Method for selecting a type of implantable medical device to treat an aneurysm

Abstract

This presentation concerns a method for selecting a type of implantable medical device (IMD) from a set of IMD types, the IMD being intended to be positioned in a vascular structure including a region of interest. The method comprises the following steps: obtaining an image of a vascular structure including a region of interest; extracting a plurality of features of the vascular structure from the image obtained; processing the extracted features by a supervised learning model to obtain, for each IMD type, a compatibility probability quantifying the compatibility of the IMD type with the region of interest of the vascular structure; and selecting, from among the extracted features, and based on each compatibility probability, at least one essential feature influencing the compatibility probability associated with each IMD type. Figure for the abstract: Fig. 3

Description

Title of the invention: Method for selecting a type of implantable medical device to treat an aneurysm. Technical field

[0001] The present exposition relates to the field of treatment of a vascular structure, such as an artery, undergoing a local pathology such as an aneurysm, by implantation of an implantable medical device (or IMD).

[0002] More specifically, the presentation concerns the choice of the right type of DMI to implant in the vascular structure for which a three-dimensional image is available. STATE OF THE ART

[0003] Many vascular accidents are linked to the presence of an aneurysm in the walls of a vascular structure. Moreover, the risk of vascular accidents increases with the expansion of the aneurysm.

[0004] It is therefore necessary to prevent the expansion and rupture of the aneurysm, as well as to prevent clots formed in the aneurysm from migrating into the vascular structure and locally blocking an artery.

[0005] It is common to use an implantable medical device (IMD) for the treatment of an artery affected by an aneurysm. Various types of IMDs exist for this purpose, such as a stent, an intrasaccular cage, a flow diverter (or flow diverter, according to commonly used Anglo-Saxon terminology), or endovascular embolization using microcoils (or coils, according to commonly used Anglo-Saxon terminology). Each of these types of IMDs is available in several sizes.

[0006] Before selecting the size of the implantable MRI device, a clinician must therefore choose from the different types of MRI devices available. Indeed, each type is more or less well suited to each aneurysm.

[0007] The clinician currently has no other aid in choosing the type of DMI to treat the specific aneurysm he is facing than his own experience or the experience of his colleagues physically present to advise him. GENERAL STATEMENT

[0008] One aim of the presentation is to improve the treatment of aneurysms by advising the clinician on the choice of the type of DMI to use according to the particular characteristics of the aneurysm.

[0009] To this end, according to one aspect of the present exposition, a method for selecting a type of implantable medical device (IMD) from among a set of IMD types is proposed, the The DMI is intended to be positioned within a vascular structure containing an area of ​​interest. The procedure comprises the following steps: - obtaining an image of a vascular structure including an area of ​​interest; - extraction of a plurality of characteristics of the vascular structure from the image obtained; - processing by a supervised learning model of the extracted features to obtain, for each type of DMI, a compatibility probability quantifying the compatibility of the DMI type with the area of ​​interest of the vascular structure; and - selection from the extracted characteristics, and according to each probability of compatibility, of at least one essential characteristic influencing the probability of compatibility associated with each type of DMI.

[0010] This selection process makes it possible to improve the treatment of aneurysms by making the choice of the type of MDI to use safer and faster.

[0011] This method increases the clinician's confidence in the choice of the type of medical device proposed by the method. Indeed, the method allows the clinician to understand the reasons for choosing one type of medical device and not another.

[0012] The method makes it possible to automate the choice of the type of DMI to use by taking into account previously trained databases.

[0013] Advantageously, but optionally, the described method includes at least one of the following features, taken alone or in any combination:

[0014] - the selection classifies each type of DMI as an appropriate type, if the probability compatibility of the type of DMI is greater than a previously set threshold, or as an inappropriate type, if the probability of compatibility of the type of DMI is less than the threshold;

[0015] - the selection is implemented based on a pattern database appropriate and an inappropriate pattern database, the appropriate pattern database and the inappropriate pattern database each comprising at least one explainability pattern for each type of MDI, each explainability pattern comprising at least one influential feature and an impact level associated with the influential feature, the influential feature being a feature having a non-zero associated impact level, the impact level qualifying the impact of each feature in calculating the probability of compatibility of the type of MDI considered,

[0016] the explainability pattern of the appropriate pattern database being common to at least one reference vascular structure for which the type of DMI considered is an appropriate type, and

[0017] the explainability pattern of the inappropriate pattern database being common to at least one reference vascular structure for which the type of DMI considered is an inappropriate type;

[0018] - the selection determines for each type of DMI the essential characteristic:

[0019] - by comparing the characteristics of the vascular structure to the patterns of explainability of the database of appropriate patterns if the type of DMI considered is an appropriate type, or of explainability of the database of inappropriate patterns if the type of DMI considered is an inappropriate type; and

[0020] - retaining as essential characteristic(s) the characteristic(s) influential common to the characteristics of the vascular structure, the influential characteristic(s) being included in one or more patterns of explainability having the most influential characteristics common to the characteristics of the vascular structure;

[0021] - a generation of the model from a database of features including features of each of the images of a plurality of reference vascular structures from a training image database;

[0022] - the compatibility probabilities of each type of DMI for the structures Reference vascular values ​​are calculated by the model from the feature database;

[0023] - an initial formation of a database of appropriate transactions and a database of inappropriate transactions,

[0024] - the appropriate transaction database including for each type of DMI a transaction including the influential characteristics of a reference vascular structure for which the type of DMI is considered an appropriate type, and

[0025] - the database of inappropriate transactions comprising for each type of DMI a transaction including the influential characteristics of a reference vascular structure for which the type of DMI is considered an inappropriate type;

[0026] - each transaction is calculated based on the probability of compatibility of the type of DMI considered for each reference vascular structure and an explainability table including for each reference vascular structure the level of impact of each characteristic for the probability of compatibility of the type of DMI considered, each transaction including the influential characteristics of a reference vascular structure and the level of impact associated with each of the influential characteristics;

[0027] - a second training of the database of appropriate patterns and of the database of inappropriate reasons, each including for each type of DMI at least one explainability reason, such as:

[0028] - the explainability pattern of the appropriate pattern database includes a influential characteristic of the type of DMI considered and the level of impact associated with the influential characteristic, the influential characteristic and the level of impact being common to at least one transaction of the type of DMI considered in the relevant transaction database, and

[0029] - the explainability pattern of the inappropriate pattern database includes an influential characteristic of the type of DMI considered and the level of impact associated with the influential characteristic, the influential characteristic and the level of impact being common to at least one transaction of the type of DMI considered in the database of inappropriate transactions;

[0030] - the set of DMI types includes the DMI types of "stent", " intrasaccular cage”, “flow diverter” or endovascular embolization by microcoils;

[0031] - a display of the probability of compatibility of each type of DMI and of essential characteristics of the vascular structure for each probability of compatibility.

[0032] According to another aspect, a computer program product is proposed comprising code instructions for the execution of a process as previously described, when said program is executed on a computer. DESCRIPTION OF THE FIGURES

[0033] Other features, purposes and advantages will become apparent from the following description, which is purely illustrative and not limiting, and which should be read in conjunction with the accompanying drawings on which:

[0034] Fig. 1 illustrates a schematic view of a DMI positioned in a vascular structure comprising an area of ​​interest, according to one embodiment of the present description.

[0035] Fig. 2 schematically illustrates the steps of an initialization of the process of choosing a type of DMI, according to a particular implementation of the present presentation.

[0036] Fig. 3 schematically illustrates the classification steps of the process for choosing a type of DMI, according to a particular implementation of the present exposition.

[0037] Throughout the figures, similar elements bear identical references. DETAILED DESCRIPTION

[0038] Aneurysm and DMI

[0039] Figure 1 schematically illustrates a vascular structure 1. The vascular structure 1 includes a region of interest 2. The region of interest 2 is an area of ​​a wall of the vascular structure 1 that presents a defect. The region of interest 2 may be an aneurysm or any other pathology related to the vascular structure 1. The aneurysm illustrated in Figure 1 includes a sac called an aneurysmal sac, but this discussion concerns any other type of aneurysm. Hereafter, for simplicity, any type of region of interest 2 will be referred to as an aneurysm 2.

[0040] To treat an aneurysm, a clinician must choose an implantable medical device (IMD). Choosing an IMD may involve selecting a type of IMD and a size of IMD. Here, we will limit ourselves to choosing the type, or model, of IMD. The different types of IMDs from which the clinician must choose are generally a stent, an intrasaccular cage, a flow diverter (as illustrated in Fig. 1), or an endovascular microcoil embolization device. Each of these types of IMDs also comes in different sizes, from which the clinician will subsequently have to make a choice.

[0041] The DMI 3 is configured to be inserted and then positioned in the vascular structure 1. The positioning of the DMI 3 in the vascular structure 1 prevents expansion or rupture of the aneurysm 2 or any other problem related to the presence of the aneurysm 2 in the vascular structure 1, such as the creation of a dried blood clot in the aneurysm 2 or a migration of a blood clot created in the aneurysm 2 to the vascular structure 1.

[0042] The type of DMI 3 must therefore be chosen according to the vascular structure 1 as well as according to the aneurysm 2. In other words, the vascular structure 1, and therefore the aneurysm 2, include characteristics 4 which are specific to it and which must be taken into account in a process of choosing the type of DMI 3 in order to treat the aneurysm 2. Indeed, each type of DMI 3 is more or less suited to the treatment of each aneurysm 2.

[0043] The characteristics 4 of the vascular structure 1 are physical and morphological characteristics 4 that influence the choice of the type of DMI 3. Each characteristic 4 includes degrees that allow the characteristic to be qualified. Indeed, each vascular structure 1 exhibits a particular degree of the characteristic for each characteristic 4. For example, the characteristics 4 and their degrees may be, but are not limited to: Degree 1 Degree 2 Degree 3 Characteristic Structure shape straight arched angled structure size small medium large Aneurysm shape spherical oval tubular Aneurysm volume small medium large Neck width small medium large Bifurcation Yes No

[0044] Of course, there are other characteristics 4 allowing to characterize the vascular structure 1.

[0045] Hereafter, we will refer to characteristic 4 of a vascular structure 1 to refer to the characteristic and degree of characteristic 4 for the vascular structure 1 under consideration. For example, for vascular structures 1 A and B and a characteristic 4 fl comprising three degrees 1, 2 and 3; if the degree of A for fl is 1 and the degree of B for fl is 3, we will refer to fl = 2 for characteristic 4 of A and fl = 3 for characteristic 4 of B.

[0046] Method for selecting a type of DMI

[0047] A method for selecting the type of DMI 3 to be positioned in the vascular structure 1 to treat the aneurysm 2 is described below. The method is configured to classify the type(s) of DMI 3 most suitable for treating the aneurysm 2 to be treated, based on an image of the vascular structure 1 including the aneurysm 2. Advantageously, the method provides, for each type of DMI 3, a score quantifying the suitability of the considered DMI 3 type for treating the aneurysm 2. Furthermore, the method advantageously provides an explanation of the score, that is, the characteristics 4 of the vascular structure 1 that led to this score and not another score.

[0048] The selection process includes an initialization El and a classification E2. The initialization El is implemented prior to the implementation of the classification E2 in order to allow the implementation of the classification E2.

[0049] The initialization El includes a generation Eli of a supervised learning model 7 and a creation E14 of one or more pattern database(s) 151, 152.

[0050] The E2 classification is carried out by implementing the trained supervised learning model 7 and by using the pattern database(s) 151,152.

[0051] To generate the supervised learning model 7, various algorithms can be used, such as a random forest (or "random forest" according to Anglo-Saxon terminology), or even a neural network. The model Supervised learning 7 is first trained and then configured to calculate, for each type of DMI 3 in the set of DMI 3 types, when implemented during classification E2 for example, a probability, called the compatibility probability 6. The compatibility probability 6 quantifies the compatibility of a DMI 3 with the aneurysm 2 of the vascular structure 1. In other words, the compatibility probability 6 provides a score quantifying the suitability of the considered type of DMI 3 to treat the aneurysm 2.

[0052] Furthermore, the pattern database(s) 151,152 are created following the generation Eli of the supervised learning model 7 to provide a library of explainability patterns 14 providing an explanation of the compatibility probability 6 of each type of DMI 3. In other words, the pattern database(s) 151,152 allow the selection of the characteristics 4 of the vascular structure 1 that were decisive in the calculation of the compatibility probability 6 of each type of DMI 3.

[0053] The method can be implemented on a server. Advantageously, the initialization E1 can be implemented by a first server and the classification E2 can be implemented by a second server, separate and independent from the first server. Advantageously, the first and second servers each comprise a processing unit. The processing unit of the first and second servers is, for example, a processor configured to implement the steps of the method that will be described below.

[0054] The first and second servers may each include a communication interface enabling communication between them. Such communication can, for example, be implemented via a wired or wireless connection through any type of communication network, such as the Internet. The first and second servers can, however, operate independently of each other. Once the E1 initialization has been performed by the first server and its results transmitted to the second server, the E2 classification can be implemented by the second server without requiring communication between them. Indeed, the results of the E1 initialization can be stored in a remote and decentralized storage service (or Cloud, according to commonly used Anglo-Saxon terminology).

[0055] Advantageously, the first and second servers can each also include a storage unit, for example, a hard drive. Typically, the first server can store one or more databases used for the selection process in the storage unit or have access to this or these databases. The architecture can be advantageously supplemented by a random access memory (RAM) which allows the intermediate calculations to be stored during the initialization El.

[0056] A person skilled in the art will understand without difficulty that other architectures for the implementation of the process are possible, the first server and the second server being able in particular to be merged into a single server.

[0057] Fig. 2 schematically illustrates steps of a particular implementation of the initialization El, i.e. the generation steps Eli of the supervised learning model 7 and the creation E14 of the pattern database(s) 151, 152.

[0058] Model learning

[0059] The method includes the generation Eli of the supervised learning model 7, hereafter referred to as the model for simplicity. The generation Eli of model 7 allows the model 7, once trained, to calculate, for each type of DMI 3 in the set of DMI 3 types, the probability of compatibility 6 of the DMI 3 with the vascular structure 1, or, in other words, with the area of ​​interest 2 of the vascular structure 1.

[0060] The Eli generation includes an E12 extraction of features 4 and an E13 training of the model 7 from the extracted features 4.

[0061] The extraction E12 of features 4 is performed on reference vascular structures la derived from reference images previously recorded and stored in a reference database (not shown). The previously recorded reference images can be simple images of the reference vascular structures la or 3D surface meshes extracted from the images of the reference vascular structures la. Advantageously, each of the reference vascular structures la includes at least one pathology such as an aneurysm 2. The reference vascular structures la and the aneurysms 2 of the reference vascular structures la include numerous specific features 4. The reference database includes, for each vascular structure la, several features 4 of each reference vascular structure la, as well as a type of DMI 3 adapted to the pathology.

[0062] Once extracted, the features 4 populate a feature database 5. The feature database 5 includes, for each image of reference vascular structures la, the features 4 specific to the reference vascular structure 1 considered as well as a mention of the type of DMI 3 associated with the treatment of the reference vascular structure la.

[0063] The E13 training of model 7 is implemented using the feature database 5. The presence in the feature database 5 of the features 4 specific to a reference vascular structure la and of the type of DMI 3 used to process the considered reference vascular structure la allows to train model 7. Indeed, model 7 uses an algorithm to calculate the links between the presence of one or more characteristics 4 and the choice of a particular type of DMI 3. Since the characteristics 4 of each extracted reference vascular structure are linked in the characteristic database 5 to a type of DMI 3, model 7 can perform calculations on all these relationships to deduce compatibility probabilities 6 of each type of DMI 3 with a vascular structure based on the characteristics 4 of the vascular structure 1 under consideration. In other words, the E13 training of model 7 allows model 7 to establish links between one or more characteristics 4 and the compatibility of each type of DMI 3 with a vascular structure 1 exhibiting that characteristic 4.

[0064] Model 7 is trained (step E13) from the characteristic database 5. Once trained, model 7 is configured to calculate (step E15), based on the specific characteristics 4 of a vascular structure 1, a compatibility probability 6 of each type of DMI 3 with the vascular structure 1 under consideration.

[0065] Creation of the pattern database(s)

[0066] The method also includes the creation E14 of a pattern database 151,152. Preferably, the method includes the creation E14 of two pattern databases 151,152. The creation E14 of the pattern databases 151,152 subsequently allows the selection, for the compatibility probability 6 of each type of DMI 3 for a vascular structure 1 to be treated, of the characteristics 4 of the vascular structure 1 considered having been decisive in the calculation of the compatibility probability 6.

[0067] The characteristics 4 of the vascular structure l having been decisive in the calculation of the probability of compatibility 6 are the characteristics 4 selected (step E24) by the implementation of the pattern databases 151,152, once the creation E14 has been carried out and are said to be essential characteristics 40.

[0068] The creation E14 of the pattern databases 151,152 includes a calculation E15 of compatibility probability 6 and of an explainability table 9, a first formation E16 of one or more transaction database(s) 11,12, an extraction E17 of explainability patterns 14 and a second formation E18 of the pattern databases 151,152.

[0069] Calculation E15 of the compatibility probability 6 of each type of DMI 3 with the reference vascular structures la uses the characteristic database 5. Model 7 then gives, for each reference vascular structure la in the characteristic database 5, a compatibility probability 6 of each type of DMI 3. For each reference vascular structure la, the compatibility probability 6 of each type of DMI 3 is between 0 and 1 and the sum the probabilities of compatibility 6 of the types of DMI 3 for each of the reference vascular structures is equal to 1.

[0070] The explainability table 9 is calculated by applying an explainability method to the characteristic data base 5 and the compatibility probabilities 6. The explainability method is configured to quantify, for each reference vascular structure la, an impact level 13 of each characteristic 4 in the calculation of the compatibility probability 6 of each type of DMI 3. The explainability table 9 therefore includes, for each reference vascular structure la, the impact level 13 of each characteristic 4 in the calculation of the compatibility probability 6 of each type of DMI 3. The impact level 13 reflects the weight of characteristic 4 in the result of the calculation of the compatibility probability 6 of each type of DMI 3 for the reference vascular structure la considered.

[0071] Advantageously, the explainability method is an interpretability method such as, for example, a LIME algorithm (or "Local Interpretable Model-agnostic Explanations", according to Anglo-Saxon terminology) or a method for estimating Shapley values, called the SHAP algorithm.

[0072] The explainability method associates, for each reference vascular structure la, with each characteristic 4 (in other words, with each degree of characteristic) an impact level 13. The impact level 13 can range from -3 to +3, with -3 expressing a significant negative impact level 13, +3 expressing a significant positive impact level 13, and 0 expressing no impact level 13. In other words, for each reference vascular structure la, the explainability method indicates for each characteristic 4 whether this characteristic 4 (in other words, whether the degree of this characteristic) has tended to increase or decrease the compatibility probability 6 of each type of DMI 3. Furthermore, the explainability method indicates for each characteristic 4 how this characteristic 4 has tended to increase or decrease the compatibility probability 6 of each type of DMI 3, for example, whether the impact is low or high.

[0073] Advantageously, the impact level 13 can be discretized into a predetermined number of values ​​to facilitate understanding of the explainability table 9. For example, the impact level 13 can be discretized into 7 values, ranging from -3 (strong negative impact) to +3 (strong positive impact), passing through 0 (neutral impact). The value of the impact level 13 can thus reflect a very strong / strong / weak positive / negative impact and a neutral impact.

[0074] The creation E14 of the pattern databases 151,152 also includes, for each reference vascular structure la, a classification of each type of DMI 3. Each type of DMI 3 is classified as appropriate or inappropriate for the reference vascular structure la considered. The classification is carried out according to the probability of compatibility 6 of the type of DMI 3 for the reference vascular structure considered.

[0075] Thus, for each reference vascular structure la, each type of DMI 3 is classified as type: - appropriate if the probability of compatibility 6 of type 3 DMI is greater than a predetermined threshold, and - inappropriate if the probability of compatibility 6 of type DMI 3 is less than the threshold.

[0076] The appropriate DMI 3 types are the DMI 3 types most suitable for properly treating aneurysm 2 of the reference vascular structure considered.

[0077] The threshold used to classify DMI 3 types as appropriate and inappropriate can be set based on equiprobability. In other words, if the number of DMI 3 types in the DMI 3 set is four, the threshold will be 0.25. Thus, any DMI 3 type whose probability of compatibility with a reference vascular structure is greater than 0.25 is an appropriate type for the considered reference vascular structure, and conversely for any DMI 3 type whose probability of compatibility with a reference vascular structure is less than 0.25.

[0078] The creation E14 of the pattern databases 151,152 includes the first formation E16 of one or more transaction database(s) 121,122. Each transaction database 121,122, includes, for each type of DMI 3, one or more transactions 12. Each transaction 12 of the transaction database 121,122, is specific to a type of DMI 3. The transaction 12 of a type of DMI 3 includes influential characteristics 4a of a reference vascular structure la for the type of DMI 3 considered.

[0079] The influential characteristics 4a are the characteristics 4 of the reference vascular structure considered whose impact level 13 is not equal to zero for the type of DMI 3 considered.

[0080] Preferably, a first transaction database 121,122, called the appropriate transaction database 121 and a second transaction database 121,122, called the inappropriate transaction database 122, are formed (step E16) during the creation E14 of the pattern databases 151, 152.

[0081] The appropriate transaction database 121 comprises, for each type of DMI 3, the transactions 12 of the reference vascular structures la for each of which the DMI 3 type is classified as an appropriate type. The inappropriate transaction database 122 comprises, for each type of DMI 3, the transactions 12 of the reference vascular structures la for each of which the DMI 3 type is classified as an inappropriate type.

[0082] The appropriate transaction database 121 therefore includes for each type of DMI 3 one or more transactions 12. Each transaction 12 of the appropriate transaction database 121 includes the influential characteristics 4a, and the associated impact levels 13, of the reference vascular structures la for which the type of DMI 3 considered is classified as an appropriate type.

[0083] The inappropriate transaction database 122 therefore includes for each type of DMI 3 one or more transactions 12. Each transaction 12 of the inappropriate transaction database 122 includes the influential characteristics 4a, and the associated impact levels 13, of the reference vascular structures la for which the type of DMI 3 considered is classified as an inappropriate type.

[0084] The creation E14 of the pattern databases 151,152 then includes an extraction E17 of the explainability patterns 14. The explainability patterns 14 are extracted from the transaction databases 121,122. Advantageously, in the case where a database of appropriate transactions 121 and a database of inappropriate transactions 122 have been created, the explainability patterns 14 are extracted from the database of appropriate transactions 121 and the database of inappropriate transactions 122.

[0085] The explainability patterns 14 are specific to each type of DMI 3. Indeed, each explainability pattern 14 is extracted from a transaction 12, which is itself specific to a type of DMI 3. The explainability pattern 14 is a pair, from a transaction 12, comprising an influential characteristic 4a and an impact level 13 associated with the influential characteristic 4a, the impact level 13 of the influential characteristic 4a being non-zero and the pair being common to one or more other transactions 12 of the type of DMI 3 considered.

[0086] Advantageously, the explainability pattern 14 can include several pairs extracted from the same transaction 12 considered if several influential characteristics 4a of the transaction 12 considered are common to one or more other transactions 12 of the same type of DMI 3.

[0087] Advantageously, a method for extracting explainability patterns 14 can be implemented. The extraction method is based on a frequent pattern extraction algorithm (or "frequent itemset extraction algorithm") such as, for example, an Apriori algorithm or an FPGrowth algorithm. A pair comprising an influential feature 4a and an impact level 13, associated with the influential feature 4a, of a transaction 12, is considered to be an explainability pattern 14 if it is common to at least K transactions 12, K being a parameter fixed by a user of the selection process.

[0088] The extraction method may further include a means for selecting a so-called maximal explainability pattern 14. An explainability pattern 14 is maximal if it is the most frequent explainability pattern 141e for the type of DMI in question 3.

[0089] The explainability pattern 14 of a type of DMI 3 is: - an appropriate reason 141 if the explainability reason 14 arises from a transaction 12 for reference vascular structures for which the type of MID 3 considered is an appropriate type and is common to one or more transactions 12 for reference vascular structures for which the type of MID 3 considered is an appropriate type; and - an inappropriate reason 142 if the explainability reason 14 is from a transaction 12 for reference vascular structures for which the type of DMI 3 considered is an inappropriate type and is common to one or more transactions 12 for reference vascular structures for which the type of DMI 3 considered is an inappropriate type.

[0090] In the case where a database of appropriate transactions 121 and a database of inappropriate transactions 122 have been created, the explainability pattern 14 of a type of DMI 3 is: - an appropriate reason 141 if the explainability reason 14 is extracted from a transaction 12 from the appropriate transaction database 121, or - an inappropriate reason 142 if the explainability reason 14 is extracted from a transaction 12 from the inappropriate transaction database 122.

[0091] The creation E14 of the pattern databases 151,152 then includes the second training E18 during which the extracted explainability patterns 14 are recorded in a pattern database 151,152. Advantageously, the second training E18 includes the creation of a suitable pattern database 151 and the creation of an unsuitable pattern database 152. The suitable patterns 141 are recorded in the suitable pattern database 151 and the unsuitable patterns 142 are recorded in the unsuitable pattern database 152.

[0092] Preferably, the database of appropriate patterns 151 consists, for each type of DMI 3, of one or more explainability patterns 14, each extracted from a transaction 12 of the database of appropriate transactions 121. And, the database of inappropriate patterns 152 consists, for each type of DMI 3, of one or more explainability patterns 14, each extracted from a transaction 12 of the database of inappropriate transactions 122.

[0093] Advantageously, the explainability patterns 14 of the appropriate pattern database 151 and the inappropriate pattern database 152 can each include the influential features 4a that affected the probability calculation compatibility 6 of a type of DMI 3 considered, whether in such a way as to increase or decrease the probability of compatibility 6 of the type of DMI 3 considered.

[0094] The appropriate pattern database 151 and the inappropriate pattern database 152 link the explainability patterns 14, including influential characteristics 4a, and their associated impact level 13. The appropriate pattern database 151 and the inappropriate pattern database 152 thus make it possible to select, for each new vascular structure 1 for which compatibility probabilities 6 of each type of MDI 3 have been calculated, one or more of the characteristics 4 of the vascular structure 1, referred to as the essential characteristic 40, along with their impact level 13. The determination of the essential characteristics 40 is detailed below.

[0095] Optionally, the appropriate transaction database 121 and the inappropriate transaction database 122 can be formed in a single global database and the appropriate pattern database 151 and the inappropriate pattern database 152 can also be formed in a single global database.

[0096] Classification

[0097] The method includes the implementation of the E2 classification. The E2 classification includes the implementation of the model 7 previously trained by the initialization El and the use of the database of appropriate patterns 151 and the database of inappropriate patterns 152 created during the initialization El.

[0098] Advantageously, the E2 classification is implemented independently of the El initialization; in other words, the E2 classification can be carried out without the El initialization necessarily being carried out as well; it is only necessary that the El initialization have been carried out once in order to train the model 7 and to create the database of appropriate patterns 151 and the database of inappropriate patterns 152. The selection process can therefore be implemented simply by carrying out the E2 classification once the El initialization has been previously carried out.

[0099] The E2 classification includes several steps to choose the most suitable type of DMI 3 for the treatment of an aneurysm 2 present in a vascular structure 1 to be treated, indicating the reasons for this choice.

[0100] Classification E2 includes obtaining an image E21 of vascular structure 1 including aneurysm 2. The image of vascular structure 1 can be obtained in various ways and is transmitted to the processing unit of the second server. From now on, vascular structure 1 will be understood to mean the vascular structure 1 containing the aneurysm 2 to be treated. This vascular structure 1 is distinct from the reference vascular structures 1a.

[0101] The image of the vascular structure 1 can be a two-dimensional, 2D image, or a three-dimensional, 3D image.

[0102] The E2 classification then implements an E22 extraction of features 4 from the vascular structure 1. The E22 extraction can be performed in the same way as the E12 extraction of features 4 from the reference vascular structures 1a. The extracted features 4 are then processed by the trained model 7.

[0103] The model 7 performs an E23 processing of the features 4 of the vascular structure 1 previously extracted from the image of the vascular structure 1. The E23 processing makes it possible to calculate the compatibility probability 6 of each type of DMI 3 for the vascular structure 1 to be processed. Advantageously, the compatibility probabilities 6 of each type of DMI 3 are between 0 and 1. The sum of the compatibility probabilities 6 of each of the types of DMI 3 is equal to 1.

[0104] The E2 classification then allows a selection E24 of at least one essential characteristic 40 that has influenced the probability of compatibility 6 associated with each type of MID 3 through the use of the database of appropriate patterns 151 and the database of inappropriate patterns 152.

[0105] Selection E24 first classifies each type of DMI 3 according to the probability of compatibility 6 of the type of DMI 3 with the vascular structure 1. Each type of DMI 3 is therefore classified as a suitable type or as an unsuitable type for the vascular structure 1. The type of DMI 3 is: - a suitable type if the probability of compatibility 6 of the DMI type 3 is greater than a previously set threshold, or - an inappropriate type if the probability of compatibility 6 of the type of DMI 3 is less than the threshold.

[0106] The threshold can be the same as the threshold set during the implementation of the initialization El or be recalculated according to the same method or according to a different method.

[0107] The E24 selection is then configured to select for each type of DMI 3 one or more essential characteristic(s) 40 of the vascular structure 1.

[0108] Selection E24 determines the essential characteristics 40 of the vascular structure 1 for each type of DMI 3 using the database of appropriate patterns 151 and the database of inappropriate patterns 152.

[0109] The essential characteristics 40 are selected by comparing the characteristics 4 of the vascular structure 1 with the explainability patterns 14 associated with the type of DMI 3 considered, recorded in the database of appropriate patterns 151 or the database of inappropriate patterns 152. In other words, the essential characteristics 40 are selected by comparing the characteristics 4 of the vascular structure 1: - to the appropriate reasons 141 of the type of DMI 3 considered if the type of DMI 3 is an appropriate type for vascular structure 1, or - to inappropriate patterns 142 of type DMI 3 if type DMI 3 is an inappropriate type for vascular structure 1.

[0110] For each type of DMI 3, selection E24 compares the characteristics 4 of the vascular structure 1 with the influential characteristics 4a of the explainability patterns 14 associated with the type of DMI 3 considered. Furthermore, selection E24 retains the explainability pattern(s) 14 with the most influential characteristics 4a common to the characteristics 4 of the vascular structure 1 to be treated, among the explainability patterns 14 of the type of DMI 3 considered. Finally, selection E24 retains as essential characteristics 40 of the vascular structure 1 all the influential characteristics 4a of the retained explainability patterns 14 common to the characteristics 4 of the vascular structure 1 to be treated.

[0111] Advantageously, selection E24 further retains the level of impact 13 associated with the influential characteristics 4a selected as essential characteristics 40 of the vascular structure 1.

[0112] Thus, the essential characteristics 40 of a vascular structure 1 for a type of MID 3 are the characteristic(s) of this vascular structure 1 selected as having influenced the probability of compatibility 6 of the type of MID 3. The essential characteristics 40 may be characteristics 4 of the vascular structure 1 that favored the choice of the type of MID 3 considered, if the type of MID 3 is proposed as appropriate for treating the vascular structure 1, and favored a rejection of the type of MID 3 considered, if the type of MID 3 is proposed as inappropriate for treating the vascular structure 1. This allows a user to know the reasons for the choice of a type of MID based on the vascular structure 1 considered. But, the essential characteristics 40 may also be characteristics 4 of the vascular structure 1 that tended to: - to discourage the choice of the type of DMI 3 considered, in the case of a type of DMI 3 proposed as appropriate for treating vascular structure 1, or - to discourage a refusal of the type of DMI 3 considered, in the case of a type of DMI 3 proposed as inappropriate to treat vascular structure 1.

[0113] This allows a user to know the elements of the vascular structure 1 considered which could call into question the choice of a type of DMI depending on the situation.

[0114] Finally, after processing E23 and selection E24, the process may advantageously include a display step E25 of the type of MID 3 having the highest probability of compatibility 6 with the vascular structure 1, or of the type(s) of MID 3 classified as suitable types. Optionally, the display E25 may carry on the compatibility probabilities 6 of all types of DMI 3 for vascular structure 1.

[0115] Advantageously, the E25 display can also relate to the essential characteristics 40 of the vascular structure 1 for the types of MDI 3 whose compatibility probability 6 is displayed. The impact level 13 can further be associated with the E25 display of each essential characteristic 40 shown.

[0116] Example of implementation of the steps of the selection process

[0117] Subsequently, an example of the implementation of the process for choosing a type of DMI 3 is explained.

[0118] In this example, the method implements an initialization from images of four reference vascular structures, labeled C1 to C4. The vascular structures 1 comprise three features 4, labeled f1, f2, f3, each comprising three degrees (1, 2, 3). The set of DMI types 3 comprises three different DMI types A, B, and C.

[0119] The following table presents the results of the E22 extraction by the probabilistic module of the features 4 fl, f2, f3, of each reference vascular structure Cl, C2, C3 and C4. These features 4 as well as the mention of the type of DMI 3 used for the processing of each reference vascular structure form the feature database 5 which allows the training E13 of the model 7.

[0120] The processing by model 7 of each reference vascular structure la of the database of characteristics 5 makes it possible to obtain the probability of compatibility 6 for each reference vascular structure la of each type of DMI 3 A, B and C, denoted P(A) for type of DMI A, P(B) for type of DMI B and P(C) for type of DMI C. Characteristic Image fl f2 f3 P(A) P(B) P(C) Cl 1 1 3 0.60 0.30 0.10 C2 1 3 1 0.45 0.20 0.35 C3 3 2 1 0.25 0.35 0.40 C4 3 2 3 0.15 0.70 0.15

[0121] Thus, for the vascular structure Cl having a characteristic fl of degree 1, a characteristic f2 of degree 1 and a characteristic f3 of degree 3, the probability of compatibility 6 of the type of MDI A, denoted P(A) is 60%, that of type B is 30% and that of type C is 10%.

[0122] Implementing the explainability method based on the characteristic database 5 as a function of the compatibility probabilities 6 of each type of DMI 3 obtained for each reference vascular structure 1 makes it possible to quantify an impact level 13 (very strong / strong / weak positive / negative impact and a neutral impact) of each characteristic 4 in the calculation E15 of the probability of compatibility 6 of each type of DMI 3. Case Solutions Impact fl Impact f2 Impact f3 Cl A strong not. weak neg. very strong pos. Cl B very strong ne g. neutral neutral Cl C strong neg. strong pos. very strong pos. C2 A strong not. neutral very strong pos. C2 B very strong neg. neutral weak neg. C2 C neutral very strong pos. weak pos. C3 A strong neg. weak neg weak pos. C3 B very strong pos. very strong pos. weak neg. C3 C weak pos. very weak neg. weak pos. C4 A strong neg weak neg neutral C4 B very strong pos. very strong pos. strong neg. C4 C strong pos. neutral very strong pos.

[0123]

[0124] Thus, for vascular structure Cl, the characteristic fl, due to its degree of 1, strongly positively influenced the probability of compatibility 6 of type DM A, and the characteristic f3, due to its degree of 3, very strongly positively influenced the probability of compatibility 6 of type DM A; while the characteristic f2, due to its degree of 1, weakly positively influenced the probability of compatibility 6 of type DM A. In other words, the presence of a characteristic f1=1 favors the use of type DM A for vascular structure Cl, the presence of a characteristic f3=3 favors it very strongly, while the presence of a characteristic f2=1 tends to disfavor the use, and therefore lower the probability of compatibility 6, of type DM A for vascular structure Cl. The appropriate transaction database 121 created includes for each type of DMI 3 the transactions 12 including the influential characteristics 4a of the reference vascular structures la for which the type of DMI 3 considered is an appropriate type. Transactions for solution A Transactions for solution B Transactions for solution C (fl=1, strong positive), (f2=1, weak negative), (f3=1, very strong positive) (fl=3, very strong positive), (f2=2, very strong positive), (f3=1, weak negative) (f2=3, very strong positive) (f3=1, weak positive) (f1=1, strong positive), (f3=3, very strong positive) (fl=3, very strong positive), (f2=2, very strong positive), (f3=3, strong negative) (fl=3, weak positive) (f2=2, very weak negative) (f3=1, weak positive)

[0125] The inappropriate transaction database 122 created includes for each type of DMI 3 the transactions 12 including the influential characteristics 4a of the reference vascular structures la for which the type of DMI 3 considered is an inappropriate type. Transactions for solution A Transactions for solution B Transactions for solution C (fl=3, strong neg.),(f2=2, weak neg.),(f3=1, weak pos.) (fl=1, very strong neg.) (fl=1, strong neg.)f2=1, strong pos.) (f3=3, very strong neg.) (fl=3, strong neg..),(f2=2, weak neg.) (fl=1, very strong neg.),(f3=1, weak neg.) (fl=3, strong pos.) (f3=3, very strong neg.)

[0126] Thus, type DM A is a suitable type for treating a vascular structure 1 if the vascular structure 1 includes the characteristic fl=l (in other words, whose fl characteristics are of degree 1), the characteristic f2=l and the characteristic f3=3 or if the vascular structure includes the characteristic fl=l and the characteristic f3=3. And, for example, type DM B is an unsuitable type for treating a vascular structure 1 if the vascular structure 1 includes the characteristic fl=l or if the vascular structure 1 includes the characteristic fl=l and the characteristic f3=l.

[0127] Following the extraction E17 of appropriate patterns 141 from the appropriate transaction database 121, the appropriate pattern database 151 includes the influential characteristics 4a of each type of DMI 3, and their associated level of impact 13, common to several transactions 12 of the considered type of DMI 3 from the appropriate transaction database 121. Suitable patterns for A Suitable patterns for B Suitable patterns for C (fl=1, strong pos.) (fl=3, very strong pos.) (f2=2, very strong pos.) (f3=1, weak pos.)

[0128] Following the extraction E17 of inappropriate patterns 142 from the inappropriate transaction database 122, the inappropriate pattern database 152 includes the influential characteristics 4a of each type of DMI 3, and their level impact 13 associated, common to several transactions 12 of the type of DMI 3 considered from the database of inappropriate transactions 122. Inappropriate patterns for A Inappropriate patterns for B Inappropriate patterns for C (fl=3, strong neg.), (f2=2, weak neg.) (fl=1, very strong neg.) (f3=3, very strong neg.)

[0129] Thus, and in summary, type DM A is an appropriate type for treating a vascular structure 1 if the vascular structure 1 includes the feature fl=l (in other words, whose fl features are of degree 1) and type DM B is an inappropriate type for treating a vascular structure 1 if the vascular structure 1 includes the feature f 1=1.

[0130] The database of appropriate patterns 151 and the database of inappropriate patterns 152 are thus formed (step E18).

[0131] The process then includes carrying out the steps of obtaining E21, extracting E22, processing E23 and selecting E24 by algorithm 7 and using the database of suitable patterns 151 and the database of unsuitable patterns 152.

[0132] For a vascular structure 1 C5 to be treated, extraction E22 provides the characteristics 4 of the vascular structure 1 as shown below. Processing E23 by algorithm 7 provides the compatibility probabilities 6 of each type of DMI 3 A, B and C for the vascular structure 1 C5. Case fl f2 f3 P(A) P(B) P(C) C5 1 2 3 0.60 0.15 0.25

[0133] Selection E24 allows DMI type 3A to be classified as appropriate and DMI types 3B and C as inappropriate. DMI type 3A is classified as appropriate because the probability of compatibility 6 of DMI type 3A for vascular structure 1C5 is greater than 1 / 3, or 0.33, the number of DMI type 3 being three. Similarly, DMI types 3B and C are classified as inappropriate for vascular structure 1C5 because their probability of compatibility 6 is less than 1 / 3. Selection E24 also allows the selection of the essential characteristics 40 of vascular structure 1C5 for each DMI type 3, as well as the impact level associated with the essential characteristics 40, through the use of the appropriate pattern database 151 and the inappropriate pattern database 152.

[0134] Finally, display E25 allows the following table to be presented. Medical Device Type 13 Probability of Compatibility 6 Essential Related Characteristics 40 P(A) 0.60 (fl=f strong positive) P(B) 0.15 (fl=l, very strong ne g) P(C) 0.25 (f3=3, very strong ne g)

[0135] Thus, for the C5 vascular structure, type A DM is the most suitable for treating the aneurysm, its P(A) compatibility probability being the highest. Furthermore, this choice is explained by the presence of the characteristic f 1=1 in the C5 vascular structure. This characteristic fl=l is therefore an essential characteristic of the C5 vascular structure, and its level of impact on the choice of type A DM is very positive; in other words, the characteristic fl=la strongly influenced the choice of type A DM for treating the C5 vascular structure.

Claims

Demands

1. A method for selecting a type of implantable medical device (3), IMD (3), from a set of IMD types (3), the IMD (3) being intended to be positioned in a vascular structure (1) comprising a region of interest (2), the method comprising the following steps: - obtaining (E21) an image of a vascular structure (1) comprising a region of interest (2); - extracting (E22) a plurality of features (4) of the vascular structure (1) from the image obtained; - processing (E23) by a supervised learning model (7) of the extracted features (4) to obtain, for each type of IMD (3), a compatibility probability (6) quantifying a compatibility of the type of IMD (3) with the region of interest (2) of the vascular structure (1);and - selection (E24) from among the extracted characteristics (4), and according to each compatibility probability (6), of at least one essential characteristic (40) influencing the compatibility probability (6) associated with each type of DMI (3).;

2. A method according to claim 1, wherein the selection (E24) classifies each type of DMI (3) as a suitable type, if the probability of compatibility (6) of the type of DMI (3) is greater than a previously fixed threshold, or as an unsuitable type, if the probability of compatibility (6) of the type of DMI (3) is less than the threshold.

3. A method according to claim 2, wherein the selection (E24) is implemented based on a database of suitable patterns (151) and a database of unsuitable patterns (152), the database of suitable patterns (151) and the database of unsuitable patterns (152) each comprising at least one explainability pattern (14) for each type of DMI (3), each explainability pattern (14) comprising at least one influencing feature (4a) and an impact level (13) associated with the influencing feature (4a), the influencing feature (4a) being a feature (4) having a non-zero associated impact level (13), the impact level (13) qualifying the impact of each feature (4) in the calculation of the probability of compatibility (6) of the type of MDI (3) considered; the explainability pattern (14) of the database of appropriate patterns (151) being common to at least one reference vascular structure (la) for which the type of MDI (3) considered is an appropriate type; and the explainability pattern (14) of the database of inappropriate patterns (152) being common to at least one reference vascular structure (la) for which the type of MDI (3) considered is an inappropriate type.

4. A method according to claim 3, wherein the selection (E24) determines for each type of DMI (3) the essential characteristic (40): - by comparing the characteristics (4) of the vascular structure (1) to the explainability patterns (14) of the database of appropriate patterns (151) if the type of DMI (3) considered is an appropriate type or to the explainability patterns (14) of the database of inappropriate patterns (152) if the type of DMI (3) considered is an inappropriate type; and - by retaining as essential characteristic(s) (40) the influential characteristic(s) (4a) common to the characteristics (4) of the vascular structure (1), the influential characteristic(s) (4a) being included in one or more explainability patterns (14) having the most influential characteristics (4a) common to the characteristics (4) of the vascular structure (1).

5. A method according to any one of claims 1 to 4, comprising a generation (E1 1) of the model (7) from a feature database (5) comprising features (4) of each of the images of a plurality of reference vascular structures (la) from a training image database.

6. A method according to claim 5, wherein the compatibility probabilities (6) of each type of DMI (3) for the reference vascular structures (la) are calculated (El5) by the model (7) from the characteristic database (5).

7. A method according to any one of claims 3 to 6, comprising a first formation (El6) of a database of appropriate transactions (121) and a database of inappropriate transactions (122); - the database of appropriate transactions (121) comprising for each type of DMI (3) a transaction (12) comprising the influential characteristics (4a) of a reference vascular structure (la) for which the type of DMI (3) is considered to be an appropriate type; and - the database of inappropriate transactions (122) comprising for each type of DMI (3) a transaction (12) comprising the influential characteristics (4a) of a reference vascular structure (la) for which the type of DMI (3) is considered to be an inappropriate type.

8. A method according to claim 7, wherein each transaction (12) is calculated as a function of the compatibility probability (6) of the type of MDI (3) considered for each reference vascular structure (la) and an explainability table (9) comprising for each reference vascular structure (la) the impact level (13) of each characteristic (4) for the compatibility probability (6) of the type of MDI (3) considered, each transaction (12) comprising the influential characteristics (4a) of a reference vascular structure (la) and the impact level (13) associated with each of the influential characteristics (4a).

9. A method according to any one of claims 7 to 8, comprising a second formation (El8) of the appropriate motive database (151) and the inappropriate motive database (152), each comprising, for each type of DMI (3), at least one explainability motive (14), such that: - the explainability motive (14) of the appropriate motive database (151) comprises an influential feature (4a) of the considered DMI type (3) and the impact level (13) associated with the influential feature (4a), the influential feature (4a) and the impact level (13) being common to at least one transaction (12) of the considered DMI type (3) in the appropriate transaction database (121); and - the explainability pattern (14) of the inappropriate pattern database (152) includes an influential feature (4a) of the type of MID (3) considered and the impact level (13) associated with the influential feature (4a), the influential feature (4a) and the impact level (13) being common to at least one transaction (12) of the type of MID (3) considered in the inappropriate transaction database (122).

10. A method according to any one of claims 1 to 9, wherein the set of DMI types (3) comprises the DMI types (3) of the "stent", "intrasaccular cage", "flow diverter" or endovascular microcoil embolization types.

11. A method according to any one of claims 1 to 10, comprising a display (E25) of the compatibility probability (6) of each type of MDI (3) and of essential characteristics (40) of the vascular structure (1) for each compatibility probability (6).

12. Product computer program comprising code instructions for the execution of a process according to any one of claims 1 to 11, when said program is executed on a computer.

Images