DESC:FIELD OF THE INVENTION:
This invention relates to the field of medical engineering.

Particularly, this invention relates to a system and method for pre-surgical planning using non-invasive imaging techniques.

BACKGROUND OF INVENTION:
Recent advances in information technology and bio-medicine have prompted Computer Aided Design (CAD) to find many novel applications in bio-medical engineering. An integration of CAD and medical technology is referred to as bio-CAD. Three-dimensional (3D) bio-CAD model reconstruction from CT medical image has recently become the issue of much attention. It is particularly important in bio-medical engineering since CAD with the help of medical imaging and freeform-fabrication technologies like Reverse Engineering (RE) and Rapid Prototyping (RP) has the capacity to create anatomic models which have diagnostic, therapeutic and rehabilitatory medical applications. Bio-CAD is widely used in many applications such as Computer-Aided Surgery, Structural Modeling of Tissue, Design of Orthopedic Device, Implants, Tissue Scaffolds and Freeform Fabrication or Bio-Manufacturing. These models also have non-medical applications in the field of passenger safety design and crash analysis. These can also be used to fabricate prostheses, to perform various simulation and analytical tasks. Number of open source and commercial products for 3D biomechanical construction are available, but still, these tools does not appear to be a simple, accurate and reliable for bio-image acquisition and analysis.

The reverse engineering procedure consists of measure steps viz. Point Data capture, pre-processing of Point Cloud data (mainly for noise reduction, smoothening, filtering, etc.), segmentation and CAD model creation.In general, human body models are obtained by a Reverse Engineering process based on non-invasive techniques (i.e. a medical procedure is strictly defined as non-invasive when no break in the skin) viz. Computer Tomography (CT), Magnetic Resonance Imaging (MRI), etc. A CT medical image is limited by its 2D image presentation in that it does not allow doctors to quickly diagnose illness and explain treatments to patients. CT scan images are in Digital Imaging and Communications in Medicines (DICOM format). Medical models in 3D solid models are therefore very important in the diagnosis and treatment process. In last few years some commercial programs were developed and used to construct a CAD based model from medical images. However, none of these programs has been widely accepted due to inherent complexity of human body’s anatomical structures.

Hence, more effective methods for conversion of CT data into CAD models still need to be developed. Although non-invasive medical imaging is able to produce three dimensional descriptions of the internal body details, the capture of internal detail of the body is essentially a voxel based representation. Voxel based representation suffers from problems viz. aliasing, loss of geometric information after voxelization, huge memory requirements, and inconvenient model editing. Implicit and explicit functions, though easy to manipulate, are unable to address geometrical and material complexity of a bio-model. The B-splines are known to represent the freeform objects closely. The control point based model is a direct extension of parametric curves, surfaces and volume where, along with geometrical information, addition material information is stored at control points. Both these pieces of information can easily be controlled by changing control points. Hence the B-spline based control point method is selected for preparing human body models. The watertight 3D surface model will be generated by using B-spline interpolation scheme from these point cloud data. In the method, a base surface is to be generated by creating a smooth implicit surface from point cloud data. This methodology has the potential to represent internal body details accurately with fewer digital input data, with applications in FEM analysis, freeform-fabrication and tissue engineering.

For reconstruction of human bone from medical images, either the volume based approach or contour based approach is most popularly used. In majority of commercially available Bio-medical software, to interpret CT image data, volume pixels i.e. voxel representation is most commonly used. This volumetric model is represented by brick-like components, each representing a set height, width, and depth. The voxel data from medical image data can be directly transferred to stl format and printed on a rapid prototyping system. It is attractive in visualizing human anatomies since it can it generates high resolution surfaces and efficient in terms of computation and data structure. Voxel based representation suffers from problems like aliasing, loss of geometric information, huge memory requirements and inconvenient model editing. Also, triangular patched model from the marching cube is difficult for manufacturing and engineering applications without pre-processing. These disadvantages of voxel models can be overcome by converting it into a CAD-based solid model, which relies upon ‘Boundary Representation’ techniques. In this method a solid model is defined by the boundaries that enclose it. These bounding surfaces are mathematically described by polynomial functions such as B-Spline curves, Non-Uniform Rational B-Spline (NURBs) functions, etc. It offers additional advantage of achieving Tangential (C1) and Curvature (C2) continuity over stl data, which offers only Positional (C0) continuity. These CAD based methods offers the construction of the model by minimizing the size of the files and ensuring the closure of bounding surfaces. In addition to being a closer approximation to most 3D structures, the boundary represented CAD model is capable of being altered through Boolean operations and analyzed by enabling CAD and analysis software packages.

PRIOR ART:
For rapid product development in both research and industrial area, Reverse Engineering and Rapid Prototyping have received extensive attention in recent time. Reverse Engineering is an important method for reconstructing the Computer Aided Design (CAD) model from a physical part which is in existence. The process starts from digitizing the existing part, i.e., capturing a point cloud data. The measured points are then transformed into a compact CAD model using fitting or interpolation techniques. The reverse engineering procedure adopted for current research work consists of the following basic phases:
1. Generation of Point Cloud Data from CT scan (DICOM)
2. Segmentation of Point Cloud Data
2.1 Region Based Segmentation
2.2 Edge based segmentation
2.3 Combination of Edge based and Region based segmentation
3. Construction of water-tight surface
3.1 Curve generation
3.2 Surface generation
3.3 Construction of Water-tight Surface Model
3.4 Conversion of Surface model to Solid model
4. Testing of Solid Model for its validity
5. Physical Model construction through Rapid Prototyping
6. Experimental Validation of Results

During previous reports, exhaustive literature survey based on PCD generation, Segmentation techniques is covered. Based on literature survey, following limitations were identified in Point Cloud Data Segmentation
a. A major problem with surface based segmentation was the presence of noise in point clouds. In a scan, surfaces are supposed to be infinitely thin, but noise has the effect of making surfaces thick. This affects the accurate estimation of surface properties in region growing tests (Leonardis, 1993) often resulting in over segmentation.
b. In edge based segmentation edges become more difficult to detect and this leads to under segmentation. To resolve the noise problem, hybrid methods which combine edge detection and surface based segmentation algorithms have been proposed (Yokoya and Levine, 1997; Checchin et al, 1997).
c. Surface based segmentation algorithms described surfaces using explicit surface functions (cylinders, planes, surface patches). This complicates the segmentation of implicit surfaces. Because of this algorithms are designed for specific scenes, for example, Rabbani et al, 2006, for industrial installations. As a result, there are very few generic segmentation algorithms.
d. Many segmentation algorithms perform neighborhood searches. In the absence of a space partitioning scheme this can lead to unacceptable computational overheads, particularly for very large point clouds.

Recent advances in information technology and bio-medicine have prompted Computer Aided Design (CAD) to find many novel applications in biomedical engineering. CAD with the help of medical imaging and Rapid Prototyping (RP) technologies has the capacity to create anatomic models which have diagnostic, therapeutic, and rehabilitatory medical applications. These models also have non-medical applications in the field of passenger safety design and crash analysis. Generally, human body models are prepared by reverse engineering process using non-invasive imaging techniques like CT/MRI. These models are used for creating biomedical implants and tissue scaffolds or as a surgical/diagnostic aid. Usually, a 3D bio-CAD model is reconstructed through either Segmentation or Volumetric Representation.

Various attempts were made to define material heterogeneity on geometric models, especially for non-medical applications. Depending on the material specification modalities, such schemes can be classified into four categories viz. Voxel Based, Explicit Function Based, Implicit Function Based and Control Point Based Modeling Methods.
a. In the voxel based method, the entire model space is discretized into small elemental volume blocks called voxels. Geometrical coordinates, as well as material composition, are specified at the voxel.
b. In an explicit function based method, analytical functions are used to define material heterogeneity. Grading source is defined as the `origin of material variation'. Material composition functions are linear or nonlinear real-valued mathematical functions. They are defined in term of the distance from the `grading source'. As the mathematical functions like linear, exponential, parabolic and power functions are used for material distribution with explicit functions, the complex heterogeneities are difficult to express by such explicit functions.
c. While the explicit function based method employs boundary representation (B-rep) for geometry and explicit functions for material representations, the implicit function based method uses functional representation (F-rep) for both geometry and material representation. The implicit functions are used for defining material heterogeneity. Adzheiv utilized F-rep to model heterogeneity of geological structures and adaptive finite element mesh generation. Biswas et al. used R-function with distance field, where the inverse distance weighing function with transfinite interpolation was employed with a multiple material feature. Model coverage is poor for this method and, as such, it is difficult to find an implicit function for a domain enclosed by NURBS or spline curves.
d. In the control point based method, geometry-material values are specified at control points and interpolated by shape functions. The shape functions usually used are B-spline, Bezier and NURBS. Due to the large number of control points, this method has excellent model coverage but a high degree of freedom. On standardization of heterogeneous object representation, Lalit et al. suggested an information model to represent heterogeneous object for ISO 10303.

In the prior art, there are three methods that can be applied to reconstruct a 3D solid model for bio-medical imaging from its 2D CT image:
a. The first method is via voxel to stack and construct the model by using the marching-cube algorithm
Limitation:
i. The voxel stacking technique may make holes, aliases and saw-toothed paths within the voxel connection.
ii. Requires huge memory storage and lacks geometric topological relations. Hence, they are unsuitable for bio-medical analysis and simulation area which demands a vector based CAD solid modelling environment.
iii. Moreover, generating the complete solid model without defects such as holes and overlapping is one of the most time-consuming processes requiring a considerable amount of manual intervention.

Figure 1 illustrates Voxel data by Marching Cube Algorithm

b. With the second method, contour detection in each layer is used to construct the mixed layers in the triangular STL model for RP fabrication by connecting the vertices of two parallel polygons
Limitation:
i. This method suffers from drawbacks in the contour detection for each CT layer
ii. File errors in the construction and use of STL meshes.
iii. branching and correspondence problems often arise

Figure 2 (a) illustrates contours of point cloud data, (b) STL meshes, (c) RP Model

c. The third method involves a swept blend from the contours of each layer in point data
Limitation:
i. The curves swept blend method is extremely complicated and it cannot be applied without drawing the curve model, since the spline must first be constructed before modelling can take place. [Wang]
ii. they are unsuitable for bio-medical analysis and simulation area which demands a vector based CAD solid modelling environment
iii. 3D surface reconstruction from the segmented contours is extremely complicated in data organization and handling, since it cannot be applied without constructing the curves in each CT slice in advance.

Figure 3 illustrates a3D model reconstruction using swept blend curve function of Skull and Femur bone.

Figure 4 illustrates B-spline surfaces.

OBJECTS OF THE INVENTION:
An object of the invention is to provide a relatively more effective system and method for conversion of CT data into CAD models.

Another object of the invention is to design and develop Graphical User Interface (GUI) for conversion of non-invasive CT raw scan data of patients in DICOM format.

Another object of the invention is to provide a system and method which comprises a single step converter without much intermediate operations using save us from performing tedious manual operations.

Yet another object of the invention is to develop an Efficient Surface Reconstruction method which enables to create accurate CAD model from scanned point cloud data in consistent framework.

Still another object of the invention is to construct a water-tight surface / solid model.

An additional object of the invention is to provide a system and method which export files in *.stl format which will enable to construct rapid prototype model.

Yet an additional object of the invention is to print 3D CAD model by using 3D printing techniques to help surgeon to visualize bone structure and deformity in spinal structure.

SUMMARY OF THE INVENTION:
According to this invention, there is provided a system for pre-surgical planning using non-invasive imaging techniques, said system comprises:
- an input mechanism configured to receive CT scan images;
- a computation engine configured to compute HU Values from an input CT scan image;
- a selection mechanism communicably coupled with said computation engine configured to select appropriate HU values based on type of data required, selection criteria being based on statistical data and empirical data specific to a patient correlating to said CT scan images;
- an estimation computation engine configured to estimate point cloud data based on a threshold technique, said data comprising dimensional items in all axes along with a density value;
- a pre-processing computation engine configured for pre-processing point cloud data, for data smoothening;
- a quality computation mechanism configured to check for quality of point cloud data by identifying deviation with respect to pre-defined threshold values;
- a segmentation engine configured to segment point cloud data using a segmentation technique in order to classify said point cloud data into geometric primitive shapes based on normal estimation of neighbouring points;
- a construction computation engine uses point cloud data to construct a surface;
- a surface quality check computation configured to conduct a surface quality check of said constructed surface in order to check if said surface is watertight;
- a repair mechanism configured to repair a surface based on output of said surface quality check computation engine in order to successfully construct a solid model;
- a conversion engine converts surface data into a neutral format for constructing a solid model; and
- a verification engine configured to verify said solid model generated by comparing with said point cloud data.

Typically, said threshold technique being selected from at least a local thresholding technique or at least a global thresholding technique, characterised in that, said local thresholding technique being correlative to a single image being analysed and applying rules to a current image only.

Typically, said threshold technique being selected from at least a local thresholding technique or at least a global thresholding technique, characterised in that, said global thresholding technique being correlative to a series of images per patient and applying rules to all images.

Typically, said data from said estimation computation engine is stored in a storage device comprising 4 dimensions, said dimensions being: a) x-direction dimension; b) y-direction dimension; c) z-direction dimension; and d) density value dimension.

Typically, said data from said estimation computation engine is stored in a storage device comprising 4 dimensions, said dimensions being: a) x-direction dimension; b) y-direction dimension; c) z-direction dimension; and d) density value dimension, characterised in that, said x-direction dimension comprising image resolution data, said y-direction dimension comprising image resolution data, z-direction dimension comprising image resolution slice thickness data.

Typically, said pre-processing computation engine being configured to de-noising by setting distance to its ‘K’ nearest neighbours wherein distances of all neighbouring points are calculated and threshold distance set to one standard deviation from the mean of average distance to neighbors of all points.

DETAILED DESCRIPTION OF THE ACCOMPANYING DRAWINGS:
According to this invention, there is provided a system and method for pre-surgical planning using non-invasive imaging techniques. Typically, this is a rapid prototyping machine.

Figure 5 illustrates Conversion of CT scan Data to STL File

In accordance with an embodiment of this invention, the machine comprises an input mechanism configured to receive CT images.

In accordance with another embodiment of this invention, the machine further comprises a computation engine configured to compute HU Values from an input CT scan image. In CT scan, Hounsfield Unit (HU) is proportional to the degree of x-ray attenuation by the tissue. A selection mechanism communicably coupled with the computation engine is configured to select proper HU values based on type of data required. Selection criteria based on statistical data and empirical data is altered to be patient specific. HU values are based on tissue density and vary with person to person for bone density based on health conditions. A histogram of HU values can be displayed along with dynamic display of field of interest. Now, a user can pick appropriate HU value looking at display.
Step 1: Selection of proper HU Values based on type of data required

In accordance with yet another embodiment of this invention, an estimation computation engine is configured to estimate point cloud data based on a proper threshold algorithm. This is Step 2.

Threshold Algorithms are classified in LOCAL and GLOBAL thresholding algorithms. CT scan images are processed in batches. If, one considers single image analyse it and after statistical analysis applies rules to current image only, then it is called as LOCAL thresholding. If applied for whole series, then it is called as Global thresholding. e.g. Threshold Value: 226 HU for bone, -800 HU for skin, etc.

The data is stored as a 4 dimensional array with dimensions in x, y, z direction and 4th dimension as density value. The x and y dimensions are based on image resolution and z value is “Slice thickness” in mm. The accuracy as well as smoothness depends upon x, y and z values. Lower the values, better is the surface fitting. To have lower values of x, y, and z, the 4D array can be resampled with smaller value. This can be achieved by “Linear”, “Cubic”, or “Spline” interpolation. For linear interpolation, C1 continuity, faster with lower memory requirement can be achieved. With Cubic or spline interpolation, higher level continuity i.e. “C2”, but slower and higher memory requirement.

In accordance with still another embodiment of this invention, a pre-processing computation engine is configured for pre-processing point cloud data for removal of outliers or noise, for data smoothening, and the like. This is Step 3.

The output data of previous step is millions of points based on data we are handling. It contents outliers i.e. the points of body, which are not actual bone but predicted as bone. (E.g. Shirt buttons. This can be manually removed as well as by using software.) These require huge memory space and a large amount of time consumed in computations.

The Step 3 further includes a step of denoising i.e. removal of outliers, as follows:
• This step s achieved by setting distance to its ‘K’ nearest neighbours. The distances of all neighbouring points are calculated and threshold distance set to one standard deviation from the mean of average distance to neighbors of all points. A point is considered to be an outlier if the average distance to its K nearest neighbors is above a threshold.

The Step 3 further includes a step of Data down sampling, as follows:
• Objective is to reduce no. of data pointsto increase speed
• Part of point cloud data segmentation, where data points can be reduced randomly or based on geometric features viz. corners, edges, etc.

Figures 6a and 6b illustrate results of the two sub-steps of Step 3for various bone point cloud data.

In accordance with an additional embodiment of this invention, a quality computation mechanism checks for quality of point cloud data. This is Step 4.

Figure 7 illustrates steps in detail which result in output of step 4.
1. Point Cloud Data and final surface model imported in the system of this invention;
2. Direct registration is performed;
3. Surface to Point Cloud Difference, deviation is found out;
4. If deviation is in within limits, part is Okay.
Basically, distance is the normal distance from point to the surface.

In accordance with yet an additional embodiment of this invention, a segmentation engine is configured to segment point cloud data using an edge / surface / hybrid segmentation algorithm. This is Step 5. The output of this aids in classification of point cloud data, in geometric primitive shapes like planes, cylinders, etc. Also, it classifies points on edges, etc. based on normal estimation of neighbouring points. This step is applied in data downsampling.

In accordance with still an additional embodiment of this invention, a construction computation engine uses point cloud data to construct a surface. This is Step 6.

In accordance with another additional embodiment of this invention, a surface quality check computation engine is configured to conduct a surface quality check of the constructed surface. This checks if the surface is watertight. This is Step 7.

In accordance with yet another embodiment of this invention, a repair mechanism is configured to repair a surface based on output of the surface quality check computation engine. This is Step 8. For successful construction of solid models, surface must be water-tight. This can be verified by using following formula>>>
The Euler's formula or Euler-Poincare is: F – E + V – L = 2(B – G)
where, F - number of faces,
E - number of edges,
V - number of vertices,
L - number of faces inner loops,
B - number of bodies and
G - number of genus.
The simplest version of this equation is: F – E + V = 2

Figure 8 illustrates some examples of watertight surfaces.

Figure 9 illustrates an example which is implemented using the system and method of this invention.

In accordance with still another embodiment of this invention, a conversion engine converts surface data into a CAD neutral format viz. stl, iges/igs, step, etc. This is Step 9.

In accordance with an additional embodiment of this invention, a rapid prototype of the model can be made using data from step 9. This is Step 10.

In accordance with yet an additional embodiment of this invention, a verification engine conducts verification of part generated by comparing with PCD generated. This is Step 11.

Applications –
a. Pre-surgical models for applications for patient, students training, etc.
b. Construct bone scaffold models
c. Facial reconstruction surgeries like plastic surgery by using reverse engineering approach

While this detailed description has disclosed certain specific embodiments for illustrative purposes, various modifications will be apparent to those skilled in the art which do not constitute departures from the spirit and scope of the invention as defined in the following claims, and it is to be distinctly understood that the foregoing descriptive matter is to be interpreted merely as illustrative of the invention and not as a limitation.
,CLAIMS:WE CLAIM,

1. A system for pre-surgical planning using non-invasive imaging techniques, said system comprising:
- an input mechanism configured to receive CT scan images;
- a computation engine configured to compute HU Values from an input CT scan image;
- a selection mechanism communicably coupled with said computation engine configured to select appropriate HU values based on type of data required, selection criteria being based on statistical data and empirical data specific to a patient correlating to said CT scan images;
- an estimation computation engine configured to estimate point cloud data based on a threshold technique, said data comprising dimensional items in all axes along with a density value;
- a pre-processing computation engine configured for pre-processing point cloud data, for data smoothening;
- a quality computation mechanism configured to check for quality of point cloud data by identifying deviation with respect to pre-defined threshold values;
- a segmentation engine configured to segment point cloud data using a segmentation technique in order to classify said point cloud data into geometric primitive shapes based on normal estimation of neighbouring points;
- a construction computation engine uses point cloud data to construct a surface;
- a surface quality check computation configured to conduct a surface quality check of said constructed surface in order to check if said surface is watertight;
- a repair mechanism configured to repair a surface based on output of said surface quality check computation engine in order to successfully construct a solid model;
- a conversion engine converts surface data into a neutral format for constructing a solid model; and
- a verification engine configured to verify said solid model generated by comparing with said point cloud data.

2. A system for pre-surgical planning using non-invasive imaging techniques as claimed in claim 1 wherein, said threshold technique being selected from at least a local thresholding technique or at least a global thresholding technique, characterised in that, said local thresholding technique being correlative to a single image being analysed and applying rules to a current image only.

3. A system for pre-surgical planning using non-invasive imaging techniques as claimed in claim 1 wherein, said threshold technique being selected from at least a local thresholding technique or at least a global thresholding technique, characterised in that, said global thresholding technique being correlative to a series of images per patient and applying rules to all images.

4. A system for pre-surgical planning using non-invasive imaging techniques as claimed in claim 1 wherein, said data from said estimation computation engine is stored in a storage device comprising 4 dimensions, said dimensions being: a) x-direction dimension; b) y-direction dimension; c) z-direction dimension; and d) density value dimension.

5. A system for pre-surgical planning using non-invasive imaging techniques as claimed in claim 1 wherein, said data from said estimation computation engine is stored in a storage device comprising 4 dimensions, said dimensions being: a) x-direction dimension; b) y-direction dimension; c) z-direction dimension; and d) density value dimension, characterised in that, said x-direction dimension comprising image resolution data, said y-direction dimension comprising image resolution data, z-direction dimension comprising image resolution slice thickness data.

6. A system for pre-surgical planning using non-invasive imaging techniques as claimed in claim 1 wherein, said pre-processing computation engine being configured to de-noising by setting distance to its ‘K’ nearest neighbours wherein distances of all neighbouring points are calculated and threshold distance set to one standard deviation from the mean of average distance to neighbors of all points.

Dated this 10th day of December, 2017

CHIRAG TANNA
of INK IDEÉ
APPLICANT’S PATENT AGENT