WDARMS Reznov-Jackal™

Abstract

The methods, process, and apparatus of the present invention produces a structurally representative organ model using magnetic resonance imaging scan information of a organ and a patient's medical history and physiology. This is accomplished by mathematically convolving the scan information with a second order ranked tensor matrix encoded with the patient's physiological information as it relates to the scanned organ and their medical profile. The convolved scan information and encoded matrix are computer processed to produce a 3D printer driver file which is used to print a structurally representative organ model conforming to the patient's physiology.

Background of the Invention

(1) The field of the invention is biological organ replication in a non-biological medium, also known as organ model production. Replicating a complex organ while the organ model maintains the properties of the tissue as the tissue characteristics vary in cross section while also maintaining the complex structure of the external and internal features of the original organ is accomplished with the present invention. Producing an organ model with the look and feel of the original organ provides medical professionals with an opportunity to practice surgical skills in an improved training environment. Currently, organ models are created in a manner which does not retain the look and feel of a real organ because they do not reproduce the nuances of tissue variousness throughout an organ nor reproduce complex internal and external features driven by cellular structure. Described is a novel process and a system to produce an organ model providing the nuances of tissue variousness and complex internal and external features present in the original biological organ. The source dataset, which is uniquely processed as described herein, is derived from readily available magnetic resonance imaging techniques. The end device which produces the replicated organ is an additive layer printer driven by software instructions generated by the unique software process of the present invention. The present invention combines software processing, system hardware, and network connectivity to produce an organ model, having the mechanical and the structural features of the originally scanned biological organ, for use as a novel solution to improve surgical and diagnostic training for highly skilled medical professionals and students.

Brief Summary of Invention

(2) The present invention provides hospital staff, medical school students, medical school residents, or other persons, the means to practice or refine interventional radiology surgical skills, surgical motor skills, surgical planning, surgical coordination among the various medical fields, cardiovascular interventional radiology, and neurological interventional radiology, by providing an additive layer printed cardiovascular model specific to a particular patient's physiology. The cardiovascular model maintains mechanical accuracy and cardiovascular spatial accuracy by mathematically convolving the magnetic resonance imaging (MRI) information of the scanned organ with a second order ranked tensor matrix encoded with the patient's physiological information as it relates to the scanned organ and their medical profile. In a magnetic resonance system, one or more components that manipulate, enhance or otherwise manipulate image data is an image processor. Encoding the second order ranked tensor with the patient's physiological data, the organ's cellular structure, patient age, patient health, and the patient's race is novel and is not in practice in the field of producing organ models. The result of convolving the MRI information with the encoded matrix is a dyadic product. The dyadic product is further computer processed, in a number of steps, resulting in a driver file suitable for execution on an additive layering printer using a thermoplastic filament. The system by which the organ model is produced includes a number of computer systems, large amounts of computer memory, a network to establish electrical communication amongst the computer systems as well as the additive layering printer. Though the invention described herein specifically describes a cardiovascular organ the invention may be used to produce a model of any body part, of any biological species, which may be scanned using MRI technology.

Brief Description of the Drawings

(1) The features described above, other features, aspects, and advantages of the present invention will become better understood with regard to the following description, appended claims, and accompanying drawings where:

(2) FIG. 1 is a drawing depicting the major process steps and their preferred order of execution in order to practice the present invention.

(3) FIG. 2 is a drawing showing the network structure and element interconnections to implement the preferred embodiment of the invention.

(4) FIG. 3 depicts the functional configuration of the elements to practice the method and apparatus of the present invention and is not meant to be taken as a literal depiction of element arrangement.

Detailed Description of the Invention

(5) The present invention provides hospital staff, medical school students, medical school residents, or other persons, the ability to produce a 3-D cardiovascular model which is both cardiovascular-mechanically accurate and cardiovascular-spatially accurate relative to an MRI scan of a patient's heart. The MRI, as a course of normal operation, produces scan information in the form of k-space data 145 set, where k-space data 145 set is an array of numerical values representing spatial frequencies corresponding to the physical characteristics of the cardiovascular organ being scanned. This cardiovascular model accuracy is attributed to the process of performing a mathematical convolution 140 of a nine-term, second order rank tensor 116 a K Space data 145 set forming a dyadic product 106. The dependent variables 128 of the nine-element, second order ranked tensor 116 are elasticity, plasticity, and fracture. The independent variables 122 of the nine-term, second order rank tensor 116 contain cellular composition information such as fibroblast cells, epithelial cells, and endothelial cells, where the fibroblast cells, epithelial cells, and endothelial cells optimize the organ's representation for both mechanical property space and the spatial property space in the resultant 3-D organ model 120. Each of the independent variables 122 within the nine-term, second order ranked tensor 116 is scaled by the tensor's rank, where the scaling 134 represents the patient's age, patient's race, and state of health in the resultant 3-D model. The values used for scaling 134 the nine-term, second order ranked tensor 116 are encoded by patient health quantifiable factors, communicated on bite stream 226, which are defined as patient age, race and state of health submitted to a vector embedding database 105 as a patient history form. The vector embedding database 105 is used to encode the digital logic of the nine-term, second order ranked tensor 116 where the preferred embodiment uses Pinecone implemented by the Python programming language. The vector database program performs “vector embeddings” onto the vector database's input, which is the patient history form. In other words, vector databases and vector embeddings 106 work together, and are distinct from one another.

(6) Specifically, vector embeddings 106 are a collection of words from a given input (i.e. the patient history form) in which these words are converted into arrays of numbers (i.e. vectors), whereby these arrays contain patterns of relationships. The combinatorics of these given arrays comprise a multi-dimensional vector projection as a means to measure similarity. Where the idea of similarity in this context is the projecting of the information found within the given patient history form and mapping those data points to a scaling of those data points, in which such a scaling is parameterized against patient age and patient ethnicity. Once the vector embeddings 106 are created, these embeddings may be stored in a database.

(7) The dyadic product 106 serves as the input of the tensor decomposition software operations 108. The tensor decomposition software 108101 on a Client computer 340. A description of Tensor Decomposition software operations 108 is a simple and general framework for extracting correlations and low-dimensional structure from tensor-embedded datasets. The preferred embodiment uses Non-Negative Matrix Factorization to implement Dimensionality Reduction in Tensor Decomposition software operations 108. The software operations 108 is available as on off the shelf product known as Alpha Tensor.

(8) The output of the convolution operation is time-acquisition optimized by tensor decomposition software operations 108. Time-acquisition optimization is defined by the initialization of a dimensionality reduction of time complexity which occurs at the time the patient health quantifiable factors are inputted into the vector embedding database 105, which is communicated on bite stream 288, this dimensionality reduction minimizes the time complexity of the invention's process as a means to produce the 3-D printed model faster than current methods and processes.

(9) Dimensionality Reduction is invoked to extract simple structure from large-scale datasets. The techniques of Dimensionality Reduction re-orders the rows and columns of large-scale datasets (i.e. high-dimensional matrices such as the K-space data set 145) to reveal their underlying structure. This is to say, Dimensionality Reduction performed by tensor decomposition “prunes-out” elements of a large-scale dataset which do not contribute to the function of the structure. Non-Negative Matrix Factorization is a technique to implement the necessary Dimensionality Reduction.

(10) The benefits of Non-Negative Matrix Factorization are sparse representation of the embedded-data (i.e. the data within the matrix which is used to generate the structure of the printed organ model is minimized within the dataset which speeds up organ production). Recombination of the voxels within the cells of the dataset being subjected to Non-Negative Matrix Factorization requires a summation of those discrete voxels within the dataset while avoiding shearing of those discrete voxels within the dataset to reconstruct the whole organ structure. Shearing is present when voxels on one plane of both a feature space and a function space remain unchanged while all other voxels of both the feature space and function space are shifted parallel to that plane. A feature space is the space containing all possible anatomic components of a 3-D printed organ model, and a function space is the space containing all possible physiologic components of a 3-D printed organ model; and discrete voxels can be embedded with whole anatomical features and do not require multiple voxels within the matrix to represent a specific feature of a given organ. The anatomical features are those features contained in the vector embedding 105.

(11) The Host Computer 330 collects the K-space dataset 145 via the MRI console interface 332. The Client computer 340 performs a phase encoding step where element k in the space of encoding steps P is obtained from the MRI console interface 332. For each key/of an arbitrary element k where: l∈{elasticity, plasticity, fracture is associated with the paired key m where: m∈{x, y, z} The Client Computer 340 then interprets the given patient history obtained from the patient history form 134 which is used as input to the supervised learning gradient methods also using respective values of the paired key m, as defined above. The Client Computer 340 projects this embedded data onto the feature space given by: l∈{elasticity, plasticity, fracture} and then interpolates the embedded data onto the function space given by: P∈{k.sub.1, . . . , k.sub.n} The result of convolving 140 the vector embedded second ranked tensor 116 with K-space dataset 145 results in a transformation 106 leading to an interpreted voxel volume 118 representing the mechanical, structural, and cellular composition of the originally scanned organ.

(12) A classification of the problem concerning large-scale datasets that do not undergo Dimensionality Reduction is identified as a Co-Clustering, a problem which should be avoided.

(13) The described process of generating the 3-D organ model retains greater structural representation, in comparison to currently produced methods and processes, due to the use of fibroblast cells, epithelial cells, and endothelial cells as the independent variables 122 in the nine-term, second order ranked tensor 116; elasticity, plasticity, and fracture for the dependent variables 128 in the nine-term, second order ranked tensor 116; patient age, and patient ethnicity for the rank in the nine-term, second order ranked tensor 116. The dimensionality reduction is defined as the process of transforming data from an initial dimensional space and then reducing this data to a lower dimensional space. The vector embedded nine-term, second order ranked tensor 116 which is an input to the convolution operation 140 captures the deformation of continuous mediums wherein the deformation of continuous mediums is the voxel value parameterization to conserve the mass of a patient's heart that is being printed on a 3-D printer hotbed plate; deformation of continuous mediums is to perform structural transformations of an object (i.e., 3-D models) using mediums (i.e., filaments), in which these mediums, when being used in lieu of the structural transformation, conserves the mass of the object, where such continuous mediums are transformed into physically continuous 3-D printed models using continuous 3-D printer filaments 230, such as continuous polymer-based filaments, continuous metallic-based filaments, or continuous organic-materials-based filaments (identified on bite stream 232) of the 3-D printer 114. The output of the tensor decomposition software operations 108, serves as an input for image and surface registration techniques 110, where the registration techniques are processed using Gradient Estimation Volume Rendering, which may be performed by parallel processing hardware 346 such as a NVIDIA GPU under network control 287. Supervising learning gradients is a powerful technique used to achieve high accuracy in implementing the image and surface processing tasks. Gradient information is used to detect features in images, such as edges, corners, and blobs related to the original organ. Gradient information can be used to analyze the texture of an image, such as its smoothness, coarseness, and homogeneity related to the original organ. Another image and surface registration technique uses transformation manifolds. A transformation manifold is used to perform surface reconstruction using a set of images or surfaces that are related to each other by a smooth transformation; a smooth transformation is a geometric summation of each computational output of a transformation matrix and a phase encoding step of k-space, for all phase encoding steps within the k-space. This means that the images or surfaces can be smoothly deformed into each other without any sharp discontinuities as part of the processing to produce the organ model. Merging resources is used as one image and surface registration technique and results in merging multiple sets of points or measurements of a surface can create a more accurate and complete 3D model of the organ model's surface.

Brief Description of the Drawings

(2) FIG. 1 is a drawing depicting the major process steps and their preferred order of execution in order to practice the present invention.

(3) FIG. 2 is a drawing showing the network structure and element interconnections to implement the preferred embodiment of the invention.

Detailed Description of the Invention

(5) The present invention provides hospital staff, medical school students, medical school residents, or other persons, the ability to produce a 3-D cardiovascular model which is both cardiovascular-mechanically accurate and cardiovascular-spatially accurate relative to an MRI scan of a patient's heart. The MRI, as a course of normal operation, produces scan information in the form of k-space data 145 set, where k-space data 145 set is an array of numerical values representing spatial frequencies corresponding to the physical characteristics of the cardiovascular organ being scanned. This cardiovascular model accuracy is attributed to the process of performing a mathematical convolution 140 of a nine-term, second order rank tensor 116 a K Space data 145 set forming a dyadic product 106. The dependent variables 128 of the nine-element, second order ranked tensor 116 are elasticity, plasticity, and fracture. The independent variables 122 of the nine-term, second order rank tensor 116 contain cellular composition information such as fibroblast cells, epithelial cells, and endothelial cells, where the fibroblast cells, epithelial cells, and endothelial cells optimize the organ's representation for both mechanical property space and the spatial property space in the resultant 3-D organ model 120. Each of the independent variables 122 within the nine-term, second order ranked tensor 116 is scaled by the tensor's rank, where the scaling 134 represents the patient's age, patient's race, and state of health in the resultant 3-D model. The values used for scaling 134 the nine-term, second order ranked tensor 116 are encoded by patient health quantifiable factors, communicated on bite stream 226, which are defined as patient age, race and state of health submitted to a vector embedding database 105 as a patient history form. The vector embedding database 105 is used to encode the digital logic of the nine-term, second order ranked tensor 116 where the preferred embodiment uses Pinecone implemented by the Python programming language. The vector database program performs “vector embeddings” onto the vector database's input, which is the patient history form. In other words, vector databases and vector embeddings 106 work together, and are distinct from one another.

(7) The dyadic product 106 serves as the input of the tensor decomposition software operations 108. The tensor decomposition software 108 is hosted on software processing platforms, the preferred embodiment uses AlphaTensor to perform tensor decomposition, which initializes 101 on a Client computer 340. A description of Tensor Decomposition software operations 108 is a simple and general framework for extracting correlations and low-dimensional structure from tensor-embedded datasets. The preferred embodiment uses Non-Negative Matrix Factorization to implement Dimensionality Reduction in Tensor Decomposition software operations 108. The software operations 108 is available as on off the shelf product known as Alpha Tensor.

(8) The output of the convolution operation is time-acquisition optimized by tensor decomposition software operations 108. Time-acquisition optimization is defined by the initialization of a dimensionality reduction of time complexity which occurs at the time the patient health quantifiable factors are inputted into the vector embedding database 105, which is communicated on bite stream 288, this dimensionality reduction minimizes the time complexity of the invention's process as a means to produce the 3-D printed model faster than current methods and processes.

(9) Dimensionality Reduction is invoked to extract simple structure from large-scale datasets. The techniques of Dimensionality Reduction re-orders the rows and columns of large-scale datasets (i.e. high-dimensional matrices such as the K-space data set 145) to reveal their underlying structure. This is to say, Dimensionality Reduction performed by tensor decomposition “prunes-out” elements of a large-scale dataset which do not contribute to the function of the structure. Non-Negative Matrix Factorization is a technique to implement the necessary Dimensionality Reduction.

(11) The Host Computer 330 collects the K-space dataset 145 via the MRI console interface 332. The Client computer 340 performs a phase encoding step where element k in the space of encoding steps P is obtained from the MRI console interface 332. For each key/of an arbitrary element k where: l∈{elasticity, plasticity, fracture} is associated with the paired key m where: m∈{x, y, z} The Client Computer 340 then interprets the given patient history obtained from the patient history form 134 which is used as input to the supervised learning gradient methods also using respective values of the paired key m, as defined above. The Client Computer 340 projects this embedded data onto the feature space given by: l∈{elasticity, plasticity, fracture} and then interpolates the embedded data onto the function space given by: P∈{k.sub.1, . . . , k.sub.n} The result of convolving 140 the vector embedded second ranked tensor 116 with K-space dataset 145 results in a transformation 106 leading to an interpreted voxel volume 118 representing the mechanical, structural, and cellular composition of the originally scanned organ.