3D data analysis method, clinical decision support system, 3D data analysis program, and computer readable media

Abstract

To support a person who inspects 3D.SOLUTION: A three dimensional data analysis method for analyzing three dimensional data (D) acquired by a three dimensional imaging procedure in an examination area comprising at least one body part of a patient comprises receiving the three dimensional data (D), inputting the three dimensional data (D) into an artificial neural network (8) to generate a three dimensional model (M) of the body part and a classifier (C) indicating at least one anomaly in the three dimensional model (M) of the body part by performing an inference operation based on the three dimensional data (D), and determining the three dimensional model (M) of the body part from the three dimensional data (D), and outputting the three dimensional model (M) together with the classifier (C) at a user interface (10).SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a 3D data analysis method for analyzing 3D data acquired by a 3D imaging procedure in an examination area comprising at least one body part of a patient's body, and further to a computer-based clinical decision support system (CDSS) for carrying out the 3D data analysis method.

[0002] Furthermore, the present invention relates to a 3D data analysis program and a computer readable medium. [Background technology]

[0003] 3D imaging procedures, such as computed tomography (CT) or magnetic resonance imaging (MRI), generate large amounts of image data or data from which images can be derived. This data is manually examined by a radiologist or surgeon. The data can then be processed to generate a 3D model. In this process, specific body structures relevant to the surgical procedure are typically considered. In other words, a 3D model is rendered for a specific part of the body that will be the subject of a subsequent surgical procedure, such as an internal organ such as the liver, kidney, or lung. The 3D data contains a large amount of information. When reviewing the data, the radiologist or surgeon focuses on a single pathology or a single body part. This is done to efficiently analyze the data. However, the need to focus on relevant parts of the data can lead to situations where abnormalities in parts of the data that are currently thought to be irrelevant go unrecognized.

[0004] A similar situation can occur when generating 3D models. In an attempt to focus on the most relevant information, 3D models are typically rendered only for body parts that are deemed to be directly relevant to, for example, a surgical procedure. For example, if a surgical procedure is planned for the liver or kidney, only a portion of the 3D data imaging this organ is rendered on the 3D model. The 3D model can later be used to assist the surgeon in performing the surgical procedure on this particular internal organ. Because of this approach, a large amount of additional information in the 3D data is not analyzed or even considered. However, the additional information in the 3D data may be relevant or useful to the surgical procedure. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] U.S. Patent No. 8,532,359 Summary of the Invention [Problem to be solved by the invention]

[0006] An object of the present invention is to provide a 3D data analysis method, a clinical decision support system, a 3D data analysis program, and a computer-readable medium that can assist a person examining 3D data. [Means for solving the problem]

[0007] The object is to provide a 3D data analysis method for analyzing 3D data acquired by a 3D image generation procedure in an examination area comprising at least one body part of a patient's body, the method comprising: receiving 3D data; determining a 3D model of the body part from the 3D data by inputting the 3D data into an artificial intelligence model and performing an inference operation based on the 3D data to generate a 3D model of the body part and a classifier that indicates at least one abnormality in the 3D model of the body part; outputting the 3D model together with the classifier in a user interface; This is solved by 3D data analysis methods including

[0008] The 3D data is acquired, in particular, by magnetic resonance imaging (MRI) and / or computed tomography (CT) by scanning the patient's body in the examination area. A 3D data analysis method according to an embodiment of the present invention can include acquiring the 3D data. In another embodiment, acquiring the 3D data does not form part of a 3D data analysis method according to an embodiment of the present invention. The 3D data analysis method can be a computer-implemented method.

[0009] The 3D data analysis method according to an embodiment of the present invention can be applied to assist those examining the 3D data, such as radiologists or surgeons. Information obtained from the 3D data analysis method can be used or applied in preparation for a surgical procedure. Furthermore, information obtained from the 3D data analysis method can be used or applied during a surgical procedure. The 3D data analysis method can include performing a surgical procedure. In another embodiment, the 3D data analysis method according to an embodiment of the present invention does not include performing a surgical procedure. The surgical procedure can be performed subsequent to the method of analyzing the 3D data, i.e., after the completion of the 3D data analysis method. The classifier can assist those examining the 3D data. As mentioned above, radiologists or surgeons traditionally focus on a specific body structure or body part relevant to the surgical procedure, which may be a patient's internal organ. The use of a 3D model limited to a single organ emphasizes this highly focused approach and also increases the risk of overlooking other findings. Early in the development of technology, i.e., without the application of single-organ 3D models, radiologists would at least superficially examine all image data, which is a time-consuming task. However, in view of the large amount of data acquired, for example, in computed tomography (CT) or magnetic resonance imaging (MRI), these traditional approaches inevitably lead to a situation where information that may be relevant to the surgical procedure goes unexamined.

[0010] In contrast, the 3D data analysis method according to an embodiment of the present invention allows the entire data set to be analyzed without spending an excessive amount of time on this analysis. The surgeon or radiologist can work as usual and focus on the portion of the data set that is relevant to the subsequent surgical procedure. The remaining data can be analyzed in an automated or assisted process by applying the method according to an embodiment of the present invention to this portion of the data. According to one embodiment of the present invention, the 3D data analysis method is applied to the entire data set. Based on the classifier, the person examining the 3D data can decide to take additional information into account. The classifier provides a sound basis for decision-making, helping, for example, to focus on more distant structures that would be overlooked if one focused only on the area or body part of interest.

[0011] According to an advantageous embodiment, the 3D data analysis method is characterized in that the artificial intelligence model is a neural network trained on 3D training data, and the 3D training data meets the following criteria: a) the 3D training data describes a standard body structure model of at least one body part; b) the 3D training data describes a standard body structure model of at least one body part whose deviation from the standard body structure is less than a predetermined deviation threshold; At least one of the following applies: It is enhanced in that the degree of abnormality indicated by the classifier is a function of the deviation that occurs when a 3D model of the body part is rendered against the 3D training data.

[0012] The 3D training data can be obtained from literature or from an appropriate dataset reflecting a standard anatomical structure. The 3D training data can also be obtained from actual clinical findings or examinations performed during a 3D imaging procedure. Only data containing clinical findings considered clinically insignificant are used as 3D training data. In other words, the 3D training data always represent a healthy state. MRI and / or CT datasets that can be used as 3D training data show deviations from standard textbook anatomical structures, but only up to a predetermined deviation threshold. In other words, only 3D datasets showing body parts that conform to standard anatomical structures to a certain degree considered healthy are used as 3D training data. Training the model on real data also enriches the data pool representing standard anatomical structures and enhances the training of the neural network. Therefore, the data used as a reference is a standard anatomical model, supplemented with real-world anatomical data showing some anatomical variations that are common but not clinically significant. The neural network can be trained using supervised or unsupervised learning.

[0013] The artificial intelligence model can perform the rendering process to match the 3D data with a standard anatomy model or a model that represents a range of common variations. For this matching, the artificial intelligence model can first be trained to efficiently detect anomalies and identify potentially relevant findings relative to the standard anatomy. The findings are represented by a classifier. For this purpose, the artificial intelligence model can be trained with a dataset that most closely resembles a standard textbook anatomy. The artificial intelligence model, e.g., an algorithm or neural network, can be trained to flag any deviations from the training dataset. In a second step, the artificial intelligence model can be trained to use a 3D dataset obtained from real-world 3D imaging, e.g., CT and / or MRI data showing a patient's real-world healthy body parts or internal organs. This data enriches the artificial intelligence model's data pool.

[0014] This 3D data analysis method is implemented by an artificial intelligence model. a) the classifier indicates the highest priority if the 3D model of the body part is not found in the standard body structure; b) the classifier indicates a medium priority if at least one deviation of the body structure from a standard body structure is found for the 3D model and this deviation is within a predetermined deviation interval; c) the classifier indicates a low priority if deviations of the anatomical structure from the standard anatomical structure are found for the 3D model and the deviations are below the lower limit of a predetermined deviation interval. Thus, it can be further enhanced in that it is trained to perform inferential operations.

[0015] Classifiers aid or assist the person examining the 3D data in focusing on parts of the data that would otherwise be overlooked in traditional analysis. Regions or body parts that have deviations from standard body structure can be given a relevance flag, which is an example of a classifier, so that this region can receive more attention.

[0016] The highest priority elements are those not found in the standard body structure. This includes, for example, lesions and other significant body structure variations. Medium priority body structure variations are close to those found in the standard body structure. However, there are deviations within a certain deviation range, i.e., within a predetermined deviation interval. This interval includes deviations in shape from the standard body structure, such as small lesions. Furthermore, the deviation interval includes structures found in the standard body structure (i.e., no or small deviations in shape) but with deviations in position. In other words, a detected structure, e.g., an organ, cannot be found in exactly the same position as it is in the standard body structure. Low priority includes deviations in body structure that fall below the lower limit of the above-mentioned predetermined deviation interval. Therefore, deviations not covered by categories a) or b) are classified into category c).

[0017] According to yet another advantageous embodiment, the artificial intelligence model further performs said inference operation based on the 3D data to generate a classifier to further indicate a type of at least one abnormality in the 3D model of the body part. The abnormality may be, for example, a lesion. Different lesions may be classified as different types of lesions.

[0018] This embodiment can be further enhanced by prioritizing the anomalies in the anomaly list, in particular by further training an artificial intelligence model to perform inference operations based on the 3D data to generate classifiers further indicating the relevance of the anomalies, and prioritizing the anomaly list based on relevance, and more particularly by including a relevance value in the classifier for each anomaly found.

[0019] The list of deviations can help the person inspecting the 3D data to perform a prioritization of the findings. The prioritization can be performed based on a navigation classification, and the deviations can be classified using one or more of the following parameters: size of the abnormality, type of abnormality of the lesion, and / or amount of deviation of the abnormality from the standard body structure.

[0020] The 3D data analysis method can be further enhanced in that the anomaly list is displayed incrementally, one anomaly at a time. Areas of the rendered 3D model corresponding to each anomaly can be highlighted and / or enlarged. In particular, additional information characterizing each anomaly can be added by user input. To facilitate easier review, the 3D data analysis method can implement a viewer function (e.g., by a software module) that allows stepping through the deviations. This can be done before the start of the surgical procedure. The radiologist or surgeon can, for example, categorize the deviations and decide to focus on a particular deviation. The viewer function can highlight and enlarge each area for quick review. Furthermore, during this review, the person examining the 3D data can add information to the deviations by entering user content. These user comments can be added to the dataset or included in the dataset for later use, such as during the surgical procedure. This allows the dataset to contain not only direct 3D information about the body part, but also user information and content.

[0021] The 3D data analysis method can be further enhanced in that the examination area includes a first area and a second area, the first and second areas are not identical, a first body part is located in the first area and a second body part is located in the second area, the 3D data includes information about the first and second body parts, the first area and the first body part are labeled with targets and areas of the surgical procedure, a first 3D model is determined for the first body part and a second 3D model is determined for the second body part, the second 3D data of at least the second body part is input into an artificial intelligence model to generate a second 3D model of the second body part, and the first and second 3D models are output via a user interface along with a classifier indicating at least one abnormality in the second 3D data of the second body part.

[0022] According to yet another advantageous embodiment, the 3D data analysis method can be further enhanced by a step of outputting prioritization information together with the classifier, the prioritization information indicating the distance between the first body part and the second body part, in particular the prioritization information indicating the distance between the first body part and the abnormalities detected in the second body part.

[0023] Prioritization information that can be added to the data can include information regarding the distance of the discovered abnormality from the primary region of investigation, which may be the area where the surgical procedure will be performed. This information can assist the surgeon in planning or performing a subsequent surgical procedure.

[0024] Furthermore, the object is solved by a computer-based clinical decision support system (CDSS) comprising a processing unit, a user interface and a display, and configured to perform the 3D data analysis method according to one or more of the previously mentioned aspects.

[0025] The same or similar advantages as those described for 3D data analysis methods apply in the same or similar manner to CDSSs, and will not be repeated.

[0026] Furthermore, the object is also solved by a 3D data analysis program that causes a computer to execute the 3D data analysis method according to one or more of the above-mentioned aspects.

[0027] Furthermore, the object is also solved by a computer readable medium comprising instructions for a computer to cause the computer to perform a 3D data analysis method according to one or more of the previously mentioned aspects.

[0028] The 3D data analysis program and computer readable medium also provide the same or similar advantages as those described with respect to the 3D data analysis method according to the embodiments of the present invention.

[0029] Further features of the invention will become apparent from the description of the embodiments according to the invention as well as from the claims and the included drawings. The embodiments according to the invention can be realized by individual features or by a combination of several features.

[0030] The present invention is described below on the basis of exemplary embodiments, without limiting the general spirit of the invention, and explicit reference is made to the drawings for the disclosure of all details according to the invention that are not described in more detail in the text. [Brief explanation of the drawings]

[0031] [Figure 1] 1 illustrates a clinical decision support system (CDSS) configured to implement a 3D data analysis method for analyzing 3D data acquired by a 3D image processing procedure in an examination area. DETAILED DESCRIPTION OF THE INVENTION

[0032] In the drawings, elements of the same or similar type or corresponding parts are given the same reference numerals so as not to have to be introduced again.

[0033] FIG. 1 shows a schematic diagram of an exemplary computer-based clinical decision support system (CDSS) 2 (hereinafter referred to as system 2) configured to output a 3D model M of a body part, such as, for example, a 3D model M of a patient's internal organs. System 2 outputs the 3D model M together with a classifier C indicating at least one abnormality in the 3D data D or 3D model M of the body part. The 3D data D is input to system 2 via an input interface 4. The 3D data D is generated by a 3D image generation procedure, for example, computed tomography (CT) and / or magnetic resonance imaging (MRI). For this purpose, system 2 may include a 3D image generation device 6, for example, a computed tomography or MRI device.

[0034] In various embodiments, the system 2 includes an input interface 4 that provides 3D data D for each patient as input features to an artificial intelligence (AI) model 8 of a processor 16. The processor 16 performs inference operations by applying the 3D data D to the AI ​​model 8 to generate a 3D model M and a classifier C. This information is output via an output interface 10 to, for example, a display 20. The display 20 is a user interface of the system 2 that communicates the 3D model M and the classifier C to a user, such as a clinician. The output can serve as a basis for planning a surgical procedure. Furthermore, the output can also be used during the surgical procedure. The AI ​​model 8 receives multiple input features via the input interface 4. Each input feature is represented by image information in the 3D data D. CT or MRI data contains a vast amount of image information, meaning the AI ​​model 8 receives a large number of input features. The input features are processed by an input layer and multiple hidden layers of the AI ​​model 8. The data is processed through a neural network, which may be the AI ​​model 8, to its output layer. The output layer data is sent to the output interface 10.

[0035] In some embodiments, the input interface 4 may be a direct data link between the system 2 and one or more 3D image generating devices 6 that generate at least some of the input features representing the 3D data D. For example, the input interface 4 may directly communicate the 3D data D to the system 2 during a therapeutic and / or diagnostic medical procedure. Additionally or alternatively, the input interface 4 may be a classic user interface that facilitates interaction between a user and the system 2. For example, the input interface 4 may facilitate a user interface in which a user can manually input the 3D data D using, for example, a portable data storage device.

[0036] Additionally or alternatively, the input interface 4 may provide the system 2 with access to a database 12 that holds electronic patient records 14 from which, among other things, the 3D data D can be extracted. In either of these cases, the input interface 4 is configured to collect the 3D data D in association with a particular patient data set at or before the time the system 2 is used to generate the 3D model M and the classifier C.

[0037] Based on the 3D data D, the processor 16 holding the AI ​​model 8 performs inference operations using the AI ​​model 8 to generate a 3D model M and a classifier C. For example, the input interface 4 receiving the 3D data D can deliver CT or MRI data to the input layer of the AI ​​model 8, which propagates this image data as input features through the AI ​​model 8 to the output layer. The AI ​​model 8 can provide a computer system with the ability to perform tasks without being explicitly programmed by making inferences based on patterns seen in an analysis of the data. The AI ​​model 8 explores the research and construction of algorithms (e.g., machine learning algorithms) that can learn from existing data (which is 3D training data, described in more detail below) and make predictions about new data. Such algorithms operate by constructing the AI ​​model 8 from exemplary training data to make predictions or decisions based on the data, which is presented as an output or assessment.

[0038] There are two general aspects of machine learning (ML): supervised ML and unsupervised ML. Supervised ML uses prior knowledge (e.g., examples correlating inputs to outputs or outcomes) to learn the relationship between inputs and outputs. The goal of supervised ML is to learn a function that best approximates the relationship between training inputs and outputs given some training data, so that an ML model can implement the same relationship given an input and generate a corresponding output. Unsupervised ML is the training of an ML algorithm using unclassified or unlabeled information, allowing the algorithm to operate on that information without guidance. Unsupervised ML is useful for exploratory analysis because it can automatically identify structures in data.

[0039] The AI ​​model 8 of the system 2 is, for example, a neural network trained with 3D training data, where at least one of the following criteria applies: the 3D training data describes a standard body structure model of at least one body part; the 3D training data can be obtained, for example, from literature; the 3D training data can describe a standard body structure model of at least one body part whose deviation from the standard body structure is less than a predetermined deviation threshold; in other words, the 3D training data does not perfectly fit the textbook literature but allows for certain deviations. The 3D training data can also be obtained from real-world measurements in 3D image generation procedures such as CT or MRI. This real-world data represents healthy body parts and enriches the database or 3D training data of the AI ​​model 8 with additional variations that often occur in practice. Training is performed taking into account the 3D model M and the classifier C. The degree of abnormality indicated by the classifier C is a function of the deviations that occur when rendering the 3D model M of the body part against the 3D training data.

[0040] The training of the AI ​​model 8 is further performed in that the AI ​​model 8 is trained to perform an inference operation such that classifier C indicates the highest priority if the 3D model M of the body part is not found to be in a standard body structure. Further, during training, the AI ​​model 8 is configured such that classifier C indicates a medium priority if at least one body structure deviation from the standard body structure is found for the 3D model M and this deviation is within a predetermined deviation interval. Finally, the AI ​​model 8 is trained to set classifier C to a low priority if a deviation that does not fit either the first case or the second case is detected. The low-priority classifier C indicates a body structure deviation from the standard body structure for the 3D model M that is below the lower limit of the predetermined deviation interval.

[0041] The AI ​​model 8 can be further trained to perform inference operations based on the 3D data D, in that the classifier C further indicates the type of abnormality of the body part seen in the 3D data D. For example, the classifier C can describe the type of lesion.

[0042] Common tasks in supervised ML are classification and regression problems. Classification problems, also known as categorization problems, aim to classify items (in this case, anomalies) into one of several categorical values ​​(e.g., is this object an apple or an orange?). Regression algorithms aim to quantify some items (e.g., by providing a score for the value of some input). Some examples of commonly used supervised ML algorithms are logistic regression (LR), naive Bayes, random forest (RF), neural networks (NN), deep neural networks (DNN), matrix factorization, and support vector machines (SVM).

[0043] Some common tasks in unsupervised ML include clustering, representation learning, and density estimation. Some examples of commonly used unsupervised ML algorithms are K-means clustering, principal component analysis, and autoencoders.

[0044] Another type of ML is federated learning (also known as collaborative learning), which trains algorithms across multiple distributed devices that hold local data without exchanging data. This approach contrasts with traditional centralized machine learning techniques, in which all local datasets are uploaded to a single server, as well as more classical distributed approaches that often assume that local data samples are identically distributed. Federated learning allows multiple parties to build a common, robust machine learning model without sharing data, thus addressing important issues such as data privacy, data security, data access rights, and access to heterogeneous data.

[0045] In some examples, the AI ​​model 8 may be continuously or periodically trained prior to execution of an inference operation by the processor 16. Then, in the inference operation, the input features for each patient, i.e., the 3D data D, provided to the AI ​​model 8 may be propagated from an input layer through one or more hidden layers and finally to an output layer, which corresponds to the output of the system, which is the 3D model M and classifier C.

[0046] For example, CT data of an examination area includes 3D data of two internal organs of a patient. This 3D data D propagates through an AI model 8, which outputs 3D models M of the two internal organs along with two classifiers C, one for each 3D model M. Each classifier C indicates an anomaly in the internal organ, indicating a discrepancy with the standard body structure due to, for example, a pathology. The classifiers C can also indicate the degree of abnormality.

[0047] During and / or after the inference operation, the 3D model M and the classifier C can be communicated to a user via a user interface, such as the display 20, and / or another surgical instrument can automatically perform the desired operation. For example, a surgical procedure can be suggested by the system 2 based on the identified abnormality. The system 2 can inform the clinician of the AI-generated output for each patient, which can be a report of the suggested diagnostic and / or treatment options and the corresponding AI-generated confidence level.

[0048] The AI ​​model 8 can be further trained to identify multiple anomalies in the 3D data D of the body part. It can be further trained to generate an anomaly list consisting of the identified anomalies, which is output via the user interface 10 on the display 20 along with the 3D model M and the classifier C. In the anomaly list, the anomalies can be prioritized, for example, according to their severity. The AI ​​model 8 can be further trained to perform inference operations based on the 3D model M and generate a classifier C that further indicates the relevance of the anomalies, which may be their severity, and prioritization in the anomaly list is performed based on the relevance of the anomalies. The list can be displayed on the display 20 in stages, with one anomaly displayed at a time. Regions of the rendered 3D model M corresponding to each anomaly can be highlighted and / or enlarged. A user of the system 2 can have the option to input additional information characterizing each anomaly. Data can be entered via the user interface 18, e.g., a keyboard or mouse pointer, and can be stored for later use, e.g., during the preparation or execution of a surgical procedure.

[0049] According to another embodiment, the 3D data analysis method for analyzing the 3D data D is applicable to the following situation: an examination area analyzed, for example, by CT or MRI, includes a first area and a second area that are not identical. The first and second areas can be arranged so that they do not overlap. However, this is not always the case, even if the areas are not identical. For example, the first area can include an internal organ, such as the liver. The second area can be, for example, the area surrounding the liver. There is overlap of body structures and overlap in image data. However, the areas are not identical. A first body part is located in the first area. A second body part is located in the second area. The 3D data D includes information about the first and second body parts. The first area and the first body part are labeled with a target and area of ​​a surgical procedure. In other words, the first body part is, for example, a body part where a surgical procedure is to be performed. A first 3D model M is determined for the first body part, and this 3D model M can be used to assist a surgeon. In addition to this first model M, a second 3D model M is determined for a second body part. The second body part may, for example, be another internal organ of the patient that is further away or remote from the primary surgical field. According to a further embodiment, multiple further body parts, i.e., third, fourth, and fifth body parts, may be included in the 3D data D. Respective 3D models M are generated for these further body parts. At least one second 3D model M of the second body part is analyzed by the AI ​​model 8. Subsequently, the first 3D model M, which serves as a basis for the surgical procedure, and the at least one second 3D model M are output via the output interface 10, for example on the display 20, along with a classifier C. If more than one second body part is examined, two or more classifiers C are output. The classifier C indicates at least one abnormality in the 3D data D of the at least one second body part. This information can potentially assist the surgeon in focusing on more distant organs or body parts within the primary surgical field.

[0050] The output may be a list, along with prioritization information indicating the distance between a first body part (e.g., the subject of the surgical procedure) and at least one second body part (not the primary subject of the surgical procedure). The prioritization information may indicate the distance between the first body part and the abnormality detected in the at least one second body part.

[0051] The database 12 can function as a computer-readable medium that stores instructions (3D data analysis programs) that cause a computer, such as the processor 16 that may be implemented in the system 2, to execute a 3D data analysis method for analyzing 3D data.

[0052] All the above-mentioned features, including features that can be obtained only from the drawings, and individual features disclosed in combination with other features, are considered to be essential to the invention, whether alone or in combination. Embodiments according to the invention can be realized by individual features or by a combination of several features. Features combined with the expression "particularly" or "especially" should be treated as preferred embodiments. [Explanation of symbols]

[0053] 2 Clinical Decision Support Systems (CDSS) 4 Input Interface 6. 3D image generation device 8 AI models 10 Output Interface 12 Databases 14 Patient Records 16 Processing Unit 18 User Interface 20 Display D 3D data M 3D model C classifier

Claims

1. 1. A 3D data analysis method for analyzing 3D data acquired by a 3D imaging procedure in an examination area comprising at least one body part of a patient's body, comprising: receiving the 3D data; determining a 3D model of the body part from the 3D data by inputting the 3D data into an artificial intelligence model and performing inference operations based on the 3D data to generate a 3D model of the body part and a classifier that indicates at least one abnormality in the 3D model of the body part; outputting the 3D model together with the classifier in a user interface; A 3D data analysis method comprising:

2. The artificial intelligence model is a neural network trained on 3D training data, and the 3D training data meets the following criteria: a) the 3D training data describes a standard anatomy model of the at least one body part; b) the 3D training data describes a standard body structure model of the at least one body part that deviates from the standard body structure by less than a predetermined deviation threshold; At least one of the following applies: The method of claim 1 , wherein the degree of abnormality indicated by the classifier is a function of deviations that occur when the 3D model of the body part is rendered against the 3D training data.

3. The artificial intelligence model is a) the classifier indicates the highest priority if the 3D model of the body part is not found in a standard body structure; b) the classifier indicates a medium priority if deviations of at least one body structure from the standard body structure are found for the 3D model and the deviations are within a predetermined deviation interval; c) the classifier indicates a low priority if deviations of the 3D model's body structure from the standard body structure are found and the deviations are below the lower limit of the predetermined deviation interval. The 3D data analysis method of claim 1 , wherein the 3D data is trained to perform the inference operation as follows:

4. 2. The 3D data analysis method of claim 1, wherein the artificial intelligence model further performs the inference operation based on the 3D data to generate the classifier to further indicate a type of at least one abnormality in the 3D model of the body part.

5. 2. The 3D data analysis method of claim 1, further comprising: identifying a plurality of anomalies in the 3D model of the body part; and generating an anomaly list consisting of the identified anomalies; and outputting the anomaly list together with the 3D model via the user interface.

6. 6. The 3D data analysis method of claim 5, wherein the anomalies are prioritized in the anomaly list, and the artificial intelligence model is further trained to perform inference operations based on the 3D data to generate the classifiers further indicative of relevance of the anomalies, and the prioritization in the anomaly list is based on the relevance, and further wherein the classifier for each found anomaly includes a relevance value.

7. the list of anomalies is displayed incrementally, one anomaly at a time, and an area of ​​the 3D model corresponding to each anomaly is highlighted and / or enlarged; The 3D data analysis method of claim 5 , wherein additional information characterizing each anomaly is added by user input.

8. 2. The method of claim 1, wherein the examination area includes a first area and a second area, the first and second areas are not identical, a first body part is located in the first area, and a second body part is located in the second area, the 3D data includes information about the first and second body parts, the first area and the first body part are labeled with a target and area of ​​a surgical procedure, a first 3D model is determined for the first body part, and a second 3D model is determined for the second body part, the second 3D data of at least the second body part is input to the artificial intelligence model to generate the second 3D model of the second body part, and the first and second 3D models are output via the user interface together with the classifier indicating at least one abnormality in the second 3D data of the second body part.

9. 9. The 3D data analysis method of claim 8, further comprising outputting additional prioritization information together with the classifier, the prioritization information indicating a distance between the first body part and anomalies detected in the second body part.

10. A computer-based clinical decision support system comprising a processing unit, a user interface, and a display, the computer-based clinical decision support system configured to perform the 3D data analysis method of claim 1.

11. A 3D data analysis program that causes a computer to execute the 3D data analysis method according to claim 1.

12. A computer-readable medium containing instructions for a computer to cause the computer to perform the 3D data analysis method of claim 1.

Images