Device for predicting the optimal treatment type for percutaneous lesion ablation

Abstract

The invention relates to a device for implementing a method (100) for selecting an optimal treatment for the percutaneous ablation of a lesion (13) within a patient's anatomy of interest (14). The method uses a machine learning algorithm (20) trained to calculate, from a medical image (12) on which the lesion (13) is visible, and for each of a plurality of available treatments, a confidence index whose value represents the probability of success of said treatment for the ablation of the lesion. The machine learning algorithm is pre-trained from a set of training elements (21), each comprising a medical image on which a lesion is visible within the anatomy of interest of another patient, a treatment chosen from among the various available treatments for treating said other patient, and a confidence index of the chosen treatment on said other patient.The optimal treatment is then selected based on the confidence indices calculated for the different available treatments. Figure for the abstract: Fig. 1.

Description

Title of the invention: Device for predicting an optimal treatment type for percutaneous ablation of a lesion Field of the invention

[0001] The present invention belongs to the field of planning a minimally invasive medical intervention. In particular, the invention relates to a computer-type electronic device implementing a method for selecting an optimal treatment from among several available treatments for the percutaneous ablation of a lesion in an anatomy of interest of a patient. State of the art

[0002] To prepare for a medical procedure to treat a lesion (e.g., a tumor) in a patient's anatomy of interest (e.g., a lung, kidney, liver, bone structure, etc.), a practitioner typically plans the procedure based on a medical image.

[0003] The first step in this planning is to select a type of treatment. There are some broad clinical practice guidelines for selecting a treatment based on a type of pathology. For example, the first step is to decide whether to resort to ablation, transplantation, chemoembolization treatment, or radiotherapy.

[0004] However, it remains difficult to accurately select a more precise type of treatment.

[0005] More particularly, in the case of percutaneous ablation of a tumor, different treatment methods can be applied: radiofrequency, microwave, laser, cryotherapy, electroporation, or brachytherapy. Methods based on radiofrequency waves, microwave, laser or cryotherapy are thermal methods. Electroporation consists of delivering between two electrodes tens of radiofrequency pulses of a few ps at very high intensity to cause irreversible alterations of the membrane functions of the cells. Electroporation is a relatively "gentle" ablation technology which makes it possible to consider treating tumors which could not previously be treated with thermal ablation technologies, either because of the location of the tumor or because of the fragility of the patient. Brachytherapy, on the other hand, aims to deliver radioactive doses inside or near the tumor.

[0006] Whatever the technique used (thermal energy, electrical energy, radioactive energy, etc.), percutaneous ablation requires the placement of one or more applicators to deliver energy in situ. The energy delivered induces at the level irreversible cellular and tissue destructuring. In some cases, it may be necessary to perform overlapping ablations sequentially by several successive insertions of an applicator, or simultaneously by inserting and activating several applicators at the same time.

[0007] Furthermore, when several applicators are activated simultaneously, there are different strategies for the deposition of energy by the applicators. In particular, a distinction is made between so-called “centripetal convergent” ablations and so-called “centrifugal” ablations.

[0008] The centrifugal ablation strategy is currently the most widely used because it is relatively simple to implement. The centripetal convergent ablation strategy, however, offers better adaptability to the ablation zone and its environment. The centripetal convergent ablation strategy allows for better control of the destruction limits, which allows for a certain “modeling” of the ablation zones according to the shape and location of the tumors. On the other hand, it involves a greater sacrifice of healthy tissue.

[0009] These different methods, which conceptually may appear very similar, are in reality quite different both in terms of their practical use and the results they allow to be obtained.

[0010] The choice of a particular type of treatment is largely dependent on the specificities of each clinical situation. Each particular type of treatment may offer the best benefit / risk ratio in a particular clinical situation.

[0011] The choice of a particular type of treatment is generally made by a practitioner based on his habits or preferences. As a result, the treatment selected to treat a lesion in a patient's anatomy of interest may not be optimal, and this may lead to a recurrence of the pathology in the patient, or collateral lesions.

[0012] Also, however, there is currently no satisfactory solution to help a practitioner choose the optimal treatment from a set of treatments available for the ablation of a lesion in an anatomy of interest of a patient. Presentation of the invention

[0013] The present invention aims to remedy all or part of the drawbacks of the prior art, in particular those set out above.

[0014] To this end, and according to a first aspect, the present invention provides an electronic device comprising at least one processor and a computer memory. The computer memory stores program code instructions which, when executed by the processor, configure the processor to implement a method for selecting at least one optimal processing from among several treatments available for the percutaneous ablation of a lesion within a patient's anatomy of interest. The method implemented includes: - obtaining a medical image on which the lesion is visible within the patient's anatomy of interest, - a calculation, for each of the different ablation treatments available, from the medical image and using a machine learning algorithm, of a confidence index whose value is representative of a probability of success of said treatment for the ablation of the lesion in the anatomy of interest of the patient, the machine learning algorithm having been previously trained from a set of training elements, each training element comprising: • a medical image on which a lesion is visible within the anatomy of interest of another patient, • an identification of a treatment chosen from among the different ablation treatments available to treat said other patient, • a confidence index of said treatment on said other patient, - a selection of at least one optimal treatment based on the confidence indices calculated for the different treatments available.

[0015] "Percutaneous ablation" means a minimally invasive intervention through the patient's skin to treat a lesion in an anatomy of interest.

[0016] The anatomy of interest is, for example, the liver, a lung, a kidney, or a bone structure. The lesion is, for example, a tumor, a cyst, or an aneurysm. Different treatments may be considered to ablate the lesion in the anatomy of interest. A treatment may in particular be characterized by the choice of a technology (radiofrequencies, microwaves, laser, cryotherapy, electroporation, etc.) and an application method (for each technology, different energy deposition strategies are possible).

[0017] The machine learning algorithm makes it possible to calculate, from the medical image of the patient's anatomy of interest, a confidence index for each available treatment. An optimal treatment can then be selected based on the confidence indices thus calculated. For example, the treatment for which the confidence index is the highest corresponds to the treatment for which the probability of treatment success is the highest. The confidence index may in particular be representative of a risk of recurrence. There may possibly be several possible optimal treatments, for example if different available treatments are associated with a confidence index of the same value.

[0018] The machine learning algorithm is previously trained from a set of training elements. Each training element is for example associated with an intervention carried out in the past on the anatomy of interest of another patient according to one of the available treatments. A training element then includes a medical image of the anatomy of interest before treatment, an identification of the treatment which was carried out on the patient, and a confidence index for this treatment (the confidence index is for example defined according to the observation or not of a recurrence following the treatment).

[0019] A training element can, however, also be defined without an intervention having actually taken place on another patient. Indeed, the confidence index of a treatment envisaged to treat a lesion visible on a medical image of the anatomy of interest of a patient can be determined theoretically. For example, the confidence index can be estimated by a panel of medical experts on the basis of the medical image. This estimate can possibly be weighted by recommendations from scientific publications and / or empirical studies carried out on groups of patients presenting similar lesions.

[0020] The set of training elements corresponds to a database for training the machine learning algorithm to calculate a confidence index for a particular treatment from a medical image showing a lesion to be treated. A medical establishment (hospital, clinic, etc.) can create and enrich its own database for the treatments available within said medical establishment (for each new intervention carried out in this establishment, a new training element can be added to the database).

[0021] In addition to the medical image, the machine learning algorithm may also be based on a set of parameters (metadata) relating to the lesion, the anatomy of interest or the patient. These parameters may be determined automatically on the medical image (for example via segmentation or image processing algorithms based on artificial intelligence), or manually by a user. These parameters may correspond, for example, to the size of the lesion, the position of the lesion relative to the anatomy of interest or relative to other anatomical structures, a physical and medical condition of the patient, etc.

[0022] Different types of machine learning algorithms can be envisaged to implement the invention: decision tree forest (“random forest” in the English literature), neural network, support vector machines (SVM for “Support Vector Machine” in English), partitioning into k-means (“k-means” in English), etc.

[0023] The invention has various advantages. In particular, it makes it possible to obtain a reliable prediction of an optimal treatment for the percutaneous ablation of the lesion. The reliability of the prediction increases with the number of training elements. It is therefore preferable to train the machine learning algorithm from a large number of training elements. On the other hand, the invention makes it possible to identify an optimal treatment very quickly. In addition, the device according to the invention can be used by people who are not necessarily experts in percutaneous ablation (for example by interns or doctors undergoing training for this practice). The proposed solution therefore allows for the democratization of percutaneous ablation treatments.

[0024] In particular embodiments, the invention may further comprise one or more of the following characteristics, taken in isolation or in all technically possible combinations.

[0025] In particular embodiments, each available treatment is characterized by an ablation technology chosen from radiofrequencies, microwaves, laser, cryotherapy, electroporation or brachytherapy.

[0026] In particular embodiments, each available treatment is characterized by an energy deposition strategy by one or more applicators, said strategy being chosen from a centrifugal type energy deposition and a centripetal convergent type energy deposition.

[0027] In particular embodiments, each available treatment is characterized by a number of applicators, and / or by the positions of said applicators relative to the tumor.

[0028] In particular modes of implementation, for each training element, the confidence index is defined as a function of: - the observation or not of a recurrence for the said other patient with the chosen treatment, and / or - the duration of the period elapsed between the end of treatment and a current date without a recurrence having been observed for the said other patient, and / or - the duration of the period between the end of treatment and a recurrence for the said other patient, and / or - an estimate, by one or more medical experts, of the probability of success of the treatment chosen on the said other patient.

[0029] In particular embodiments, the method comprises an association of a set of parameters with the medical image of the patient, said set of parameters comprising one or more parameters relating to the lesion, to the anatomy of interest and / or to characteristics of the patient. Each training element comprises a set of similar parameters relating to the lesion, to the anatomy of interest and / or to characteristics of the other patient for which the medical image corresponding to said reference element was obtained.

[0030] In particular embodiments, the set of parameters comprises one or more parameters chosen from: - the age, sex, weight and / or height of the patient for whom the medical image was obtained, - a comorbidity presented by the patient for whom the medical image was obtained, - a value representative of the size and / or volume of the lesion, - a distance between the lesion and a capsule of the anatomy of interest, - a distance between the lesion and a blood vessel close to the lesion, - when the anatomy of interest is the liver, a distance between the lesion and a bile duct close to the lesion.

[0031] In particular embodiments, the method comprises a transformation of the medical image of the patient, the medical image of each training element having undergone a similar transformation before being used to train the machine learning algorithm.

[0032] In particular modes of implementation, the transformation of the medical image includes: - segmentation of the lesion on the medical image, and / or - a segmentation of the anatomy of interest on the medical image, and / or - a segmentation of blood vessels on the medical image, and / or - when the anatomy of interest is the liver, a segmentation of bile ducts on the medical image.

[0033] In particular modes of implementation, the transformation of the medical image comprises a reframing of the image around the lesion according to a frame whose dimensions are predetermined, said frame being common to all the medical images of the training elements, the position of the lesion relative to the frame being variable from one medical image to another.

[0034] In particular embodiments, the method comprises: - for each training element corresponding to the selected optimal treatment, a calculation of a similarity value representative of the similarity between the medical image of the patient to be treated and the medical image of the training element, and - a selection of a reference training element from among the set of training elements based on the calculated similarity values.

[0035] In particular embodiments, the medical images were acquired by computed tomography, magnetic resonance imaging or ultrasound.

[0036] In particular embodiments, the medical images are three-dimensional.

[0037] In particular embodiments, the anatomy of interest is the liver, a kidney or a lung, and the lesion is a tumor or a cyst.

[0038] In particular embodiments, the set of training elements comprises training elements whose medical images do not present any lesion in the anatomy of interest. Presentation of figures

[0039] The invention will be better understood on reading the following description, given by way of non-limiting example, and made with reference to Figures 1 to 4 which represent:

[0040] [Fig.l] a schematic representation of the main steps of a method for selecting an optimal treatment for the ablation of a lesion in an anatomy of interest of a patient,

[0041] [Fig.2] a schematic representation of an electronic device allowing the selection of an optimal treatment for the ablation of a lesion in an anatomy of interest of a patient,

[0042] [Fig.3] a schematic representation of a step of collecting a set of reference elements used to train the machine learning algorithm,

[0043] [Fig.4] a schematic representation of a particular mode of implementation of a method for selecting an optimal treatment, in which a reference training element is selected.

[0044] In these figures, identical references from one figure to another designate identical or similar elements. For reasons of clarity, the elements represented are not necessarily on the same scale, unless otherwise stated.

[0045] Detailed description of an embodiment of the invention

[0046] [Fig.l] schematically represents the main steps of a method 100 for selecting an optimal treatment for the ablation of a lesion in an anatomy of interest of a patient. The optimal treatment is chosen from several available treatments. It should be noted that several treatments may possibly be indicated as optimal by the method 100 without it being possible to separate them.

[0047] The method 100 is implemented by an electronic device of the computer, tablet, mobile phone, etc. type. [Fig.2] schematically illustrates an exemplary embodiment of such an electronic device 30. The electronic device 30 comprises at least one processor 32 and a computer memory 31 storing program code instructions which, when executed by the processor 32, configure the processor 32 to implement the method 100 for selecting a processing according to the invention.

[0048] The computer memory 31 therefore corresponds to a recording medium readable by the electronic device 30, and comprising instructions which, when they are executed by said electronic device 30, lead the latter to implement the method 100 for selecting a treatment.

[0049] The invention may also take the form of a computer program comprising instructions which, when the program is executed by the electronic device 30, cause the latter to implement the method 100 for selecting a treatment.

[0050] The method 100 comprises a step 101 of obtaining a medical image 12 on which the lesion 13 is visible within the anatomy of interest 14 of the patient.

[0051] In the example considered, the medical image 12 is a two-dimensional medical image acquired by computed tomography (CT-scan for “Computerized To-mography Scan” in the English literature). However, nothing would prevent the medical image from being acquired using another medical imaging method, for example by MRI (acronym for “Magnetic Resonance Imaging”) or by ultrasound. Nothing would also prevent the medical image from being a three-dimensional image.

[0052] The medical image 12 is for example stored in the computer memory 31 of the electronic device 30 and processed by the processor 32. The medical image 12 is for example transmitted to the electronic device 30 via wired communication means or wireless communication means (these means are not shown in [Fig. 2]). According to another example, the electronic device 30 can be connected to an external memory device, for example a USB key (acronym for “Universal Serial Bus”) on which the medical image 12 is recorded. In any event, the electronic device 30 is configured to obtain the medical image 12 which has been previously acquired on the patient to be treated.

[0053] In the example considered, and in a non-limiting manner, the anatomy of interest is the liver, and the lesion is a cancerous tumor. The invention could however also be applied to other anatomies of interest (for example a lung, a kidney, a bone structure, etc.), and / or to other types of lesions (for example a cyst, an aneurysm, a metastasis, etc.).

[0054] The available treatments, from which at least one optimal treatment is selected, correspond for example to different treatments used in a medical establishment to treat such a tumor.

[0055] An available treatment may in particular be characterized by an ablation technology chosen from radiofrequencies, microwaves, laser, cryotherapy, electroporation or brachytherapy. These different technologies are known to those skilled in the art for treating a cancerous tumor in the liver.

[0056] An available treatment may further be characterized by a deposition strategy of energy by one or more applicators (needle, electrode, probe, etc.). The energy deposition strategy can in particular be chosen from a centrifugal type of energy deposition and a centripetal convergent type of energy deposition.

[0057] For a long time, the preferred energy deposition strategy followed a centrifugal coverage scheme in which the energy is delivered isotropically from the center of the tumor (where the applicator is inserted) to the periphery of the tumor. In some cases, it may be necessary to perform overlapped ablations sequentially by several successive activations of an applicator inserted at different positions, or simultaneously by activating several applicators at different positions at the same time. The centrifugal type deposition strategy has the advantage of being a particularly simple and proven technique. However, under certain specific conditions (depending in particular on the size or volume of the tumor, or on the tissue properties of the lesion or the anatomical structures close to the lesion), the centrifugal type strategy is not always optimal.Another energy deposition strategy, consisting of converging the energy from the periphery towards the center of the tumor (centripetal convergent strategy) is then sometimes preferable.

[0058] In addition to the ablation technology and energy deposition strategy, an available treatment may further be characterized by the number of applicators used, and / or by the positions of said applicators relative to the tumor.

[0059] The position of the applicators is generally determined relative to the tumor, but it can also be determined relative to the environment of the tumor (i.e. by the anatomical structures close to the tumor: bones, blood vessels, organs at risk, etc.).

[0060] The method 100 comprises a calculation step 104, for each of the different available ablation treatments, of a confidence index whose value is representative of a probability of success of said treatment for the ablation of the lesion 13 in the anatomy of interest 14 of the patient. The confidence indices of the different available treatments are calculated from the medical image 12 by a previously trained machine learning algorithm 20.

[0061] As illustrated in [Fig.l], the machine learning algorithm is trained, during a training step 206, from a set of training elements 21. Each training element 21 is for example associated with an intervention carried out in the past to treat a lesion in the anatomy of interest of another patient according to one of the available treatments. A training element 21 then comprises a medical image of the anatomy of interest of said other patient before treatment, an identification of the treatment which was carried out on said other patient, and a confidence index for this treatment.

[0062] For each training element 21, the confidence index can be defined for example as a function of whether or not a recurrence has been observed for said other patient with the chosen treatment (if there has been no recurrence, the probability of success of the treatment is relatively high; conversely, if there has been a recurrence, the probability of success of the treatment is lower).

[0063] The length of the period elapsed between the end of treatment and the current date without a recurrence having been observed can also be a factor used to define the treatment confidence index (the longer this period, the greater the probability of treatment success).

[0064] If there has been a recurrence, it is possible to take into account the duration of the period elapsed between the end of treatment and the recurrence to determine the confidence index (the shorter this period, the lower the confidence index).

[0065] It should be noted that the step 200 of collecting the training elements and the step 206 of training the automatic learning algorithm are carried out prior to the method 100 of selecting an optimal treatment (they are therefore not part of the method 100 of selecting an optimal treatment).

[0066] It should also be noted that the training elements 21 can also be obtained without interventions having actually taken place on other patients. Indeed, the confidence index of a treatment envisaged for treating a lesion visible on a medical image of the anatomy of interest of another patient can be determined theoretically by a panel of medical experts.

[0067] Once the confidence indices have been calculated by the machine learning algorithm for the different available treatments, it becomes possible to select, during a selection step 105, at least one optimal treatment from among the different available treatments. For example, the optimal treatment corresponds to the treatment for which the calculated confidence index is the highest. As indicated previously, there may possibly be several possible optimal treatments, for example if different available treatments are associated with a confidence index of the same value.

[0068] The training element set 21 corresponds to a database used to train the machine learning algorithm to calculate a confidence index for a particular treatment from a medical image showing a lesion to be treated. This database can be enriched over time, each time a new medical image showing a lesion in the anatomy of interest of a patient is available and a confidence index has been determined for a particular treatment of this lesion.

[0069] It is possible to carry out a selection on the cases likely to be used to form training elements 21. In particular, when the elements training models 21 are constructed on the basis of interventions that actually took place, it is advantageous to consider only those interventions for which the ablation margin was sufficient to completely cover the tumor. Indeed, for an intervention where the ablation margin would not have been sufficient, a recurrence of the tumor could be attributed to the type of treatment chosen, whereas the recurrence could be mainly due to the insufficiency of the ablation margin. The ablation margin is for example defined as the smallest distance between the periphery of the lesion and the periphery of the ablation region. It is generally appropriate that the ablation region completely covers the lesion, and that the ablation margin is at least equal to a threshold value, for example five millimeters.

[0070] [Fig.3] schematically represents the collection 200 of training elements used to train the machine learning algorithm.

[0071] To obtain a training element 21, it is first necessary to collect (step 201) a medical image 22 on which a lesion 23 is visible within the anatomy of interest 24 of another patient.

[0072] This may be, for example, another patient on whom an intervention has taken place in the past to ablate the lesion 23 with a particular treatment from among the various treatments available. Preferably, this intervention took place within the same medical establishment, but it could also have taken place in another establishment. However, it is appropriate that the treatment applied for the intervention corresponds to one of the treatments available. As indicated previously, however, it is not essential that an intervention aimed at treating the lesion has actually taken place.

[0073] Next, it is appropriate to identify (step 204) a particular treatment, from among the different treatments available, to treat the lesion 23. If an intervention has taken place, this is the treatment that has been chosen to ablate the lesion 23 (this choice may have been made, in particular, using the selection method according to the invention, or from a study of the medical image 22 by one or more medical experts). It may also be the treatment that is estimated as being the most appropriate by one or more medical experts from the medical image 22.

[0074] Finally, it is appropriate to determine (step 205) a confidence index for the chosen treatment. This confidence index can be determined according to the different methods mentioned previously. If an intervention has taken place, the confidence index can in particular be defined according to the observation or not of a recurrence following the treatment, according to the duration of the period elapsed between the end of the treatment and the current date without a recurrence having been observed, or according to the duration of the period elapsed between the end of the treatment and the recurrence, etc. The confidence index can also be estimated by one or more medical experts on the basis of the medical image (in particular in the case where there has not yet been an intervention on the lesion 23, or if it is not possible to obtain information on whether or not a recurrence occurs after the intervention).

[0075] The machine learning algorithm 20 may also be based on a set of parameters (metadata) relating to the lesion, the anatomy of interest or the patient. This makes it possible to strengthen the accuracy of the prediction made by the machine learning algorithm 20.

[0076] The parameters are for example determined manually by a user. Alternatively, or in addition, the parameters can be determined automatically on the medical image (for example via segmentation or image processing algorithms based on artificial intelligence).

[0077] These parameters may in particular correspond to physical or medical characteristics of the patient: the age, sex, weight and / or height of the patient, or the identification of certain comorbidities from which the patient suffers (heart failure, immunodeficiency, alcoholism, etc.). They may also correspond to a value representative of the size or volume of the lesion, to a distance between the lesion and a capsule of the anatomy of interest, or to a distance between the lesion and certain anatomical structures close to the lesion (blood vessel, bile duct, etc.). All these characteristics may in fact have an impact on the effectiveness or risks associated with a particular treatment (certain treatments are not indicated for large lesions, the presence of blood vessels close to the lesion impacts the effectiveness of thermal treatments, certain treatments are too risky for fragile patients, etc.).

[0078] Preferably, the same parameters used to create the training elements 21 are also used during the method 100 for selecting an optimal treatment. For this purpose, and as illustrated in FIGS. 1 and 3, the step 200 of collecting training elements 21 comprises, for each training element 21, a step 202 of associating a set of parameters with the medical image 22 corresponding to said training element 21. The method 100 also comprises a step 102 of associating a set of similar parameters with the medical image 12 of the patient to be treated. The sets of parameters associated respectively with the medical images 22 of the training elements 21, and the set of parameters associated with the medical image 12 of the patient to be treated are all similar to each other.This means that parameter sets all have parameters of the same type (however, parameter values ​​are usually different from one parameter set to another).

[0079] As illustrated in [Fig.3], the collection of a training element 21 may also comprise a transformation step 203 of the medical image 22 corresponding to said training element 21. This transformation 203 may correspond for example to a segmentation 25 of the lesion 23 on the medical image 22. According to other examples, the transformation 203 could also correspond to a segmentation of the anatomy of interest 24 and / or of particular anatomical structures on the medical image 22 (blood vessels, bile ducts, etc.). In the example illustrated in [Fig. 3], the transformed image 22' obtained comprises a segmentation 25 of the lesion. As illustrated in [Fig. 1], the method 100 for selecting an optimal treatment comprises a similar step of transformation 103 of the medical image 12 of the patient to be treated. In the example illustrated in [Fig. 1], the transformed image 12' obtained comprises a segmentation 15 of the lesion 13.

[0080] In particular embodiments, the transformation 103, 203 of a medical image 12, 22 comprises a cropping of the image 12, 22 around the lesion 13, 23 according to a frame whose dimensions are predetermined, for example 128x128 pixels. The dimensions of the frame are the same for all the medical images 22 of the training elements 21. However, and advantageously, the position of the lesion 23 relative to the frame varies from one medical image 22 to another. Such arrangements make it possible to reduce the prediction errors of the automatic learning algorithm 20 (it is indeed necessary to prevent the algorithm from learning that the lesion to be treated is mainly at the center of the medical image, which is not necessarily always true).

[0081] The step 102, 202 of associating a set of parameters and the step 103, 203 of transforming the medical image 12, 22 are optional (which is why they are shown in dotted lines in FIGS. 1 and 3). The machine learning algorithm 20 can in fact be based exclusively on the information contained in the medical images 12, 22. The step 102, 202 of associating a set of parameters and / or the step 103, 203 of transforming the medical image 12, 22, however, make it possible to provide additional information to the machine learning algorithm 20, thus improving the accuracy of the prediction.

[0082] Different types of machine learning algorithms are conceivable for implementing the selection method 100 according to the invention. The machine learning algorithm must predict a confidence index for each treatment available for ablating the lesion 13 visible on the medical image 12 acquired on the patient to be treated.

[0083] Let Q = {(Xi, T;, h) I i = 1 : N) be a set of training elements 21, where: • Xi = [xib xi2, ..., Xy, ..., xiK] is a point of a K-dimensional space corresponding to K characteristics extracted from the medical image 22 associated with the training element of index i (these parameters can possibly also be extracted from the transformed medical image 22' and / or from the set of parameters associated with the training element of index i), • T; is a particular treatment among the different treatments available, • li is the confidence index representative of a probability of success of the treatment T; for the ablation of the lesion visible on the medical image 22 associated with the training element of index i, • N is the number of elements in the training set considered.

[0084] The confidence index f corresponds for example to a score out of ten (for example the index h takes an integer value between 0 and 10). According to another example, the confidence index f can correspond to a standardized value between 0 and 1 representative of a probability of success of the treatment T).

[0085] If the value of N is sufficiently large (for example at least equal to one thousand, or even at least equal to five thousand), it is possible to train a machine learning algorithm to calculate a value I, for each available treatment T, from a set X of characteristics linked to a medical image of a patient to be treated (and possibly to the patient himself).

[0086] For this purpose, machine learning algorithms of the “neural network” type can be used. At least two different methods are possible. A first method consists of applying the different neurons to the different points Xi in order to predict a value f for a particular treatment T;. In this case, the characteristics xy are defined by hand (“hand-crafted” in English). However, this requires in-depth knowledge of the domain. A second method consists of delegating the choice of characteristics to the neural network. In this second method, the input of the neural network is the medical image 12 on which the lesion 13 to be treated is visible, and the characteristics are extracted by the convolution filters of the neural network defined during the training process. The use of additional characteristics (metadata) not forming part of the medical image 12 is however always possible.These additional features can be introduced in particular at the last layer of the neural network.

[0087] Machine learning algorithms such as "decision tree forest" can also be used. Classification trees are among the most popular algorithms in the field of machine learning and they are the basis for the most efficient methods. For example, in a tree structure, a node selects the variable xij that minimizes a classification error according to a certain rule, and a leaf represents a confidence index h for a particular treatment T;. The training set is used to define the variable xij and the specific rule to be used for each node. For example, the first node in the tree could correspond to the volume of the lesion and the rule "tumor volume less than 4 cm3". If the volume is less than 4 cm3, a second node could then for example correspond to a choice on the type of lesion (hepatocellular carcinoma, metastasis, etc.). The path through the nodes of the tree (corresponding to the verification of a set of rules) makes it possible to arrive at a prediction value for the confidence index h for a treatment T). Multiple classification trees can be created for the different treatments available by a random sampling process from the set Q of training elements 21. For a new value of X, each tree gives a value of 1 and the final prediction can then be made by a simple majority vote. Each tree is generally constructed independently of the others and has the same weight on the final decision.

[0088] According to another example, a machine learning algorithm of the "AdaBoost" type can be used (from the English "Adaptive Boosting"). The classic implementation of the AdaBoost method consists of a combination of weak classifiers (typically classification trees with a single node) where subsequent weak classifiers are adjusted to give more weight to samples misclassified by previous classifiers. Unlike decision tree forests, in the Ada-Boost method the weak classifiers are not independent of the others and they also do not have the same weights on the final classification.In the problem that concerns us, a first weak classifier would be made up for example of a single node corresponding to the volume of the lesion, and the following classifier, which would present for example a single node corresponding to the type of lesion, would be created to compensate for the classification errors of the first classifier.

[0089] According to another example, a machine learning algorithm of the “support vector machine” type can be used. This technique makes it possible, for example, to separate, in a binary manner, the points X corresponding to patients treated by a particular treatment with a first confidence index I (corresponding for example to a probability of success greater than a certain threshold), from the points X' corresponding to patients treated with the same treatment but with a different confidence index I' (corresponding for example to a probability of success lower than said threshold). For this, kernel functions are applied to the different points in order to make them separable by a hyper-plane in a higher-dimensional space.This technique can thus make it possible to predict a confidence index (informing about the probability of success in relation to a predetermined threshold) for a particular treatment (support vector machines are binary classifiers). In order to be able to manage a finer granularity of the confidence index and several possible treatments, the problem can be divided into several binary problems (an approach called in English "One-vs-One and One-vs-Rest").

[0090]

[0091]

[0092]

[0093]

[0094] In yet another example, a machine learning algorithm of the "k-means" type can be used. This is an unsupervised learning method that groups the training data X; into different groups ("clusters" in English) represented by their means (p varying for example from 1 to M, where M is the number of groups) which minimizes the sum S of the distances between the points X of a group and the associated mean: [Math.l] In the case that concerns us, for a given treatment, M corresponds for example to a number of possible values ​​taken by the confidence index (for example M = 11 if the confidence index corresponds to an integer value varying between 0 and 10). Once the algorithm is trained, the confidence index for a new patient represented by X is determined by looking for the average closest to X. Other types of machine learning algorithms could be considered, both classification algorithms and regression algorithms. The choice of a particular type of machine learning algorithm is only one variation of the invention. It may be of interest, for the practitioner who will carry out the treatment, to obtain a reference medical image corresponding to a particularly similar case having been previously treated with the same treatment as the selected optimal treatment. For this purpose, in a particular mode of implementation as illustrated in [Fig.4], the method 100 comprises a step 106 of calculating a similarity value for each training element 21 corresponding to the selected optimal treatment. The similarity value is representative of the similarity between the medical image 12 of the patient to be treated and the medical image 22 of the training element 21 considered. The method 100 then comprises a step 107 of selecting a reference training element from among the set of training elements 21 according to the calculated similarity values.For example, the reference element corresponds to the training element 21 having the highest similarity value. The medical image 22 associated with this reference element can then be used by the practitioner to compare the clinical case of the patient to be treated with the clinical case corresponding to the reference element. A high similarity value indicates that the lesion visible on the medical image of the reference element is similar to the lesion visible on the medical image of the patient to be treated. This can reassure the practitioner about the validity of the optimal treatment selected. The practitioner can also obtain information on the progress or results of a possible intervention having . took place to treat the injury of the case corresponding to the reference element.

[0095] It may happen that the benefit / risk ratio of carrying out an intervention on the patient to ablate the lesion is not sufficient (if the therapeutic benefits of the ablation are lower than the risks associated with its implementation). In such a case, it is preferable not to ablate the lesion. It is therefore conceivable that the option of not ablating the lesion is part of the available treatments. In other words, the optimal treatment selected by the method 100 may consist of not treating the lesion. This could also be the case, for example, if the element identified as a lesion by the practitioner on the medical image 12 of the patient is not in fact a lesion, or if it is a benign lesion.

[0096] For these reasons, it may be advantageous to train the machine learning algorithm 20 with medical images 22 on which there is no visible lesion in the anatomy of interest. Thus, in particular embodiments, the medical images 22 associated with some of the training elements 21 used to train the machine learning algorithm 20 do not present any visible lesion in the anatomy of interest.

[0097] The above description clearly illustrates that, through its various characteristics and their advantages, the present invention achieves the set objectives. In particular, the invention makes it possible to obtain a reliable and rapid prediction of an optimal treatment for the percutaneous ablation of a lesion in the anatomy of interest of a patient. Furthermore, the invention can be used by people who are not necessarily experts in percutaneous ablation; the proposed solution therefore allows a democratization of percutaneous ablation treatments.

Claims

Demands

1. An electronic device (30) comprising at least one processor (32) and a computer memory (31) storing program code instructions which, when executed by said processor (31), configure said processor (31) to implement a method (100) for selecting at least one optimal treatment from among several available treatments for the percutaneous ablation of a lesion (13) within an anatomy of interest (14) of a patient, said method (100) comprising: - obtaining (101) a medical image (12) previously acquired on the patient and on which the lesion (13) is visible within the anatomy of interest (14) of the patient, - a calculation (104) by a machine learning algorithm (20), from the medical image (12), for each of the different available ablation treatments, of a confidence index whose value is representative of a probability of success of said treatment for the ablation of the lesion (13) in the anatomy of interest (14) of the patient, the machine learning algorithm (20) having been previously trained from a set of training elements (21), each training element (21) comprising: • a medical image (22) on which a lesion (23) is visible within the anatomy of interest (24) of another patient, • identification of a treatment chosen from among the various ablation treatments available to treat said other patient, • a confidence index for said treatment on said other patient, - a selection (105) of at least one optimal treatment based on the confidence indices calculated for the different available treatments.

2. An electronic device (30) according to claim 1, wherein each available treatment is characterized by an ablation technology selected from radiofrequency, microwave, laser, cryotherapy, electro- poration or brachytherapy.

3. Electronic device (30) according to claim 2 wherein each available treatment is characterized by an energy deposition strategy by one or more applicators, said strategy being chosen from a centrifugal type energy deposition and a centripetal convergent type energy deposition.

4. Electronic device (30) according to claim 3 wherein each available treatment is characterized by a number of applicators, and / or by the positions of said applicators relative to the lesion.

5. Electronic device (30) according to any one of claims 1 to 4 wherein, for each training element (21), the confidence index is defined as a function of: - whether or not a relapse has been observed for said other patient with the chosen treatment, and / or - the duration of the period elapsed between the end of treatment and a current date without a relapse having been observed for said other patient, and / or - the duration of the period elapsed between the end of treatment and a relapse for said other patient, and / or - an estimate, by one or more medical experts, of the probability of success of the chosen treatment on said other patient.

6. Electronic device (30) according to any one of claims 1 to 5 wherein the method (100) comprises an association (102) of a set of parameters with the medical image (12) of the patient, said set of parameters comprising one or more parameters relating to the lesion (13), the anatomy of interest (14) and / or characteristics of the patient, each training element (21) comprising a similar set of parameters relating to the lesion (23), the anatomy of interest (24) and / or characteristics of the other patient for whom the medical image (22) corresponding to said reference element (21) was obtained.

7. An electronic device (30) according to claim 6, wherein the parameter set comprises one or more parameters selected from: - the age, sex, weight and / or height of the patient for whom the medical image (12) was obtained, - a comorbidity presented by the patient for whom the image medical (12) was obtained, - a representative value of the size and / or volume of the lesion (13), - a distance between the lesion (13, 23) and a capsule of the anatomy of interest (14), - a distance between the lesion (13) and a blood vessel close to the lesion, - when the anatomy of interest (14) is the liver, a distance between the lesion (13) and a bile duct close to the lesion.

8. Electronic device (30) according to any one of claims 1 to 7 wherein the method (100) comprises a transformation (103) of the medical image (12) of the patient, the medical image (22) of each training element (21) having undergone a similar transformation (203) before being used to train the machine learning algorithm (20).

9. Electronic device (30) according to claim 8 wherein the transformation (103) of the medical image (12) comprises: - a segmentation (15) of the lesion on the medical image (12), and / or - a segmentation of the anatomy of interest on the medical image (12), and / or - a segmentation of blood vessels on the medical image (12), and / or - when the anatomy of interest (14) is the liver, a segmentation of bile ducts on the medical image (12).

10. Electronic device (30) according to any one of claims 8 or 9 wherein the transformation (103) of the medical image (12) includes a cropping of the image (12) around the lesion (13) according to a frame whose dimensions are predetermined, said frame being common to all medical images (22) of the drive elements (21), the position of the lesion (23) relative to the frame being variable from one medical image (22) to another.

11. Electronic device (30) according to any one of claims 1 to 10 wherein the method (100) comprises: - for each training element (21) corresponding to the selected optimal treatment, a calculation (106) of a similarity value representative of the similarity between the medical image (12) of the patient to be treated and the medical image (22) of the training element (21), and - a selection (107) of a reference training element from among the set of training elements (21) according to the calculated similarity values.

12. Electronic device (30) according to any one of claims 1 to 11 for which medical images (12, 22) were acquired by computed tomography, magnetic resonance imaging or ultrasound.

13. Electronic device (30) according to any one of claims 1 to 12 wherein the medical images (12, 22) are three-dimensional.

14. Electronic device (30) according to any one of claims 1 to 13 wherein the anatomy of interest (14, 24) is the liver, a kidney or a lung, and the lesion (13, 23) is a tumor or a cyst.

15. Electronic device (30) according to any one of claims 1 to 14 wherein the machine learning algorithm (20) is further pre-trained with training elements whose medical images (22) do not show any lesion in the anatomy of interest.

Images