Ablation region determination method based on deep learning

Abstract

A post-treatment evaluation method (200) for ablation of a portion of a target anatomical structure (130) of an individual (110), the target anatomical structure comprising at least one lesion (165). The evaluation method comprises in particular the following steps: automatically determining the contour of the ablation zone by means of a machine learning method of the neural network type, analyzing post-treatment images of the target anatomical structure of the individual, pre-training (290) the machine learning method in the training phase using a database comprising a plurality of post-operative medical images of the same target anatomical structure of a group of patients, each medical image in the database being associated with an ablation zone of the target anatomical structure of the patient. The invention also relates to an electronic device (150) comprising a processor and a computer memory storing instructions for such an evaluation method.

Description

Field of the Invention

[0001] The field of the invention is the evaluation of medical interventions.

[0002] More specifically, the present invention relates to a method for post-treatment assessment of lesion ablation areas and prediction of associated recurrence risks.

[0003] The invention is particularly suitable for evaluating minimally invasive medical interventions and predicting the risk of recurrence of lesions, such as tumors or metastases. Such minimally invasive medical interventions correspond, for example, to percutaneous ablation of lesions, such as tumors in the liver, lungs, kidneys or any other organ. Percutaneous ablation generally involves the insertion of one or more needles through the skin under image guidance to reach and destroy the lesion. Background of the Invention

[0005] Techniques for assessing the effectiveness of interventions and predicting the risk of lesion recurrence are known in the art.

[0006] Such techniques include, for example, determining, after a lesion has been ablated, the extent to which the ablation zone covers the lesion. By comparing the volume of the ablation zone with the volume of the lesion, the ablation margin can be determined. In practice, an ablation margin of at least 5 mm is generally recommended.

[0007] To determine these margins, the volume of the lesion is usually determined when planning the intervention and compared with the volume of the ablation zone segmented by the operator on at least one postoperative image.

[0008] The main disadvantage is that the volume of the ablation zone is usually determined with little accuracy, often to some extent depending on the operator performing the segmentation. In addition, the quality of postoperative images is often poor, which introduces uncertainty in the segmentation. Therefore, it is difficult to determine the correlation between the ablation margin and the risk of lesion recurrence.

[0009] In order to improve the prior art techniques, it is known practice to use methods for automatic segmentation of the ablation area.

[0010] For example, Zhang et al. describe this technique in a scientific publication titled “Detection and monitoring of thermal damage caused by microwave ablation using ultrasound imaging and convolutional neural networks” published in September 2019. The segmentation method described in the publication can calculate the remainder of the ablation area by segmenting the pre-operative ultrasound image and the post-operative ultrasound image.

[0011] However, the segmentation method described in this publication cannot predict the risk of recurrence due to the low accuracy of the automatic segmentation of the ablation zone. First, the segmentation method is limited to a sub-sampling matrix of the ablation zone image, whose size is fixed, generally equal to 4 mm. 2, thus limiting the use of the method to small-sized regions. In addition, in order to fix the position of the subsampling matrix, the position of the ablation region needs to be known, making the method difficult to use without constant reference points within the subsampling matrix in pre- and postoperative images. Finally, due to the nature and quality of 2D ultrasound images, the target anatomical structure may be difficult to discern, making the segmentation of the region inaccurate and, in particular, leading to the erroneous observation that the ablation region to be segmented does not include the lesion to be ablated.

[0012] Furthermore, automatic segmentation methods provide consistent results for homogeneous regions (i.e. bones, vessels or lesions, or for example when the image contains a known number of regions). In the case of ablation regions, the obtained segmentation results are less consistent, since ablation regions are highly complex regions, generally composed of different materials such as gas, necrotic cells, healthy cells, residual contrast agent, calcifications, etc. Furthermore, segmentation is generally performed on medical images, which are usually blurred and low in contrast, which makes automatic segmentation of images difficult.

[0013] None of the current systems can meet all the needs at the same time, namely to provide a technique that enables accurate assessment of ablation therapy by segmenting ablation regions, especially heterogeneous ablation regions, with better accuracy, especially based on blurry and / or low-contrast medical images. Summary of the invention

[0014] The present invention aims to overcome all or part of the above-mentioned defects of the prior art.

[0015] To this end, a subject of the present invention is a method for post-treatment evaluation of ablation of a portion of a target anatomical structure of an individual, the target anatomical structure comprising at least one lesion.

[0016] The ablation is a percutaneous or minimally invasive ablation, which usually involves inserting at least one needle into the skin to reach and destroy the lesion. Most ablation techniques can be: radiofrequency, microwave, electroporation, laser, cryotherapy, ultrasound, etc.

[0017] The target anatomical structure may be the liver, lung, kidney, or any other organ that is prone to lesions.

[0018] According to the present invention, the post-treatment evaluation method comprises the following steps:

[0019] - Acquire post-operative medical images of individual target anatomical structures;

[0020] - Automatically determine the contours of the ablation zone by means of a machine learning method of the neural network type, analyzing the post-treatment images of the target anatomical structure of the individual, wherein the machine learning method is pre-trained during the training phase using a database of multiple post-operative medical images of the same target anatomical structure of a group of patients, each medical image in the database being associated with an ablation zone of the target anatomical structure of the patient.

[0021] Thus, based on previous learning from a plurality of medical images segmented by at least one operator, preferably by a plurality of operators, the ablation zone is automatically segmented in the postoperative medical image. It should be noted that such automatic segmentation by a neural network makes it possible to dispense with the presence of an operator experienced in medical image analysis. In addition, the segmentation obtained by the neural network is generally more precise, especially in the case of analyzing three-dimensional images. Finally, in the case of images showing low contrast and / or sharpness, the automatic segmentation using this method is also of better quality.

[0022] In some specific embodiments of the invention, the training phase comprises a prior step of training using medical images of the same target anatomical structure containing non-ablated lesions.

[0023] Therefore, the neural network is better at segmenting ablation areas in individual post-operative medical images. This surprising effect can be explained by the similarity in terms of shape and location between lesions and ablation areas. Furthermore, it should be noted that the number of accessible medical images showing non-ablated lesions on a given target anatomy is generally higher than the number of accessible medical images acquired after lesion ablation.

[0024] In other specific embodiments of the present invention, the post-treatment evaluation method also includes the step of aligning a postoperative image and a medical image of the individual target anatomical structure, wherein the medical image is obtained before surgical treatment and is referred to as a preoperative medical image, and the aligned postoperative medical image and preoperative medical image form a pair of medical images of the individual target anatomical structure.

[0025] Thus, an analysis of the location of the ablation zone associated with the lesion may be performed.

[0026] “Preoperative medical image” refers to the medical image obtained before ablation therapy, and “postoperative medical image” refers to the medical image obtained after ablation therapy.

[0027] In some specific embodiments of the present invention, the post-treatment evaluation method further comprises the step of evaluating the risk of recurrence based on relative characteristics between the ablation area and the lesion, between the ablation area and the target anatomical structure, or between the lesion and the target anatomical structure.

[0028] Thus, the post-treatment evaluation method according to the present invention provides medical personnel with a better view of the ablation treatment applied to an individual by allowing them to assess the need for additional treatment when risk of recurrence is demonstrated.

[0029] The risk of recurrence is usually in the form of a binary value, for example, which may be equal to 0 or 1. A positive value will be interpreted as demonstrating a risk of recurrence, and a negative value will be interpreted as a low risk of recurrence.

[0030] However, the risk of recurrence may also take the form of a probability between 0 and 1. The risk of recurrence will then be understood to have been demonstrated when the risk value is above a threshold value, for example equal to 0.5.

[0031] In some specific embodiments of the present invention, when the risk of recurrence is confirmed, the post-treatment evaluation method also includes a step of determining the recurrence location based on relative features between the ablation area and the lesion, between the ablation area and the target anatomical structure, or between the lesion and the target anatomical structure.

[0032] In some specific embodiments of the present invention, the risk of recurrence is assessed by considering the ablation margin between the ablation area and the lesion.

[0033] For example, for a negative recurrence risk, the ablation margin is equal to or greater than 5 mm.

[0034] The ablation margin is usually defined as the minimum distance between the ablation zone and the lesion.

[0035] In some specific embodiments of the present invention, the risk of recurrence is assessed by considering the distance between the centroid of the lesion and the centroid of the ablation zone.

[0036] The reference value of the centroid depends on the ablation margin. If the ablation margin is 10 mm and the reference value of the ablation margin is 5 mm, the distance between the centroid of the lesion and the centroid of the ablation zone should be less than or equal to 5 mm.

[0037] In some specific embodiments of the present invention, the risk of recurrence is assessed by considering the uniformity and sharpness of the margin of the ablation zone relative to the surrounding healthy tissue.

[0038] "Surrounding healthy tissue" means healthy tissue of the target anatomy located within the frame of the cropped medical image.

[0039] In some specific embodiments of the present invention, the risk of recurrence is assessed by considering the ratio of the lesion volume to the ablation zone volume.

[0040] In some specific embodiments of the invention, the risk of recurrence is assessed by considering the central location of the lesion relative to the target anatomical structure.

[0041] In some specific embodiments of the present invention, the post-treatment evaluation method further comprises the step of segmenting the lesion in the pre-operative medical image of the individual target anatomical structure.

[0042] In some specific embodiments of the present invention, the post-treatment evaluation method further comprises the step of detecting lesions in a pre-operative medical image of the individual target anatomical structure.

[0043] In some specific embodiments of the present invention, all or part of the medical images in the database are cropped around an ablation area including at least one lesion, the cropping of the image is performed using a common frame of a predetermined size, and the collection of centers of the ablation areas in the cropped medical images in the database forms a cluster of different points within the common frame.

[0044] Therefore, by distributing the locations of the ablation zones in a common frame, the prediction errors in the machine learning method can be reduced. In the case where all ablation zones are at the same location in the common frame, the machine learning method will mainly consider the post-operative images featuring the ablation zone at that specific location, resulting in prediction errors in the case where the ablation zone is at another location.

[0045] In some specific embodiments of the present invention, for a set of medical images in a database, a portion of an individual's body included in the image is divided into a plurality of basic units of a single size, and the number of basic units is divided into two approximately equal parts between the portion of the human body defined by the ablation area and the rest of the individual's body included in the image.

[0046] It should be noted that the equal distribution between primitives corresponding to ablated regions and primitives corresponding to non-ablated regions can be analyzed at the image level or globally across all images.

[0047] The basic unit is often called a pixel in the context of two-dimensional images or a voxel in the context of three-dimensional images.

[0048] By "approximately equal parts" is meant when the two groups of basic units consist of the same number of basic units, or when the number of basic units in each of the two groups differs by, for example, less than 5% of the number of basic units in the two groups.

[0049] In some specific embodiments of the present invention, the postoperative medical image database includes at least one preoperative medical image, and the preoperative medical image includes at least one unablated lesion.

[0050] As a result, machine learning methods are better trained.

[0051] In some specific embodiments of the present invention, the post-treatment evaluation method further comprises the step of determining a supplemental ablation mask when the risk of recurrence is demonstrated.

[0052] Therefore, in order to eliminate the risk of recurrence, the treatment plan is estimated by the post-treatment evaluation method. It should be noted that this treatment plan is not mandatory and may or may not be followed by the medical staff.

[0053] In some specific embodiments of the present invention, the post-treatment evaluation method further comprises the step of planning a path of the medical device to a target point of the ablation zone defined by the supplementary ablation mask.

[0054] It should be noted that this planning step is performed prior to complementary treatment.

[0055] In some specific embodiments of the present invention, the post-treatment evaluation method further comprises the step of helping an operator of the medical device to follow the planned path.

[0056] In some specific embodiments of the present invention, the planned path and / or the guidance indicator are displayed in real time on the screen of the augmented reality device.

[0057] In some specific embodiments of the present invention, the medical image is a three-dimensional image.

[0058] It should be noted that a three-dimensional image may correspond to a collection of two-dimensional images taken at substantially regular intervals along a predetermined axis.

[0059] In some specific embodiments of the present invention, the same image acquisition technique is used to acquire each post-operative image.

[0060] In other words, the technique used to acquire the postoperative medical images is the same as the technique used to acquire the postoperative medical images in the training database of the machine learning method.

[0061] The present invention also relates to an electronic device comprising a processor and a computer memory storing instructions for a method according to any one of the preceding embodiments.

[0062] For example, such an electronic device can be a control device, a navigation system, a robotic device or an augmented reality device. The control device can in particular be a computer present in the operating room or a remote server. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Other advantages, objects and specific features of the invention will become apparent from the following non-limiting description of at least one particular embodiment of the device and method that is the subject of the invention, with reference to the accompanying drawings, in which:

[0064] - Figure 1 It is a schematic diagram of medical intervention;

[0065] - Figure 2 is used for Figure 1 A framework diagram of the methods for post-treatment evaluation of medical interventions;

[0066] - Figure 3 is Figure 2 An example of a three-dimensional medical image in which the ablation area is highlighted for use in the illustrated method;

[0067] - Figure 4 are four examples of medical images, each of which includes an ablation mask manually defined by an operator and Figure 2 Ablation mask predicted by the neural network in the shown method;

[0068] - Figure 5 is Figure 2 Examples of medical images in which automatic segmentation of lesions and ablation areas is performed in the steps of the method shown;

[0069] - Figure 6 Shown by Figure 2 The method shown in is optionally presented as an example of supplementary ablation. DETAILED DESCRIPTION OF THE INVENTION

[0071] This description is not limiting and each feature of one embodiment can be combined in an advantageous manner with any other feature of any other embodiment.

[0072] It should be noted hereafter that the pictures are not drawn to scale. Example

[0073] Figure 1 1 is a schematic diagram of a medical intervention in which a medical device 120 is used to treat an individual 110 lying on a table 115. In the current non-limiting example of the invention, the medical intervention corresponds to ablation of a lesion 165 in a target anatomical structure 130, in this case the liver of the individual 110, with the medical device 120, in this case a semi-rigid needle. The medical intervention in this case is a percutaneous procedure without opening the body of the individual 110.

[0074] The manipulation of the medical instrument 120 by the operator 140 may advantageously be guided by a guiding device, which in the present non-limiting example of the invention is an augmented reality device, such as a headset 150 worn by the operator 140. The medical instrument 120 may also be associated with a medical robotic device 125.

[0075] The headset 150 comprises a semi-transparent screen 155 allowing normal viewing by the operator. On the screen 155, an image is superimposed to display markings which make it possible to guide the operator in the operation of the medical device 120 with a view to ablating a region 160, called the ablation zone, around a lesion 165 identified in the target anatomical structure 130. The markings may in particular comprise an ablation mask which has been pre-estimated on a medical image 170 of the target anatomical structure 130 acquired prior to the operation. The medical image 170 will hereinafter be referred to as the pre-operative medical image 170.

[0076] Once the procedure is completed, evaluation of the surgical treatment is performed by a post-treatment evaluation method 200 for ablation, such as Figure 2 1, the instructions of which are stored in a computer memory 180 of an electronic control device 181 connected to the headset 155 by a cable or wireless technology. The post-treatment evaluation method 200, the instructions of which are processed by a computer processor 182 of the electronic control device 181, can in particular determine the ablation area and the associated risk of recurrence in order to check whether the surgical treatment performed during the operation is sufficient or whether it is better to continue the treatment by, for example, supplementary ablation.

[0077] It should be noted that the electronic device 181 may advantageously be integrated into the headset 150 .

[0078] The post-treatment evaluation method 200 includes a first step 210 of acquiring a post-operative medical image of the target anatomical structure 130 .

[0079] It should be noted that the preoperative and postoperative medical images are preferably acquired by means of a computer tomography method. Alternatively, they may be acquired using a magnetic resonance imaging device.

[0080] Preferably, the techniques used to acquire the pre-operative medical images and the post-operative medical images are similar or even the same.

[0081] However, the technology used to acquire postoperative medical images may be different from the technology used to acquire preoperative medical images.

[0082] In this non-limiting example of the invention, the pre-operative and post-operative medical images are advantageously images acquired in three dimensions. In practice, each medical image acquired in three dimensions typically corresponds to a collection of medical images in two dimensions, each corresponding to a cross section through the target anatomical structure 130 taken at regular intervals along a predetermined axis. A three-dimensional representation of the target anatomical structure can be reconstructed from this collection of two-dimensional medical images. Therefore, the term "three-dimensional image" will be understood to mean both a collection of medical images and a three-dimensional representation. The term "voxel" will be understood to mean a basic unit related to the resolution of a three-dimensional image.

[0083] Optionally, both the preoperative and postoperative medical images are acquired in two dimensions. The basic unit associated with the resolution of a two-dimensional image is usually referred to as a pixel.

[0084] The pre-operative and post-operative medical images are either images that include the entire target anatomy or images that are cropped around the ablation area using a predefined frame. In a three-dimensional image, the frame around the ablation area corresponds to a cube, while in a two-dimensional image, the frame corresponds to a square.

[0085] A frame around the ablation zone, also called a "bounding box", can be automatically generated around the ablation zone following one action of the operator. For example, such an action can correspond to the operator indicating a point in the post-operative medical image that belongs to the ablation zone, and then generating a frame around this point. For example, in the case of minimally invasive ablation treatment of small lesions, i.e. lesions with a maximum diameter of the order of 5 cm + / - 10%, for example, even preferably with a maximum diameter of the order of 3 cm + / - 10%, each side of the cube or each side of the square has a length between 5 cm and 10 cm.

[0086] Figure 3 3 is a schematic diagram of a three-dimensional medical image 300, in which an ablation area 310 is surrounded by a frame 320. The frame 320 is a cube and corresponds to the square 330 in the cross-sectional views 340, 350 and 360, which are the sagittal, axial and coronal planes, respectively.

[0087] In a second step 220 , the post-operative medical image of the target anatomical structure 130 is analyzed by a neural network, which is a machine learning method, to automatically segment the ablation region in the post-operative medical image of the target anatomical structure 130 of the individual 110 .

[0088] To this end, the neural network has previously been trained on a database of medical images of the same target anatomical structure (thus the liver in this case) of a group of patients in a preliminary training phase 290. Each medical image in the database comprises a target anatomical structure having the same function as the target anatomical structure 130.

[0089] Advantageously, the post-operative medical images of the target anatomical structure 130 of the individual 110 are acquired in the same manner as the medical images in the neural network training database.

[0090] When the postoperative medical image of the target anatomical structure 130 of the individual 110 is cropped, the size of the cube or square of the postoperative medical image is advantageously the same as the size of the cube or square used for training. In other words, the medical image in the database has the same size as the cropped postoperative medical image.

[0091] For training the neural network, the ablation area of ​​each post-operative image in the database of ablated lesions has been previously segmented by at least two operators to increase the relevance of the learning and thus the relevance of the analysis results obtained by the neural network. In particular, it may be difficult for operators to define the ablation area, especially when the contrast in the image is low, such as in Figure 4 As can be seen in the four pre-segmented post-operative medical images 400 in . Therefore, using multiple operators to annotate medical images can improve the identification of ablation regions. Therefore, in the current non-limiting example of the present invention, the ablation region associated with the registered post-operative images therefore corresponds to a combination of ablation regions proposed by the operators. As an alternative example, the ablation region associated with the registered post-operative medical images can correspond to the intersection, consensus or arbitration of ablation regions proposed by the operators. The neural network is further trained to classify voxels of the medical images in areas with or without ablation.

[0092] Alternatively, learning can be performed using a single expert annotator who delineates ablation regions in medical images. The operator's experience is important so that the neural network can arrive at well-defined ablation regions.

[0093] It should also be noted that preferably the set of images in the database includes as many voxels belonging to regions with ablation as to regions without ablation. This ratio is calculated based on the voxel classification manually determined by the operator.

[0094] In other words, the individual body part contained in each image in the database is divided into a plurality of basic units of a single size, and the number of basic units is divided into two almost equal parts between the human body part defined by the ablation area and the rest of the individual body contained in the image.

[0095] Furthermore, it is also preferred that the ablation zone is not always in the center of the frame in each medical image in the training database. In the case where the ablation zone is mainly in the center of the frame, a bias will be introduced in the neural network, which will learn that the ablation zone is mainly in the center of the frame, but this is not necessarily the case, especially if the operator makes errors in the positioning of the frame. It is therefore advantageous to add a bounded random variable to the position of the frame to limit this bias with respect to the positioning of the lesion in the center of the frame.

[0096] The collection of centers of ablation regions in cropped medical images in the database thus forms a cluster of distinct points within a common frame.

[0097] In order to limit false positives generated by the neural network, it is further preferred that the database comprises medical images comprising at least one non-ablated lesion.

[0098] The 290th stage of training a neural network is usually carried out in several steps:

[0099] - training step 291;

[0100] - verification step 292;

[0101] -Test step 293.

[0102] Therefore, the medical image database is divided into three databases including different medical images. These three databases are respectively called training database, verification database and test database. In the current non-limiting example of the present invention, 60% to 98% of the medical images in the medical image database are grouped together in the training database, 1% to 20% are grouped together in the verification database, and 1% to 20% are grouped together in the test database. The percentage usually depends on the number of images in the medical image database, which is given here as an indication.

[0103] The first two steps 291 and 292 of the phase 290 of training the neural network are the main steps that can be repeated multiple times. The third test step is optional.

[0104] In a first step 291 of the training phase 290 , the weights W and biases b of each neuron of the neural network are determined based on the medical images in the training database.

[0105] It should be noted that the training database may advantageously comprise medical images comprising at least one non-ablated lesion.

[0106] The second step 292 in the training phase 290 makes it possible to verify the weights W and bias b determined in advance for each neuron of the neural network based on the medical images in the verification database to verify the results of the neural network, in particular the prediction error, i.e., the prediction error obtained by comparing each medical image in the verification database with the ablation area obtained by segmenting the ablation area from the medical image extracted from the training database.

[0107] In case the prediction error is too large when completing this second step, both the training 291 and validation 292 steps are performed again to retrain the neural network using the same medical images to refine the values ​​of the weight W and bias b of each neuron.

[0108] Optionally, when retraining the neural network, a first step 291 uses a resampling of medical images, taking into account the training medical images in the training database and some medical images in the validation database. The remaining medical images in the validation database are then used to validate the weights W and biases b obtained when completing the first retraining step.

[0109] It should be noted that the neural network can be retrained as many times as needed until the prediction error is acceptable, ie, below a predetermined value.

[0110] Once the two steps 291 and 292 in the training phase 290 have been implemented at least once, the final performance of the neural network can be tested in a possible third test step 293 using medical images from the test database. These medical images, different from those in the training and validation databases, make it possible to verify that the neural network configured with the parameters W and b for each neuron allows segmentation of the ablation zone with high accuracy in all situations that the neural network may face. However, unlike the validation step 292, this potential third test step 293 does not lead to a new training cycle of the neural network.

[0111] It should be noted that the images used in the testing step 293 are typically carefully selected to cover different locations and sizes of ablation regions in the target anatomy to best test the predictive ability of the neural network.

[0112] Based on the post-operative medical image of the target anatomical structure 130, the neural network classifies each voxel as an area to be ablated or not ablated. This prediction can take the form of an ablation mask overlaid on the post-operative medical image of the target anatomical structure 130. The ablation mask is typically registered to the voxels belonging to the ablation area predicted by the neural network. It should be noted that in the case of a two-dimensional image, the ablation mask is typically defined by a surface or contour.

[0113] Advantageously, before the training phase 290, the neural network may have been previously trained in said pre-training phase 295 on a second database of medical images, the database comprising medical images showing lesions of the same type of target anatomical structure as the individual 110. This pre-training phase 295 cleverly uses the similarity of lesion shape and ablation zone shape, and even the similarity of position within the target anatomical structure, allowing a better segmentation of the ablation zone in the post-operative medical images of the individual 110.

[0114] It should be noted that the medical images in the second database may advantageously have been previously segmented by at least one operator. The learning in stage 295 is similar to the learning performed in stage 290 .

[0115] exist Figure 2 In the third step 230 of the illustrated method 200, the postoperative medical image is registered with the preoperative medical image 170. Registration, which allows to find correspondences between anatomical points in two medical images, is performed using methods known to those skilled in the art. Registration can be performed rigidly, i.e. all points in the image are transformed in the same way, or flexibly, i.e. each point in the image can have a specific transformation.

[0116] Then, in a fourth step 240 comprising four sub-steps 241 , 242 , 243 and 244 , an assessment of the risk of recurrence is performed.

[0117] In a possible sub-step 241, lesions 165 are detected in a pre-operative medical image of a target anatomy 130 of an individual 110. This detection may be performed automatically, or manually by an operator.

[0118] In sub-step 242, segmentation of lesion 165 is automatically performed on the pre-operative medical image of target anatomical structure 130 of individual 110. Alternatively, segmentation is performed manually by an operator.

[0119] The lesion is automatically segmented using methods known to those skilled in the art. For example, segmentation is performed using image-based histogram methods (such as Otsu's method) or using deep learning methods.

[0120] The segmentation sub-step 242 consists of Figure 5 Shown, Figure 5 A pre-operative medical image 500 is shown in which automatic segmentation based on a deep learning approach is performed in order to determine the three-dimensional position of a lesion 510 and an ablation zone 520. A different neural network than the one used in step 230 may be used to obtain equivalent results.

[0121] Then in substep 243, an ablation margin is determined between the segmentation of the lesion and the previously established ablation mask. The ablation margin corresponds to the minimum margin, i.e. the minimum distance between the segmentation of the lesion and the ablation mask. In other words, the ablation margin corresponds to the minimum distance calculated between the points of the lesion and the points of the ablation zone, and is calculated for all points of the lesion.

[0122] Determining the ablation margin can ensure that the ablation zone properly covers the lesion.

[0123] Based on the reference value of the ablation margin determined in sub-step 243, the prediction of the risk of recurrence or the determination of the location of the recurrence can be evaluated in sub-step 244 by comparing the calculated ablation margin with the value of the ablation margin associated with the recurrence status stored in the database, the recurrence status indicating whether a recurrence is observed after the operation, possibly with an associated recurrence date. For example, when the ablation margin is equal to or greater than 5 mm, the risk of recurrence can be considered to be zero.

[0124] It should be noted that predictions of risk of lesion recurrence are often in the form of a binary value, such as equal to 0 (zero or negative risk) or equal to 1 (proven or positive risk).

[0125] Additionally or alternatively, predictors of risk of recurrence other than ablation margin may be used. The risk of recurrence and the location of recurrence may be estimated by weighting all or some of these different predictors. For example, predictors of risk of recurrence are based on relative features, which may be:

[0126] -ablation margin;

[0127] - the distance between the lesion surface or a part of the lesion surface and the ablation zone;

[0128] - The distance between the centroid of the lesion and the centroid of the ablation zone;

[0129] - the distance between the lesion surface or a portion of the lesion surface and the ablation zone and the distance between the center of mass of the lesion and the center of mass of the ablation zone, taking into account the proximity of the capsule to the target anatomy, in particular in case of subcapsular lesions;

[0130] - uniformity of the edges of the ablation zone relative to the surrounding healthy tissue;

[0131] -The sharpness of the edges of the ablation zone relative to the surrounding healthy tissue;

[0132] -Ratio of lesion volume to ablation zone volume;

[0133] -Location of the lesion within the target anatomy.

[0134] The reference value of the centroid depends on the ablation margin. If the ablation margin is 10 mm and the reference value of the ablation margin is 5 mm, the distance between the centroid of the lesion and the centroid of the ablation zone should be less than or equal to 5 mm.

[0135] It should be noted that in order to take into account multiple risk predictors, the value of the risk of recurrence may advantageously be continuous rather than binary between 0 and 1. Thus, a weighting may be performed between the different risk predictors to obtain a value between 0 and 1. Then when the risk of recurrence is above a threshold, for example equal to 0.5, it will be understood to have been demonstrated.

[0136] If the risk is confirmed, the location of the recurrence can be determined in order to estimate a supplementary ablation mask in the fifth step 250 of the post-treatment evaluation method 200. For example, consider that the supplementary ablation mask performs supplementary ablation in areas where the ablation margin is less than a given value, such as five millimeters.

[0137] In step 250, automatic identification of regions where the ablation margin is less than a threshold value, such as five millimeters, may be performed.

[0138] Figure 6An example of supplementary ablation after ablation treatment of a lesion 600 with a risk of recurrence is shown. The evaluation method 200 has identified an area between the lesion 600 and the ablation zone 610 where the ablation margin 605 is insufficient, and a supplementary ablation mask 620 has been generated. It should be noted that the supplementary ablation mask 620 is generated while trying to limit the ablation zone as much as possible. In addition, the target point 630 to be reached by the ablation needle can then be defined in the mask 620.

[0139] In order to guide the operator in manipulating the medical device 120 in step 270 of guiding the medical device 120 along the planned path, the method 200 may further include step 260 of planning a path to be followed by the medical device 120 associated with the ablation mask or the supplementary ablation mask.

[0140] An example of a planning method is described in French patent application number 1914780 entitled "Method for automatically planning a trajectory of a medical intervention".

[0141] It should be noted that guidance in this non-limiting example of the present invention is visual by displaying the planned path and / or guidance indicators on the screen 155 of the headset 150 .

[0142] Alternatively, the medical device 120 may be guided by means of a navigation system providing information about the position and orientation of the medical device 120. This may be the case with mechanical guidance by a robotic device connected to such a navigation system.

[0143] It should be noted that steps 230 to 260 may be repeated until the risk of recurrence is zero or almost zero, or until the ablation margin is sufficient.

Claims

1. A post-treatment evaluation method (200) for ablation of a portion of a target anatomical structure (130) of an individual (110), the target anatomical structure comprising at least one lesion (165, 510, 600), the ablation of the portion of the target anatomical structure being defined by an ablation zone (160), characterized in that: The evaluation method comprises the following steps: acquiring (210) a post-operative medical image of a target anatomical structure of the individual, including all or part of the ablation area; Automatically determining the contours of the ablation zone by a machine learning method of the neural network type, analyzing the post-treatment images of the target anatomical structure of the individual, the machine learning method being pre-trained (290) in a training phase using a database comprising a plurality of post-operative medical images of the same target anatomical structure of a group of patients, each medical image in the database being associated with an ablation zone of the target anatomical structure of the patient; assessing the risk of recurrence based on relative characteristics between the ablation zone and the lesion, between the ablation zone and the target anatomical structure, or between the lesion and the target anatomical structure; The training phase includes a prior step of training using medical images of the same target anatomical structure containing non-ablated lesions.

2. The post-treatment evaluation method according to claim 1 further comprises a step (220) of registering a postoperative image and a medical image (170, 500) of the target anatomical structure of the individual, wherein the medical image is obtained before surgical treatment and is referred to as a preoperative medical image.

3. The post-treatment evaluation method according to claim 1 further includes, when the risk of recurrence is proven, a step of determining the recurrence location based on relative features between the ablation area and the lesion, between the ablation area and the target anatomical structure, or between the lesion and the target anatomical structure.

4. The post-treatment evaluation method according to claim 1 or 3, wherein: The relative feature is the ablation margin between the ablation zone and the lesion.

5. The post-treatment evaluation method according to claim 1 or 3, wherein: The relative features are the centroid of the lesion and the centroid of the ablation zone.

6. The post-treatment evaluation method according to claim 1 or 3, wherein: Relative characteristics are the uniformity and sharpness of the edges of the ablation zone relative to the surrounding healthy tissue.

7. The post-treatment evaluation method according to claim 1 or 3, wherein: The relative characteristic is the ratio of the volume of the lesion to the volume of the ablation zone.

8. The post-treatment evaluation method according to claim 1 or 3, wherein: The relative feature is the location of the lesion relative to the center of the target anatomy.

9. The post-treatment evaluation method of claim 1, further comprising the step of segmenting the lesion in a pre-operative medical image of the target anatomical structure of the individual.

10. The post-treatment evaluation method of claim 1, further comprising the step of detecting the lesion in a pre-operative medical image of the individual's target anatomical structure.

11. The post-treatment evaluation method according to claim 1, wherein: All or part of the training postoperative medical images in the database are cropped around an ablation area including at least one lesion, the cropping of the images is performed using a common frame of a predetermined size, and the collection of ablation area centers in the cropped postoperative medical images forms a cluster of different points within the common frame.

12. The post-treatment evaluation method according to claim 1, wherein: For the set of medical images in the database, the part of the individual body included in the image is divided into a plurality of basic units of a single size, and the number of basic units is divided into two equal parts between the part of the human body defined by the ablation area and the rest of the individual body included in the image.

13. The post-treatment evaluation method according to claim 1, wherein: The postoperative medical image database includes at least one preoperative medical image containing at least one unablated lesion.

14. The post-treatment evaluation method of claim 1, further comprising the step of determining a supplemental ablation mask when a risk of recurrence is demonstrated.

15. The post-treatment evaluation method of claim 14, further comprising the step of proposing a path to be followed by a medical device to a target point of the supplemental ablation mask.

16. The post-treatment evaluation method according to claim 1, wherein: The medical image is a three-dimensional image.

17. The post-treatment evaluation method according to claim 1, wherein: Each postoperative image was acquired using the same image acquisition technique.

18. An electronic device (150) comprising a processor and a computer memory storing instructions for a post-treatment assessment method according to any one of claims 1 to 3 or 9 to 17.

19. The electronic device according to claim 18, wherein: The device may be a control device, a navigation system, a robotic device or an augmented reality device.

Images