A device that predicts the best procedure type for percutaneous ablation of a lesion

The electronic device employs a machine learning algorithm to determine the optimal treatment for percutaneous ablation by analyzing medical images, addressing the challenge of selecting the best procedure and improving treatment accuracy and reducing recurrence.

JP2024540787AInactive Publication Date: 2024-11-06QUANTUM SURGICAL
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
JP2023572568
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2021-10-07
Filing Date
2022-10-03
Publication Date
2024-11-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Current methods lack a satisfactory solution for selecting the optimal treatment type among available procedures for percutaneous ablation of lesions in a patient's target anatomy, leading to potential suboptimal procedures and recurrence of pathology or collateral lesions.

Method used

An electronic device using a machine learning algorithm to determine a confidence index for each available ablation procedure based on medical images, pre-trained from a set of training elements, allowing selection of the optimal treatment with the highest probability of success.

Benefits of technology

Provides a reliable and rapid prediction of the optimal treatment for percutaneous ablation, enabling non-experts to perform procedures accurately and reducing the risk of recurrence.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024540787000001_ABST
    Figure 2024540787000001_ABST
Patent Text Reader

Abstract

The present invention relates to a device for performing a method (100) of selecting an optimal procedure for percutaneously ablating a lesion (13) in a target anatomical structure (14) of a patient. The method uses a machine learning algorithm (20) trained to calculate, from a medical image (12) in which the lesion (13) is visible, for each of a plurality of available procedures, a confidence score having a value representative of the likelihood that said procedure will be successful for ablation of the lesion. The machine learning algorithm is pre-trained using a set of training elements (21) each including a medical image in which the lesion is visible in the target anatomical structure of another patient, a procedure selected from a variety of procedures available for treating said other patient, and a confidence score of the selected procedure for said other patient. An optimal procedure is then selected based on the calculated confidence scores for the different available procedures.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] FIELD OF THEINVENTION This patent application is in the technical field of scheduling minimally invasive medical interventions, and in particular relates to a computer-based electronic device that implements a method for selecting the most suitable procedure from among several available procedures for the percutaneous ablation of a lesion in a target anatomical structure of a patient. [Background technology]

[0002] current level To prepare a medical intervention aimed at treating a lesion (e.g., a tumor) in a patient's target anatomical structure (e.g., lung, kidney, liver, bone structure, etc.), a practitioner typically performs scheduling of the intervention based on medical images.

[0003] The essence of this first step of scheduling is to select the treatment type. There are several major clinical practice guidelines for selecting the treatment according to the type of pathology. The aim is, for example, to first determine whether it is best to use ablation, implantation, chemoembolization treatment, or radiation therapy as a means.

[0004] However, appropriately selecting a more precise treatment type remains difficult.

[0005] More precisely, in the case of percutaneous ablation of tumors, different treatment methods can be applied: radiofrequency, microwave, laser, cryo, electroporation or brachytherapy. Methods that resort to radiofrequency, microwave, laser or cryo are thermal methods. The essence of electroporation is the delivery of tens of radiofrequency pulses of a few μs at very high intensity between two electrodes, causing irreversible modification of the membrane function of the cells. Electroporation is a relatively "gentle" ablation technique, which allows the consideration of the treatment of tumors that could not be treated before with thermal ablation techniques, either due to the location of the tumor or the frailty of the patient. As a brachytherapy, the aim is to deliver a radiation dose inside or near the tumor.

[0006] Whatever the technique used (thermal energy, electrical energy, radiofrequency energy, etc.), percutaneous ablation involves the placement at a predetermined location of one or more applicators for the in situ delivery of energy. The delivered energy induces irreversible degradation on a cellular and tissue level scale. In some cases, it may be indicated that it is necessary to perform overlapping ablations, either sequentially by the successive insertion of several applicators or simultaneously by the simultaneous insertion and activation of several applicators.

[0007] Furthermore, when several applicators are activated simultaneously, there are different strategies for depositing energy by the applicators, in particular a distinction being made between so-called "centripetal" and so-called "efferent" ablation.

[0008] The efferent ablation strategy is currently the most widely used, since it is relatively easy to implement. However, the centripetal convergent ablation strategy offers a better adaptability to the ablation zone and its environment. The centripetal convergent ablation strategy actually allows a better control of the destruction limit, thereby allowing a specific "modeling" of the ablation zone as a function of the morphology and location of the tumor. On the other hand, it entails a greater sacrifice of healthy tissue.

[0009] These various methods, which may appear conceptually very similar, are in reality quite different in terms of practical application as well as the results they make possible to obtain.

[0010] The choice of a particular treatment type will depend largely on the specific characteristics of each clinical situation, with each particular treatment type offering the best benefit / risk ratio in that particular clinical situation.

[0011] The selection of a particular procedure type is generally made by the practitioner according to the practitioner's practice or preference, and therefore the procedure selected to treat a lesion in a patient's target anatomy may not be optimal and may lead to recurrence of the patient's condition or side branch disease. Summary of the Invention [Problem to be solved by the invention]

[0012] Thus, currently, there is no satisfactory solution that assists a practitioner in selecting the optimal procedure from among a set of procedures available for ablation of a lesion in a patient's target anatomy, however.

[0013] Summary of the Invention The object of the present invention is to remedy in whole or in part the drawbacks of the prior art, in particular those mentioned above. [Means for solving the problem]

[0014] To this end, according to a first aspect, an electronic device is proposed, comprising at least one processor and a computer memory, the computer memory storing program code instructions which, when executed by the processor, configure the processor to implement a method for selecting at least one optimal procedure among several procedures available for percutaneous ablation of a lesion in a target anatomical structure of a patient, the method being implemented comprising the steps of: - obtaining a medical image in which a lesion is visible within a target anatomical structure of a patient; - calculating, for each of the different available ablation procedures, from the medical images using a machine learning algorithm, a confidence index having a value representative of the probability of success of said procedure in ablation of a lesion in a target anatomical structure of a patient, the machine learning algorithm being pre-trained from a set of training elements, each of the training elements being: Medical imaging where the lesion can be seen within the target anatomy of another patient; - identification of the procedure selected from among different ablation procedures available for said separate patient procedure; - Confidence index of the above procedure for the above different patients and - selecting at least one optimal treatment as a function of the confidence indexes calculated for the different available treatments; Includes.

[0015] "Percutaneous ablation" is understood to mean a minimally invasive intervention through a patient's skin to treat a lesion in a targeted anatomical structure.

[0016] The anatomical structure of interest is, for example, the liver, lungs, kidneys, or bone structures. The lesion is, for example, a tumor, a cyst, or an aneurysm. Various procedures can be envisaged to ablate the lesion in the anatomical structure of interest. The procedures can be characterised in particular by the choice of technique (radiofrequency, microwave, laser, cryo, electroporation, etc.) and the method of application (for each technique, different energy deposition strategies can be envisaged).

[0017] The machine learning algorithm takes as input a medical image and a treatment type, and directly provides as output a confidence index corresponding to a value representative of the success probability of said treatment. Thus, the machine learning algorithm allows to calculate a confidence index for each available treatment from a medical image of the patient's target anatomical structure. The optimal treatment can then be selected as a function of the duly calculated confidence index. For example, the treatment with the highest confidence index corresponds to the treatment with the highest probability of treatment success. The confidence index can in particular represent the risk of recurrence. In some cases, there may be several optimal treatment candidates, for example when different available treatments are associated with the same value of the confidence index.

[0018] The machine learning algorithm is pre-trained based on a set of training elements, each associated with an intervention previously performed on the target anatomical structure of another patient, for example according to one of the available procedures, where the training elements include a medical image of the target anatomical structure before the procedure, an identification of the procedure performed on the patient, and a confidence index of this procedure (the confidence index is defined, for example, as a function of whether a recurrence was observed after the procedure or not).

[0019] However, the training elements can also be defined without the intervention actually being performed on another patient. In fact, the confidence index of a procedure considered to treat a pathology visible in a medical image of the patient's target anatomical structure can be theoretically specified. For example, the confidence index can be estimated by a medical expert organization based on the medical image. This estimation can possibly be weighted by recommendations from scientific publications and / or empirical studies performed on patient groups with similar pathologies.

[0020] The set of training elements corresponds to a database that allows training a machine learning algorithm to calculate a confidence index for a particular procedure from a medical image showing the pathology to be treated. Healthcare institutions (hospitals, clinics, etc.) can create and enrich their own database of procedures available within said medical facility (for each new intervention performed in this facility, a new training element can be added to the database).

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

[0022] To implement the invention, different types of machine learning algorithms can be considered: decision tree forests ("random forests" in the literature), neural networks, support vector machines (SVM), k-means partitioning, etc.

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

[0024] In certain embodiments, the device may further comprise one or more of the following features, either alone or in all technically possible combinations:

[0025] In certain embodiments, each available treatment is characterized by an ablation technique selected from radiofrequency, microwave, laser, cryo, electroporation, or brachytherapy.

[0026] In certain embodiments, each available treatment is characterized by an energy deposition strategy with one or more applicators, said strategy being chosen among centrifugal energy deposition and centripetal focused energy deposition.

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

[0028] In certain embodiments, for each training element, the confidence index is: - recurrence or no recurrence was observed in the other patient with the selected treatment, and / or - The duration of time elapsed between the end of treatment and the current date without observed recurrence in the other patient; and / or - the duration of the period that has elapsed between the end of treatment and the recurrence of the other patient; and / or - an estimate by one or more medical professionals of the probability of success of the procedure chosen for that other patient; It is defined as a function of

[0029] In certain embodiments, the method includes associating a set of parameters with a medical image of a patient, the set of parameters including one or more parameters related to a lesion, an anatomical structure of interest, and / or a characteristic of the patient, and each training element includes a similar set of parameters related to a lesion, an anatomical structure of interest, and / or a characteristic of another patient from whom a medical image corresponding to the reference element was obtained.

[0030] In a particular embodiment, the set of parameters is: - the age, sex, weight, and / or height of the patient from whom the medical image was obtained; - comorbidities presented by the patient for whom the medical images were obtained, - a value representing the size and / or volume of the lesion; - the distance between the lesion and the capsule of the targeted anatomical structure, - the distance between the lesion and the blood vessel close to the lesion, - If the anatomical structure of interest is the liver, the distance between the lesion and the hepatic duct closest to the lesion The parameter includes one or more parameters selected from the following:

[0031] In certain embodiments, the method includes transforming medical images of the patient, where the medical images of each training element undergo a similar transformation before being used to train the machine learning algorithm.

[0032] In certain embodiments, transforming the medical image comprises: - segmenting lesions in medical images; and / or - Segmenting anatomical structures of interest in medical images; and / or - segmenting blood vessels in medical images; and / or Segmenting the hepatic ducts in medical images, where the anatomical structure of interest is the liver Includes.

[0033] In a particular embodiment, transforming the medical image includes reframing the image around the lesion according to a frame of predetermined dimensions, said frame being common to all of the medical images of the training element, and the position of the lesion relative to the frame being variable from one medical image to another.

[0034] In certain embodiments, the method comprises: - calculating, for each training element corresponding to the selected optimal treatment, 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; - selecting a reference training element from among the set of training elements as a function of the calculated similarity value; Includes.

[0035] In certain embodiments, the medical images were obtained by tomographic densitometry, magnetic resonance imaging, or ultrasound.

[0036] In certain embodiments, the medical image is three-dimensional.

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

[0038] In a particular embodiment, the set of training elements includes training elements in which the medical images do not show any pathology in the anatomical structure of interest.

[0039] Text description of the illustration image024.gif. The invention will be better understood on reading the following description, given by way of non-limiting example, and with reference to figures 1 to 4. [Brief description of the drawings]

[0040] [Figure 1] 1 is a schematic representation of the main steps of a method for selecting an optimal procedure for ablation of a lesion in a target anatomical structure of a patient. [Diagram 2]1 is a schematic representation of an electronic device that allows for the selection of an optimal procedure for ablation of a lesion in a target anatomical structure of a patient. [Diagram 3] 1 is a schematic representation of the steps of collecting a set of reference elements used to train a machine learning algorithm. [Figure 4] 1 is a schematic representation of a particular embodiment of a method for providing optimal treatment, in which a reference training element is selected. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0041] In these figures, like references indicate the same or similar elements throughout the figures, and for clarity, unless otherwise noted, elements depicted are not necessarily to scale.

[0042] Detailed Description of the Invention Embodiments 1 illustrates in schematic form the main steps of a method 100 for selecting an optimal procedure for ablation of a lesion in a target anatomical structure of a patient. The optimal procedure is selected from among several available procedures. It should be noted that in some cases, several procedures may be indicated as optimal by the method 100 without being able to determine which of them is optimal.

[0043] The method 100 may be implemented by electronic devices such as computers, tablets, mobile phones, and other such devices. Figure 2 shows a schematic of an exemplary embodiment of such an electronic device 30. The electronic device 30 comprises at least one processor 32 and a computer memory 31 having stored therein program code instructions which, when executed by the processor 32, configure the processor 32 to implement the method 100 for selecting an action according to the present invention.

[0044] Thus, the computer memory 31 corresponds to a storage medium containing instructions that can be read by the electronic device 30 and that, when executed by said electronic device 30, cause the electronic device 30 to perform the method 100 of selecting a treatment.

[0045] The invention may also take the form of a computer program comprising instructions that, when executed by the electronic device 30, cause the electronic device 30 to perform the method 100 of selecting a treatment.

[0046] The method 100 includes a step 101 of acquiring a medical image 12 in which a lesion 13 is visible within an anatomical structure of interest 14 of a patient.

[0047] In the example considered, the medical image 12 is a two-dimensional medical image acquired by tomographic densitometry ("CT scan" for "Computed Tomography Scan" in the literature). However, nothing prevents the medical image from being acquired by another medical imaging modality, for example MRI ("Magnetic Resonance Imaging") or ultrasound. Also, nothing prevents the medical image from being a three-dimensional image.

[0048] The medical images 12 are for example stored in a computer memory 31 of the electronic device 30 and processed by a processor 32. The medical images 12 are for example transmitted to the electronic device 30 via wired or wireless communication means (these means are not represented in FIG. 2). According to another example, the electronic device 30 can be connected to an external memory peripheral, for example a USB (acronym for "Universal Serial Bus") stick, on which the medical images 12 are stored. In any case, the electronic device 30 is configured to acquire medical images 12 acquired before the patient to be treated.

[0049] In the considered example, without limitation, the anatomical structure of interest is the liver and the lesion is a cancerous tumor, however, the present invention may also be applied to other anatomical structures of interest (e.g., lung, artificial, bone structures, etc.) and / or other types of lesions (e.g., cysts, aneurysms, metastases, etc.).

[0050] The available procedures from which the at least one optimal procedure is selected may correspond, for example, to a plurality of different procedures used in a medical facility to treat such a patient.

[0051] The available treatments can be characterized in particular by ablation techniques chosen from radiofrequency, microwave, laser, cryo, electroporation or brachytherapy, the various techniques being known to those skilled in the art of treating cancerous tumors of the liver.

[0052] The available treatments can further be characterized by an energy deposition strategy by one or more applicators (needles, electrodes, probes, etc.) The energy deposition strategy can be chosen in particular between centrifugal energy deposition and centripetal energy deposition.

[0053] For a long time, the preferred energy deposition strategy has followed the centrifugal coverage method, where the energy is delivered isotropically from the center of the tumor (where the applicator is inserted) to the tumor periphery. In some cases, it may be shown that it is necessary to perform overlapping ablation either sequentially by several successive activations of applicators inserted at different positions, or simultaneously by simultaneous activation of several applicators at different positions. The centrifugal deposition strategy offers the advantage of being a particularly convenient and proven technique. However, in some specific situations (particularly as a function of the size or volume of the tumor or as a function of the anatomical structures close to the lesion), the centrifugal strategy is not always optimal. In that case, another energy deposition strategy (centripetal focusing strategy), whose essence is to focus the energy from the tumor periphery to the center, may be preferred.

[0054] In addition to the ablation technique and energy deposition strategy, available treatments can be further characterized by the number of applicators used and / or the location of said applicators relative to the tumor.

[0055] The location of the applicator is generally specific to the tumor, but can also be specific to the tumor's environment (ie, anatomical structures close to the tumor: bones, blood vessels, organs at risk, etc.).

[0056] The method 100 comprises the step of calculating 104, for each of the different available ablation procedures, a confidence index having a value representative of the success probability of said procedure in ablating a lesion 13 in the patient's target anatomical structure 14. The confidence indexes of the different available procedures are calculated from the medical images 12 by a pre-trained machine learning algorithm 20.

[0057] As shown in Fig. 1, during a training step 206, the machine learning algorithm is trained from a set of training elements 21. Each training element 21 is associated with an intervention previously performed to treat a lesion in a target anatomical structure of another patient, for example according to one of the available procedures. In that case, the training element 21 comprises a medical image of the target anatomical structure of said other patient before the procedure, an identity of the procedure performed on said other patient, and a confidence index of this procedure.

[0058] For each training element 21, a confidence index can be defined, for example, as a function of whether or not a relapse has been observed in the other patient with the selected treatment (if there is no relapse, the probability of success of the treatment is relatively high, and conversely, if there is a relapse, the probability of success of the treatment is low).

[0059] The duration of the period that has elapsed between the end of treatment and the current date, without any recurrence being observed, can also be a factor used to define a confidence index for the treatment (the longer this period, the higher the probability of success of the treatment).

[0060] In case of recurrence, it is possible to take into account the duration of the period that has elapsed between the end of treatment and the recurrence in order to identify a confidence index (the shorter this period, the lower the confidence index).

[0061] It should be noted that step 200 of collecting training elements and step 206 of training the machine learning algorithm are performed prior to (and therefore do not form part of) method 100 of selecting an optimal treatment.

[0062] It is also noted that the training element 21 may be acquired without the intervention actually being performed on another patient. Indeed, a procedural confidence index considered for the treatment of a pathology visible on a medical image of the target anatomical structure of another patient can theoretically be specified by a medical expert organization.

[0063] Once the confidence indices of the different available treatments have been calculated by the machine learning algorithm, it is possible to select at least one optimal treatment from the different available treatments during a selection step 105. For example, the optimal treatment corresponds to the treatment for which the calculated confidence index is the highest. As indicated above, in some cases, there may be several optimal treatment candidates, for example when the different available treatments are associated with the same value of confidence index.

[0064] The set of training elements 21 corresponds to a database used to train the machine learning algorithm to calculate a confidence index for a particular treatment from medical images showing the lesion to be treated. This database can be enriched over time as new medical images become available showing a lesion in the patient's target anatomical structure and as confidence indices for a particular treatment for this lesion are identified.

[0065] It is possible to consider the application of the selection to cases likely to be used for forming the training element 21. In particular, if the training element 21 is built on interventions that have actually been performed, it is advantageous to consider only those interventions for which the ablation margin was sufficient to cover the tumor entirely. In fact, in the case of interventions for which the ablation margin would not have been sufficient, there is a risk that the cause of the tumor recurrence will be attributed to the selected procedure type, whereas the recurrence may have been mainly due to an insufficient ablation margin. The ablation margin is defined, for example, as the minimum distance between the periphery of the lesion and the periphery of the ablation area. In general, it is appropriate that the ablation area covers the lesion entirely and that the ablation margin is at least equal to a threshold value, for example 5 millimeters.

[0066] FIG. 3 illustrates a schematic representation of a collection 200 of training elements used to train a machine learning algorithm.

[0067] To obtain the training elements 21, the first step should be to collect (step 201) medical images 22 in which a lesion 23 is visible within the anatomical structure of interest 24 of another patient.

[0068] It can be, for example, another patient on whom an intervention has been performed in the past, using a specific procedure from among the different procedures available, to ablate the lesion 23. Preferably, this intervention has been performed in the same medical facility, but it can also be performed in another facility. However, the procedure applied for the intervention should correspond to one of the available procedures. However, as indicated above, it is not essential that an intervention aimed at treating the lesion is actually performed.

[0069] Next, a specific procedure has to be identified (step 204) among the different procedures available to treat the lesion 23. If an intervention is performed, it is the procedure chosen for ablation of the lesion 23 (this selection could in particular have been made using a selection method according to the invention or could have been made from a study of the medical images 22 by one or more medical experts). It can also be the procedure estimated to be the most appropriate by one or more medical experts on the basis of the medical images 22.

[0070] Finally, a confidence index of the chosen treatment should be identified (step 205). This confidence index can be identified according to the different methods described above. If an intervention has been performed, the confidence index can be defined in particular as a function of whether or not a recurrence has been observed after the treatment, as a function of the duration of the period that has elapsed between the end of the treatment and the current date without a recurrence being observed, or as a function of the duration of the period that has elapsed between the end of the treatment and a recurrence, etc. The confidence index may be estimated by one or more medical experts on the basis of medical images (in particular when no intervention has yet been performed on the lesion 23 or when it is not possible to obtain information on the occurrence or non-occurrence of a recurrence after the intervention).

[0071] The machine learning algorithm 20 may be based on a set of parameters (metadata) related to the lesion, the anatomical structure of interest, or the patient, which allows to reinforce the accuracy of the predictions made by the machine learning algorithm 20.

[0072] The parameters may be specified manually, for example, by a user. Alternatively or additionally, the parameters may be specified automatically in the medical image (for example, via image segmentation or artificial intelligence based processing algorithms).

[0073] These parameters may correspond in particular to the physical or medical characteristics of the patient: the patient's age, sex, weight, and / or height, or the identity of certain comorbidities the patient has (heart failure, immunodeficiency, alcoholism, etc.). The parameters may also correspond to values ​​representative of the size or volume of the lesion, the distance between the lesion and the capsule of the anatomical structure of interest, or the distance between the lesion and certain anatomical structures close to the lesion (blood vessels, hepatic ducts, etc.). All these characteristics may in fact have an effect on the efficacy or risk associated with a particular treatment (certain treatments are not indicated for large sized lesions, the presence of blood vessels near the lesion affects the efficacy of thermal treatments, some treatments are too risky for vulnerable patients, etc.).

[0074] Preferably, the same parameters used for creating the training elements 21 are also used during the method 100 for selecting the optimal treatment. For this, as shown in figures 1 and 3, the step 200 of collecting the 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 respectively associated with the medical images 22 of the set of training elements 21 and the sets of parameters associated with the medical images 12 of the patient to be treated are all similar to each other. This means that the sets of parameters all comprise the same type of parameters (however, the values ​​of the parameters are generally different for each set of parameters).

[0075] As shown in Fig. 3, collecting the training elements 21 may also include a step 203 of transforming a medical image 22 corresponding to said training elements 21. This transformation 203 may for example correspond to a segmentation 25 of a lesion 23 in the medical image 22. According to another example, the transformation 203 may also correspond to a segmentation of the anatomical structure of interest 24 and / or of a particular anatomical structure (blood vessels, hepatic ducts, etc.) in the medical image 22. In the example shown in Fig. 3, the acquired transformed image 22' includes a segmentation 25 of the lesion. As shown in Fig. 1, the method 100 for selecting an optimal treatment includes a similar step 103 of transforming a medical image 12 of a patient to be treated. In the example shown in Fig. 1, the acquired transformed image 12' includes a segmentation 15 of the lesion 13.

[0076] In a particular embodiment, the transformation 103, 203 of the medical images 12, 22 comprises reframing the images 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 medical images 22 of the training element 21. However, advantageously, the position of the lesion 23 relative to the frame differs from one medical image 22 to another. Such preparation makes it possible to reduce the prediction error of the machine learning algorithm 20 (it is important to avoid making the algorithm learn that the lesion to be treated is mainly in the center of the medical image, which is not always the case in practice).

[0077] The steps 102, 202 of associating sets of parameters and the steps 103, 203 of transforming the medical images 12, 22 are optional (which is why they are represented by dotted lines in Figures 1 and 3). The machine learning algorithm 20 can in fact be based exclusively on the information contained in the medical images 12, 22. However, the steps 102, 202 of associating sets of parameters and / or the steps 103, 203 of transforming the medical images 12, 22 allow additional information to be provided to the machine learning algorithm 20, thus enhancing the accuracy of the predictions.

[0078] Different types of machine learning algorithms can be considered for the implementation of the selection method 100 according to the invention. The machine learning algorithm must predict a confidence index for each treatment available to ablate a lesion 13 visible in a medical image 12 acquired on the patient to be treated.

[0079] Ω={(X i ,T i ,I i )|i=1:N) be the set of training elements 21, where X i =[x i1 ,x i2 ,x ij ,x iK ] is a point in a space having K dimensions corresponding to K features extracted from the medical image 22 associated with the training element of index i (these parameters may possibly be extracted from the transformed medical image 22′ and / or from the set of parameters associated with the training element of index i), T i is a particular treatment among different treatments available, I i is the treatment T for ablation of a lesion visible in the medical image 22 associated with the training element of index i i is a confidence index that represents the probability of success, · N is the number of elements in the set of training elements considered.

[0080] Trust Index I i corresponds to a score out of 10 (e.g., the index I i (T takes integer values ​​from 0 to 10). In another example, the reliability index I i is the treatment T i can correspond to a normalized value between 0 and 1 that represents the probability of success.

[0081] If the value of N is sufficiently large (for example at least equal to 1000, or even at least equal to 5000), it is possible to train a machine learning algorithm to calculate, for each available treatment T, a value I from a set of features X linked to medical images of the patient to be treated (and possibly the patient himself).

[0082] For this purpose, machine learning algorithms of the "neural network" type can be used. At least two different methods can be considered. The essence of the first method is to apply different neurons to different points Xi to generate a certain treatment T i Value of I i In this case, the goal is to predict the feature x ij are defined by hand (or "handcrafted"); however, a deep knowledge of the domain is required. The essence of the alternative method is to delegate the selection of the features to a neural network. In this second method, the input of the neural network is a medical image 12 in which the lesion 13 to be treated is visible, and the features are extracted by the convolution filters of the defined neural network during the training process. However, the use of additional features (metadata) that are not part of the medical image 12 is still possible. These additional features can be introduced in particular in the last layer of the neural network.

[0083] One can also use machine learning algorithms of the "random forest" type. Classification trees are some of the most popular algorithms in the field of machine learning and are the basis of the most powerful methods. For example, in a tree structure, the nodes select the variables xij that minimize the classification error according to certain rules, and the leaves ... i Confidence Index I i A set of training elements is used to define the variables xij and the specific rule to be used at each node. For example, the first node in the tree represents the volume of the lesion and the rule "4 cm 3 The "tumor volume less than 4cm" rule can be applied. 3If the first node corresponds to the type of lesion (hepatocellular carcinoma, metastasis, etc.), then the second node corresponds to the type of lesion (hepatocellular carcinoma, metastasis, etc.). The path of the nodes of the tree (corresponding to the validation of the rule set) is i Confidence Index I i It is possible to arrive at a prediction value of I. Multiple classification trees can be created for the different treatments available by a random sampling process based on the set Ω of training elements 21. For a new value of X, each tree contributes a value of I, and then a final prediction can be made by simple majority voting. Each tree is typically built independently of the other trees and has equal weight in the final decision.

[0084] According to another example, machine learning algorithms of the "AdaBoost" type can be used (short for the term "adaptive boost"). The essence of conventional implementations of the AdaBoost method is the combination of low classifiers (typically classification trees with a single node) in which the subsequent low classifiers are adjusted to give more weight to samples that are poorly classified by the previous classifier. Unlike random forests, in the AdaBoost method, the low classifiers are not independent of the other classifiers and do not have the same weight in the final classification. In the problem addressed here, the first low classifier is composed of a single node, for example corresponding to the volume of the lesion, and the next classifier, with a single node corresponding to the type of lesion, is created to correct the classification error of the first classifier.

[0085] According to another example, a machine learning algorithm of the "support vector machine" type can be used. This technique allows, for example, in a binary way, to separate points X corresponding to patients treated with a certain treatment with a first confidence index I (e.g. corresponding to a success probability greater than a certain threshold) from points X' corresponding to patients treated with the same treatment but with a different confidence index I' (e.g. corresponding to a success probability less than said threshold). To do so, a kernel function is applied to the different points to allow separation by a hyperplane in a higher dimensional space. This technique therefore allows to predict the confidence index of a certain treatment (informing the success probability with respect to a given threshold) (support vector machines are binary classifiers). To be able to manage a finer grain of confidence indexes and several treatment candidates, the problem can be subdivided into several binary problems (a technique called "one-to-one and one-to-rest").

[0086] According to yet another example, a machine learning algorithm of the "k-means" type can be used, which is the mean associated with a point X of a group.

number

number

number

[0087] In the case considered here, for a given treatment, M corresponds, for example, to the number of values ​​that the confidence index can take (e.g., if the confidence index corresponds to integer values ​​varying from 0 to 10, then M=11). Once the algorithm is trained, it finds the confidence index of a new patient, denoted X, by the average confidence index closest to X.

number

[0088] Other types of machine learning algorithms are also contemplated, both classification and regression algorithms. The selection of a particular type of machine learning algorithm is merely a variation of the present invention.

[0089] It may be advantageous for the practitioner who will carry out the procedure to obtain, in particular, reference medical images corresponding to similar cases previously treated with the same procedure as the selected optimal procedure. For this purpose, in the particular embodiment shown 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 procedure. The similarity value represents 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 the set of training elements 21 as a function of the calculated similarity value. For example, the reference element corresponds to the training element 21 with 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 lesions visible in the medical image of the reference element are similar to the lesions visible in the medical image of the patient to be treated. This may allow the practitioner to be reassured about the validity of the selected optimal procedure. The practitioner may also take information about the progress or outcome of a candidate intervention carried out to treat the lesion of the case corresponding to the reference element.

[0090] There may be cases where the benefit / risk ratio of performing an intervention on a patient to ablate the lesion is not sufficient (when the beneficial therapeutic effect of ablation is lower than the risk linked to its implementation). In such cases, it is preferable not to ablate the lesion. Therefore, the option of not ablating the lesion can be considered as part of the treatment candidates. In other words, the optimal treatment selected by the method 100 may be to not treat the lesion. For example, an element identified as a lesion by the practitioner in the patient's medical image 12 may not actually be a lesion or may be a benign lesion.

[0091] For these reasons, it may be advantageous to train the machine learning algorithm 20 with medical images 22 that do not show pathology in the anatomical structures of interest. Thus, in certain embodiments, the medical images 22 associated with some of the training elements 21 used to train the machine learning algorithm 20 do not show any pathology visible in the anatomical structures of interest.

[0092] The above description clearly shows that the invention achieves the set objectives through its various features and advantages. In particular, the invention makes it possible to obtain a reliable and rapid prediction of the optimal treatment for percutaneous ablation of a lesion in a target anatomical structure of a patient. Furthermore, the invention can be used by people who are not necessarily experts in percutaneous ablation, thus making the proposed solution accessible to anyone for treatment by percutaneous ablation.

Claims

1. 1. An electronic device (30) comprising at least one processor (32) and a computer memory (31) storing program code instructions which, when executed by the processor (31), configure the processor (31) to perform a method (100) of selecting at least one optimal procedure from among several available procedures for percutaneous ablation of a lesion (13) in a target anatomical structure (14) of a patient, the method (100) comprising: - acquiring (101) a medical image (12) previously acquired of said patient, in which said lesion (13) is visible within said anatomical structure of interest (14) of said patient; - calculating (104) for each of the different available ablation procedures from the medical image (12) using a machine learning algorithm (20) having a confidence index having a value representative of the probability of success of the procedure with the ablation of the lesion (13) in the target anatomical structure (14) of the patient, the machine learning algorithm (20) having been pre-trained from a set of training elements (21), each of the training elements (21) comprising: a medical image (22) in which a lesion (23) can be seen within said anatomical structure of interest (24) of another patient; - an identity of a procedure selected from among the different ablation procedures available for treatment of the other patient; A confidence index for the treatment for the other patient. Calculating (104), - selecting (105) at least one optimal treatment as a function of the confidence indices calculated for the different available treatments; An electronic device (30).

2. 10. The electronic device (30) of claim 1, wherein each available treatment is characterized by an ablation technique selected from radiofrequency, microwave, laser, cryo, electroporation, or brachytherapy.

3. 3. The electronic device (30) of claim 2, wherein each available treatment is characterized by an energy deposition strategy with one or more applicators, said strategies being chosen from centrifugal energy deposition and centripetal focused energy deposition.

4. The electronic device (30) of claim 3, wherein each available treatment is characterized by a number of applicators and / or a position of the applicators relative to the lesion.

5. For each training element (21), the confidence index is - whether a recurrence was or was not observed in said other patient with said selected treatment, and / or - the duration of the period that has elapsed between the end of said treatment and the current date, without any recurrence observed in said other patient, and / or - the duration of the period that elapsed between the end of the treatment and the recurrence of the other patient, and / or - an estimate by one or more medical professionals of the probability of success of the treatment chosen for said other patient; The electronic device (30) of any one of claims 1 to 4, wherein the θ is defined as a function of

6. The method (100) comprises: Associating (102) a set of parameters with the medical image (12) of the patient, the set of parameters including one or more parameters related to the lesion (13), the anatomical structure of interest (14) and / or features of the patient, each training element (21) including a similar set of parameters related to the lesion (23), the anatomical structure of interest (24) and / or features of another patient from whom the medical image (22) corresponding to the reference element (21) was acquired; The electronic device (30) of any one of claims 1 to 5, comprising:

7. The set of parameters is - the age, sex, weight and / or height of the patient from whom the medical image (12) was acquired, - the comorbidities presented by the patient from whom the medical image (12) was obtained, - a value representative of the size and / or volume of said lesion (13), the distance between the lesion (13, 23) and the capsule of the anatomical structure of interest (14), the distance between said lesion (13) and a blood vessel close to said lesion, - if the anatomical structure of interest (14) is the liver, the distance between the lesion (13) and the hepatic duct closest to the lesion; The electronic device (30) of claim 6, further comprising one or more parameters selected from the group consisting of:

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

9. Transforming (103) the medical image (12) comprises: - segmenting (15) said lesion in said medical image (12), and / or - segmenting said anatomical structure of interest in said medical image (12), and / or - segmenting blood vessels in said medical image (12), and / or - segmenting the hepatic ducts in said medical image (12) if said anatomical structure of interest (14) is the liver; The electronic device (30) of claim 8, comprising:

10. 10. The electronic device (30) of claim 8 or 9, wherein transforming (103) the medical image (12) comprises reframing the image (12) around the lesion (13) according to a frame of predetermined dimensions, the frame being common to all of the medical images (22) of the training element (21), and the position of the lesion (23) relative to the frame being variable for each of the medical images (22).

11. The method (100) comprises: - calculating (106) for each training element (21) corresponding to the selected optimal treatment, 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); - selecting (107) a reference training element from among said set of training elements (21) as a function of said calculated similarity value; The electronic device (30) of any one of claims 1 to 10, comprising:

12. The electronic device (30) according to any one of the preceding claims, wherein the medical image (12, 22) is acquired by tomographic densitometry, magnetic resonance imaging or ultrasound.

13. The electronic device (30) of any one of claims 1 to 12, wherein the medical image (12, 22) is three-dimensional.

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

15. The electronic device (30) of any one of claims 1 to 14, wherein the machine learning algorithm (20) is further pre-trained using training elements in which medical images (22) do not show any pathology in the anatomical structure of interest.

Citation Information

Patent Citations

  • Method and system for automating the detection of inclusion or exclusion criteria

    JP2020507149A

  • Systems and methods for determining segments targeted for ablation

    JP2021510092A

  • Method for planning tissue ablation based on deep learning

    WO2020260433A1