECC) in CT Colonography (CTC) with a Unified Bayesian model, comprising:
acquiring a colonoscopy scan image data associated with at least one of a scan orientation, location of stool around air filled pockets, one or more tissue classes, inhomogeneity and enforcement of inconsistency of the one or more tissue classes;
identifying missing data by using an Expectation Maximization (EM) method based on the acquired colonoscopy scan data; and
formulating the electronic cleansing in an expectation maximization framework;
wherein the EM is executed using a sparse variant instead of using whole dataset to generate the one or more tissue classes, where the EM is used only in the one or more voxels in the vicinity of the boundaries and folds, while avoiding voxels which are at least one of tagged stool, soft tissues and air pockets.

2. The method as claimed in claim 1, wherein the EM method comprises:
obtaining values associated with a discrete observation of continuous uncleansed CT intensities;
obtaining a set of unobservable variables representing partial volume contributions;
performing a hard segmentation on the acquired colonoscopy scan image for labeling of tissues so as to maximize posterior probability;
calculating a voxel-wise probabilities for each of the tissue classes; and
identifying a spatial interaction in voxel neighborhood based on Markov random field model.

3. The method as claimed in claim 2, wherein the voxel-wise probabilities for each of the tissue classes is calculated based on a Gaussian model using mean and variance in the calculated tissue classes.

4. The method as claimed in claim 1, where in the one or more tissue classes comprises air bubbles, soft tissue and stool.

5. The method as claimed in claim 1, wherein generation of the sparse variant comprises:
computing crude segmentation of the available dataset based on the one or more tissue classes;
calculating means for the segmentation associated with the one or more tissue classes; and
computing a class variance for the segmentation associated with the one or more tissue classes.

6. The method as claimed in claim 5, wherein the stool is identified based on the classification and a preset criterion.

7. The method as claimed in claim 6, wherein the preset criterion for recognizing stool comprises a PV voxel belonging to a class with the highest mean intensity of the voxels which are at least one of tagged stool, soft tissues and air pockets on the boundary that face upwards, at least 20% exposed to air, and at 3mm of PV tissue.

Dated this the 17th day of October 2014
Signature

KEERTHI J S
Patent agent
Agent for the applicant