DEVICE AND METHOD FOR AID IN THE DIAGNOSIS OF PELVIC PATHOLOGIES

Abstract

DEVICE AND METHOD FOR AID IN THE DIAGNOSIS OF PELVIC PATHOLOGIES The present invention relates to a device and an associated method for aid in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one item of information relating to the diagnosis, said device comprising: at least one input interface configured to receive at least one volumetric image acquired with an MRI modality, said image comprising at least a portion of the pelvis of said patient;at least one processor configured to: segment said at least one image by a segmentation neural network, extract at least one first specific region of interest, from said at least one segmented image and at least one predefined rule relating to a first pathology among said pelvic pathologies, and calculate at least one information relating to the diagnosis linked to said specific region of interest by at least one first expert model associated with said first pathology; and at least one output interface configured to provide said at least one information relating to the diagnosis. Abstract figure: Fig. 1;

Description

Title of the invention: DEVICE AND METHOD FOR AID IN THE DIAGNOSIS OF PELVIC PATHOLOGIES FIELD OF THE INVENTION

[0001] The present invention relates to the field of medical image analysis. More particularly, the present invention relates to a device and a method for assisting in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one piece of information relating to the diagnosis. STATE OF THE ART

[0002] Pelvic pathologies in female patients represent a crucial area of ​​reproductive health, and the accurate diagnosis of these pathologies is of paramount importance.

[0003] Pelvic pathologies in women represent a major health concern, given their high prevalence and significant impacts on women's health. These pathologies, such as endometriosis, uterine fibroids, pelvic infections, myomas, polycystic ovarian syndrome and other gynecological disorders, are frequently diagnosed in women of all ages. Endometriosis, for example, can affect women of reproductive age, causing severe pelvic pain, menstrual disturbances and potential fertility complications. Uterine fibroids, although often benign, affect a large proportion of young women, causing symptoms such as heavy bleeding and pelvic pain.These pelvic pathologies impact not only physical health, but also mental health, often leading to emotional disturbances related to chronic pain and gynecological complications. Better understanding, early detection, reliable diagnosis and appropriate treatment options are essential to mitigate the impact of these pelvic pathologies and improve women's overall health.

[0004] The current state of technology for the diagnosis of pelvic pathologies highlights the predominance of MRI (Magnetic Resonance Imaging) as a quality reference, frequently used as a complement to validate the results from preliminary studies, such as ultrasound, despite its relative simplicity of use. However, the interpretation of female pelvic MRI images remains a major challenge due to the substantial time required (20 to 40 minutes on average, not including acquisition time) and the lack of qualified healthcare personnel and the necessary specialized skills. Only a minority of radiologists are specifically trained in the analysis of images from gynecological MRI, which which frequently leads to diagnostic errors, even when the analysis is carried out by an expert.

[0005] There is then a need to overcome these obstacles by introducing a reliable innovative approach to limit the risk of false diagnoses of pelvic pathologies while accelerating the process of processing MRI images, thus improving the overall efficiency of diagnosis.

[0006] The invention falls within this context. SUMMARY

[0007] The present invention relates to a device for assisting in the diagnosis of pelvic pathologies in a female patient, making it possible to obtain at least one piece of information relating to the diagnosis, said device comprising: - at least one input interface configured to receive at least one volumetric image acquired with an MRI modality, said image comprising at least a portion of the pelvis of said patient, - at least one processor configured to: - segmenting said at least one volumetric image by a previously trained segmentation neural network, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said volumetric image based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, said volumetric images relating to the cohort of patients being acquired with at least said MRI modality, - extracting at least one first specific region of interest, from said at least one segmented image and at least one predefined rule relating to a first pathology among said pelvic pathologies, - calculating at least one piece of information relating to the diagnosis linked to said specific region of interest by at least one first expert model associated with said first pathology, said at least one first expert model being a neural network configured to receive as input said at least one first region of interest and to provide as output at least one score indicative of the presence of said first pathology in said patient, said at least one neural network having been previously trained on a training data set comprising annotated volumetric images relating to a group of patients, said volumetric images being acquired with at least said MRI modality, and said group of patients comprising at least one subgroup of patients affected by said first pathology, - at least one output interface configured to provide said at least one piece of diagnostic information.

[0008] Said device for assisting in the diagnosis of pelvic pathologies in a female patient is capable of providing at least one piece of information relating to the diagnosis for different pelvic pathologies at the same time. Furthermore, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said image based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, makes it possible to segment different structures of interest at the same time.

[0009] In one embodiment, the device is characterized in that the predefined ontology comprises at least the organs of the pelvic area and the lesions.

[0010] In one embodiment, the device is characterized in that the at least one processor is configured to calculate, from the segmented image of at least one structure of interest, a second piece of information relating to the diagnosis being the diameter of said structure.

[0011] In one embodiment, the device is characterized in that the at least one processor is configured to calculate a third piece of information relating to the diagnosis being the most optimal view of a given lesion.

[0012] In one embodiment, the device is characterized in that the at least one processor is configured to, upstream of the segmentation step, resample the image in a predefined spatial resolution and / or normalize the intensities of the voxels of the image.

[0013] In one embodiment, the device is characterized in that the at least one first expert model associated with said first pathology is configured to further receive clinical information as input.

[0014] The present invention further relates to a method for assisting in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one piece of information relating to the diagnosis, said method comprising: - reception of at least one volumetric image acquired with an MRI modality, said image comprising at least a portion of the pelvis of said patient, - segmentation of said at least one image by a previously trained segmentation neural network, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said image based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, said volumetric images relating to the cohort of patients being acquired with at least said MRI modality, - extraction of at least one first specific region of interest, from said at least one segmented image and at least one predefined rule relating to said first pathology among said pelvic pathologies, - calculation of at least one piece of diagnostic information related to said specific region of interest by at least one first expert model associated with said first pathology, said at least one first expert model being a neural network configured to receive as input said at least one first region of interest and to provide as output at least one score indicative of the presence of said first pathology in said patient, said at least one neural network having been previously trained on a training data set comprising annotated volumetric images relating to a group of patients, said volumetric images being acquired with at least said MRI modality, and said group of patients comprising at least one subgroup of patients affected by said first pathology, - provide as output said at least one piece of information relating to the diagnosis.

[0015] The present invention further relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the latter to implement the method described above.

[0016] The present invention further relates to a non-transitory computer-readable recording medium comprising instructions which, when the program is executed by a computer, cause the computer to implement the method described above.

[0017] Such a non-transitory computer-readable recording medium may be, without limitation, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor device, or any suitable combination of the foregoing. It should be noted that the following examples, although more specific, constitute only an illustrative and non-exhaustive list, readily appreciated by those skilled in the art: a portable computer floppy disk, a hard disk, a ROM, an EPROM (Erasable Programmable ROM) or Flash memory, a portable CD-ROM (Compact-Disc ROM). DEFINITIONS

[0018] In the present invention, the terms below are defined as follows:

[0019] The term "processor" should not be interpreted as being limited to hardware hardware capable of executing software, and generally refers to a processing device, which may, for example, include a computer, a microprocessor, an integrated circuit, or a programmable logic device (PLD). The processor may also include one or more graphics processing units (GPUs), whether used for computer graphics and image processing or other functions. In addition, the instructions and / or data for executing the associated and / or resulting functionalities may be stored on any medium readable by the processor, such as, for example, an integrated circuit, a hard disk drive, a CD (Compact Disc), an optical disk such as a DVD (Digital Versatile Disc), RAM (Random-Access Memory) or ROM (Read-Only Memory).Instructions may be stored in, among other things, computer hardware, software, firmware, or any combination thereof.

[0020] A “neural network”, “neural network” or “artificial neural network (ANN)” refers to a category of machine learning algorithm comprising nodes (called neurons), and connections between neurons modeled by “weights”. For each neuron, an output is given as a function of an input or set of inputs by an “activation function”. Neurons are generally organized into several “layers”, such that neurons in one layer only connect to neurons in the immediately preceding and immediately following layers. A “loss function” or “objective function” is a measure that evaluates how far a neural network’s predictions are from the true target values ​​when training the neural network. The main objective is to minimize this loss function in order to optimize the parameters (i.e.the weights) of the neural network so that it can make more accurate predictions.

[0021] A "convolutional neural network" refers to a type of acyclic ("feed-forward") artificial neural network comprising multiple hidden layers. The hidden layers are generally convolutional layers followed by activation layers, some of which are followed by "pooling" layers.

[0022]

[0026] The terms “adapted” and “configured” are used in these description so as to broadly encompass the initial configuration, subsequent adaptation or supplementation of the current device, or any combination thereof, whether by hardware or software (including firmware) means.

[0023]

[0027] “Machine Learning” (ML) refers to traditionally computer algorithms improve automatically through experience, based on training data to adjust the parameters computer models by reducing the gap between the expected outputs extracted from the training data and the evaluated outputs calculated by the computer models.

[0024]

[0028] A “hyper-parameter” currently means a parameter used for perform upstream control of model construction, such as a memory-forgetting balance in sample selection or a time window width, as opposed to a parameter of the model itself, which depends on specific situations. In machine learning applications, hyperparameters are used to control the learning process.

[0025]

[0029] “Datasets” are collections of data used for building a mathematical machine learning model, in order to make predictions or decisions based on data. In "supervised learning" (i.e., inferring functions from known input-output examples in the form of labeled training data), three types of machine learning datasets (also referred to as ML sets) are generally dedicated to three respective types of tasks: "training", i.e., parameter tuning, "validation", i.e., tuning ML hyper-parameters (which are parameters used to control the learning process), and "testing", i.e., independent verification of a training dataset exploited to build a mathematical model that the latter model provides satisfactory results.In "unsupervised learning," which is distinguished by its lack of labeling in the training data, datasets are dedicated to specific functions without the need for known input-output examples. In contrast to "supervised learning," where the focus is on tuning parameters from labeled data, "unsupervised learning" focuses on identifying structures, patterns, or groupings inherent in unlabeled data. Datasets in this context can be used for tasks such as segmentation, dimensionality reduction, or anomaly detection, without resorting to labeled training data.The traditional training, validation, and testing phases in supervised learning may not be as clearly defined in unsupervised learning, where evaluation may focus more on the quality of discovered structures rather than pre-labeled results. DESCRIPTION OF FIGURES .

[0026] The present invention will be better understood, and other specific characteristics and advantages will appear, upon reading the following description of particular and non-restrictive embodiments, the description making reference to the accompanying drawings in which:

[0027] [Fig. 1] is a functional diagram schematically representing a particular mode of a device for assisting in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one piece of information relating to the diagnosis, in accordance with the present disclosure;

[0028] [Fig.2] is a functional diagram schematically representing a particular embodiment of a device for assisting in the diagnosis of pelvic pathologies in a female patient in which several pelvic pathologies can be studied at the same time;

[0029] [Fig.3] is a flowchart showing the successive steps performed with the device of [Fig.l];

[0030] [Fig.4] schematically represents a device integrating the functions of the device of [Fig.l].

[0031] In the figures, the drawings are not to scale and identical or similar elements are designated by the same references. DETAILED DESCRIPTION

[0032] The present description illustrates the principles of the present disclosure. It will therefore be appreciated that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the disclosure and are included within its scope.

[0033] All examples and conditional language cited herein are intended for educational purposes to assist the reader in understanding the principles of the disclosure and the concepts contributed by the inventor to advance the state of the art, and should be construed as not being limited to these specifically cited examples and conditions.

[0034] Furthermore, all statements of principles, aspects and embodiments of the disclosure, as well as their specific examples, are intended to encompass their structural and functional equivalents. Furthermore, it is intended that these equivalents include both currently known equivalents and equivalents developed in the future, i.e., all developed elements that perform the same function, regardless of their structure.

[0035] Thus, for example, those skilled in the art will understand that the block diagrams presented herein may represent conceptual views of illustrative circuits implementing the principles of the disclosure. Likewise, it will be appreciated that all flowcharts, flow diagrams, and the like represent various processes that may be substantially represented on a computer-readable medium and thereby executed by a computer or processor, whether or not that computer or processor is explicitly shown.

[0036] The functions of the various elements illustrated in the figures may be performed by the use of dedicated computer hardware as well as computer hardware capable of executing software in association with appropriate software. When performed by a processor, the functions may be performed by a single dedicated processor, a single shared processor, or a plurality of individual processors, some of which may be shared.

[0037] It is understood that the elements illustrated in the figures may be implemented in various forms of hardware, software, or combinations thereof. Preferably, these elements are implemented in a combination of hardware and software on one or more suitably programmed general-purpose devices, which may include a processor, memory, and input / output interfaces.

[0038] The present disclosure will be described with reference to a particular functional embodiment of a device 1 for assisting in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one piece of information relating to the diagnosis, as illustrated in [Fig.l].

[0039] The device 1 is in fact adapted to receive as input at least one volumetric image of the patient as well as parameters of the automatic learning model 20 and to provide as output at least one piece of information relating to the diagnosis 31 of a pelvic pathology in a female patient.

[0040] The volumetric images 21 may include in particular at least a portion of the patient's pelvis. The volumetric MRI images 21 may be acquired with different MRI modalities, for example, T1, T2, T2 thin slices, T1 Fat (T1 with fat saturation) or T1 Water (T1 with water saturation). The use of images from MRI modalities is particularly advantageous because MRI is suitable for identifying different structures of the female pelvis.

[0041] The device 1 may be an apparatus, or a physical part of an apparatus, designed, configured and / or adapted to perform the functions mentioned and produce the effects or results mentioned. In alternative implementations, the device 1 is realized in the form of a set of apparatuses or physical parts of apparatuses, whether grouped in the same machine or in different, possibly remote, machines. The device 1 may for example have functions distributed on a cloud infrastructure and be available to users as a cloud-based service, or have remote functions accessible via an APL

[0042] The device 1 can be integrated into the same device or set of devices, and intended for the same users.

[0043] In the following disclosures, modules are to be understood as functional entities rather than as physically distinct hardware components. They can therefore be materialized either as grouped into a single tangible and concrete component, or distributed across several of these components. Similarly, each of these modules is possibly itself shared between at least two physical components. Furthermore, the modules are implemented in the form of hardware, software, firmware or any mixed form thereof. They are preferably incorporated into at least one processor of the device 1.

[0044] The device 1 comprises a module 11 for receiving at least one volumetric image of the patient 21 as well as parameters of the automatic learning model 20 stored in one or more local or remote database(s) 10. The latter may take the form of storage resources available from any type of suitable storage means, which may in particular be a RAM or an EEPROM (“Electrically-Erasable Programmable Read-Only Memory”) such as a Flash memory, possibly within an SSD (“Solid-State Disk”). In certain embodiments, the parameters of the model 20 are received by a communication network.

[0045] Optionally, the device 1 further comprises a module 12 for preprocessing the volumetric images acquired with an MRI modality 21. The module 12 may in particular be adapted to normalize the intensities of the voxels of the received volumetric image(s) 21 for efficient and reliable processing. Optionally, it may resample the volumetric image(s) 21 in a predefined spatial resolution. This may improve the efficiency of the downstream processing by the device 1. Such normalization may be particularly useful when the image(s) used come from different sources, in particular different MRI modalities.

[0046] For example, the 21 row(s) may be normalized using Z-score normalization, histogram normalization, Min-Max normalization, or atlas normalization.

[0047] The device 1 further comprises a segmentation module 13. In one embodiment, the module 13 is further configured to perform a segmentation of said at least one image 21 using a previously trained segmentation neural network, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said image 21 based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, said volumetric images relating to the cohort of patients being acquired with at least one MRI modality. In particular, at least one subgroup of the images of the cohort are acquired with the same MRI modality of image(s) 21.

[0048] One of the advantages of such a configuration of the module 13 is to segment different structures of interest (i.e. organs, lesions) at the same time.

[0049] In one embodiment, the training dataset for segmentation comprises annotated volumetric images relating to a cohort of patients, the annotated volumetric images originating from a group of female patients and covering different pelvic pathologies. The training dataset further comprises images of healthy subjects annotated to highlight normal structures while indicating that each pixel is negative for each pathology, meaning that no pixel represents the presence of pathology in these images. In one embodiment, the predefined ontology comprises at least the organs of the pelvic area and the lesions. For example, the organ of the pelvic area may be the uterus, the peritoneum, the fallopian tubes, the ovaries, the cervix, the vulva, the vagina, the rectum, the colon, the bladder or the urethra.The predefined ontology of the structures of interest associates with each organ a set of anatomical structures and a set of lesions that can be found at the level of this organ, said anatomical structures and lesions being to be segmented.

[0050] In one example, the organ of the pelvic area is the uterus, the set of anatomical structures at this organ contains the uterus, the junctional zone, the endometrium and the myometrium, and the set of lesions that can be found at this organ contains subendometrial cysts, external adenomyosis, myomas and endometrial abnormalities.

[0051] In another example, the organ of the pelvic area is the cervix, the set of anatomical structures at this organ contains the cervix, and the set of lesions that can be found at this organ contains Nabothian cysts and abnormal masses of tissue or substance.

[0052] In another example, the organ of the pelvic area is one of the ovaries, the set of anatomical structures at the level of this organ contains the ovary in question, the follicles and the corpus luteum, and the set of lesions that can be found at the level of this organ contains functional cysts (>30mm), endometriomas, mature teratomas, dermoid cysts, hemorrhagic cysts and abnormal masses of tissue or substances.

[0053] In another example, the organ of the pelvic area is one of the fallopian tubes, the set of anatomical structures at the level of this organ contains the tube in question, and the set of lesions that can be found at the level of this organ contains the hematosalpinx and the hydrosalpinx.

[0054] In another example, the organ of the pelvic area is the uterus, the pelvic pathology is posterior subperitoneal endometriosis, the set of structures anatomical at the level of this organ of interest for the pelvic pathology mentioned contains the uterine torus and the uterosacral ligaments, and the set of lesions that can be found at the level of this organ contains nodules or thickening of the torus, nodules or thickening of the uterosacral ligaments, nodules on the lateral uterosacral ligaments and sacro-recto-genito-pubic laminae, nodules of the rectovaginal septum, vaginal nodules or thickening, rectosigmoid digestive lesions, other digestive lesions (small intestine, ileo-caeco-appendicular junction), external adenomyosis and nodules or lesions with bleeding due to deep hemorrhagic implants.

[0055] In another example, the organ of the pelvic area is the uterus, the pelvic pathology is superficial endometriosis, and the lesions that can be found at the level of this organ contain the nodules or lesions with bleeding due to surface hemorrhagic implants.

[0056] In another example, the organ in the pelvic area is the peritoneum, and the lesion is a Douglas fir effusion.

[0057] In another example, the organ of the pelvic area is the bladder, the set of anatomical structures at the level of this organ contains the bladder, and the set of lesions that can be found at the level of this organ contains endometriosis of the bladder. The predefined ontology can therefore contain all or part of the organs and / or lesions cited below.

[0058] In one embodiment, the segmentation neural network adopts a fully convolutional 3D neural network model, comprising an encoder and a decoder known as U-Net. The encoder consists of convolution layers and max pooling layers aimed at generating a feature map of the images while reducing their size to decrease the number of network parameters. The decoder allows expansion along a path symmetrical to the encoder's contraction path, giving the architecture a distinctive U-shape. The U-Net offers the advantage of connecting multiple outputs of the encoder to multiple inputs of the decoder, thus allowing the extraction of information at each scale (i.e., at each block preceding a max pooling) and their assembly in the decoder. This differs from the case of the auto-encoder, which only retains information from the lowest scale.

[0059] In one embodiment, the segmentation neural network is trained with “supervised learning” with a Tversky-type loss function. The use of this loss function makes it possible to prioritize the reduction of false positives over false negatives or vice versa.

[0060] In one embodiment, the segmentation neural network is prea lably trained with “unsupervised learning” using Masked Image Reconstruction.

[0061] In one embodiment, the segmentation neural network may be a residual neural network (ResNet), which is a neural network that uses skip connections to skip certain layers. These shortcuts avoid the problem of vanishing gradients. ResNet models may be implemented with two- or three-layer skips that contain non-linear activation functions (ReLU for Rectified Linear Unit) and batch normalization between the two.

[0062] The device 1 comprises, among other things, a module 14 configured to extract a first specific region of interest from the segmented image(s) obtained at the output of the segmentation module 13 and from at least one predefined rule relating to a first pathology among said pelvic pathologies.

[0063] In one embodiment, the predefined rule relating to said first pathology consists in choosing said first region of interest according to the organ of the pelvic area which is affected by said first pathology (organ of interest). For example, in the case where the pelvic pathology is an endometrioma, the organ of the pelvic area is the ovary. In another example, in the case where the pelvic pathology is adenomyosis, the organ of the pelvic area is the uterus. Extracting the region of interest consists in extracting from the segmented image a first set of voxels labeled with the organ of interest for the first pathology, plus a second set of voxels neighboring the voxels of the first set to obtain a margin around the organ. Consequently, the region of interest (i.e., the image of the region of interest) comprises the organ of interest surrounded by a margin of predefined thickness comprising the bordering tissues.

[0064] The device 1 comprises, among other things, a module 15 which consists of a first expert model associated with said first pathology, said first expert model being configured to calculate at least one piece of information relating to the diagnosis linked to said specific region of interest. The information relating to the diagnosis may be a score indicative of the presence of a pelvic pathology.

[0065] In one embodiment, said first expert model is a neural network configured to receive as input said at least one first region of interest and to provide as output at least one score indicative of the presence of said first pathology in said patient, said at least one neural network having been previously trained on a training data set comprising annotated volumetric images relating to a group of patients, said volumetric images being acquired with at least said MRI modality, and said group of patients comprising at least one subgroup of patients affected by said first pathology.

[0066] The first expert model may be configured to ensure that the score indicative of the presence of pelvic pathology is, for example, one of the following options:

[0067] - a probability value, between 0 and 1 or expressed as a percentage, indicative of the presence of pelvic pathology, or

[0068] - a level of risk (low, moderate, high) of presence of pelvic pathology, Or

[0069] - a binary value (0 or 1) indicative of the presence or absence of a pa pelvic theology.

[0070] In one embodiment, the device is characterized in that the first expert model associated with said first pathology is configured to receive as input further clinical information (i.e. clinical data). For example, the clinical data may be examination reports or volumetric images acquired with at least one MRI modality and annotated by specialists (i.e. radiologists).

[0071] In one embodiment, the device comprises at least one output interface configured to provide at least one segmented image, at least one specific region of interest, and at least one diagnostic information.

[0072] [Fig.2] represents another functional embodiment of a device 1 for assisting in the diagnosis of pelvic pathologies in a female patient, making it possible to obtain at least one piece of information relating to the diagnosis 31.

[0073] The device 1 is in fact adapted to provide at least one piece of information relating to the diagnosis 31 of a pelvic pathology in a female patient as output. The information relating to the diagnosis may be at least one of the following information: a score indicative of the presence of a pelvic pathology, a geometric measurement relating to an anatomical structure of interest, an image with the most optimal view of a given lesion.

[0074] In this embodiment, the device 1 comprises N modules (14_i, i=1,.. ,,N; i being the ith pathology or the pathology number i); each module being adapted to extract a specific region of interest from the ith pathology. For example, the 1st module (14_1) is associated with the first pathology and the Nth module (14_N) is associated with the Nth pathology.

[0075] In this embodiment, the device 1 comprises N expert models, each of said expert models being associated with a different pelvic pathology chosen from among N different pathologies (i=1,.. .,N; i being the ith pathology or the pathology number i): the 1st expert model (15_1) is associated with the first pathology, the 2nd expert model (15_2) is associated with the second pathology, the Nth expert model (15_N) is associated with the Nth pathology, and this is so for all intermediate expert models between the 2nd expert model and the Nth expert model. As a result, a specific region of interest is extracted for each of the expert models: the region of interest extracted by the module 14_i corresponds to the expert model 15_i, i=l,...,N.

[0076] In one embodiment, each of the expert models (15_i) calculates at least one piece of diagnostic information linked to said specific region of interest extracted by the module (14_i), said diagnostic information being a score indicative of the presence of the ith pathology in said patient.

[0077] In this embodiment, said device 1 comprises at least one expert model (N+j), j=0,.. .,J, the expert model (N+j) being configured to provide at least one piece of information relating to the diagnosis, said piece of information relating to the diagnosis being at least one of the following pieces of information: a geometric measurement relating to an anatomical structure of interest or an image with the most optimal view of a given lesion.

[0078] An expert model (N+j), among the J expert models, can be configured to calculate said geometric measurement relating to an anatomical structure of interest, the geometric measurement being the diameter of said structure. Since the anatomical structures have varied geometries, the diameter of the anatomical structure can be an equivalent diameter, a minimum diameter, a maximum diameter, an average diameter, a diameter along an axis of the structure (i.e. the main axis for example along which the structure extends the most), a diameter along the main and secondary axes of the ellipse which best represents the structure.

[0079] In another example, the expert model (N+j) is configured to calculate the geometric measurement relating to an anatomical structure of interest, the geometric measurement being the diameter, length, width, surface area, circumference, volume or thickness of an organ of the pelvic area and / or a lesion. The geometric measurement relating to an anatomical structure of interest can be calculated by segmentation, geometry, contour, statistical, morphological analysis methods, by an artificial intelligence model, or any suitable image processing approach.

[0080] In one embodiment, one of the expert models (N+j) is configured to provide the image with the most optimal view of a given lesion. By image with the most optimal view, we mean the image acquired along a given axis and in which the lesion is the clearest to extract useful information for diagnosis. The axis can be chosen from the three main axes used in MRI: sagittal axis, coronal axis, transverse (or axial) axis.

[0081] In one example, an expert model (N+j) is an algorithm or a model machine learning or a neural network configured to output an image taken with the most optimal view of a given lesion.

[0082] In one embodiment, the device 1 comprises J expert models of type (N+j), j=0,.. .,J, each of the expert models of type (N+j) being configured to provide at least one piece of information relating to the different diagnosis.

[0083] In one embodiment, said device 1 is characterized in that the at least one processor is configured to calculate a combination of several pieces of information relating to the diagnosis, said pieces of information relating to the diagnosis being chosen from the different options presented in the present disclosure.

[0084] The device 1 interacts with a user interface 16, through which information can be entered and retrieved by a user. The user interface 16 comprises any suitable means for entering or retrieving data, information or instructions, including visual, tactile and / or audible capabilities which may include one or more of the following means well known to those skilled in the art: a screen, a keyboard, a trackball, a touch pad, a touch screen, a loudspeaker, a voice recognition system.

[0085] In one embodiment, the device 1 is configured to generate a visual report, designed to be integrated into medical image reading software (PACS for “Picture Archiving and Communication System”), the visual report being in the form of a series of characteristic images, originating from volumetric MRI images, on which a processor is configured to print markers (i.e. frames) around a specific region of interest as well as a block of textual data indicating at least one piece of information relating to the diagnosis extracted from at least one volumetric MRI image in connection with this specific region of interest.

[0086] In its automatic actions, the device 1 can for example execute the following method ([Fig.3]): - reception 41 of at least one volumetric image acquired with an MRI modality, said image comprising at least a portion of the pelvis of said patient, - segmentation 43 of said at least one volumetric image by a previously trained segmentation neural network, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said volumetric image based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, said volumetric images relating to the cohort of patients being acquired with at least said MRI modality, - extraction 44 of at least one first specific region of interest, from said at least one segmented image and at least one predefined rule relating to a first pathology among said pelvic pathologies, - calculation 45 of at least one piece of diagnostic information linked to said specific region of interest by at least one first expert model associated with said first pathology, said at least one first expert model being a neural network configured to receive as input said at least one first region of interest and to provide as output at least one score indicative of the presence of said first pathology in said patient, said at least one neural network having been previously trained on a training data set comprising annotated volumetric images relating to a group of patients, said volumetric images being acquired with at least said MRI modality, and said group of patients comprising at least one subgroup of patients affected by said first pathology, - provide as output said at least one piece of information relating to the diagnosis 31.

[0087] In one embodiment, a preprocessing step 42 precedes the segmentation step 43 and consists of resampling the volumetric image in a predefined spatial resolution and / or normalizing the intensities of the voxels of the image.

[0088] A particular device 5, visible in [Fig.4], implements the device 1 described above. It corresponds, for example, to a workstation, a laptop, a tablet, a smartphone, or even a “head-mounted display” (HMD).

[0089] This device 5 is adapted to provide at least one piece of information relating to the diagnosis. It comprises the following elements, connected to each other by an address and data bus 55 which also carries a clock signal: - a microprocessor 51 (or CPU);

[0090] - a graphics card 52 comprising several graphics processing units (or GPU) 520 and a graphics RAM (GRAM) 521;

[0091] - a non-volatile memory of type ROM 56;

[0092] - a RAM 57;

[0093] - one or more input / output (I / O) devices 54 such as, for example, a keyboard, a mouse, a trackball, a webcam; other modes of entering commands such as voice recognition are also possible;

[0094] - a power source 58; and

[0095] - a radio frequency unit 59.

[0096] In one embodiment, the power supply 58 is external to the apparatus 5.

[0097] The apparatus 5 also comprises a display device 53 of the screen type display directly connected to the 52 graphics card to display syn images calculated and composed in the graphics card. The use of a dedicated bus to connect the display device 53 to the graphics card 52 offers the advantage of having much higher data transmission rates and therefore reducing the latency time for the display of the images composed by the graphics card.

[0098] In one embodiment, a display device is external to the apparatus 5 and is connected thereto by a cable or wirelessly to transmit the display signals. The apparatus 5, for example through the graphics card 52, comprises a transmission or connection interface adapted to transmit a display signal to an external display means such as for example an LCD or plasma screen or a video projector. In this regard, the radio frequency unit 59 can be used for wireless transmissions.

[0099] It should be noted that the word "register" used below in the description of the memories 57 and 521 can designate in each of the memories mentioned, a memory area of ​​low capacity (some binary data) as well as a memory area of ​​high capacity (making it possible to store an entire program or to calculate or display all or part of the data representative of the data). Similarly, the registers shown for the RAM 57 and the GRAM 521 can be arranged and constituted in any way, and each of them does not necessarily correspond to adjacent memory locations and can be distributed differently (which covers in particular the case where a register comprises several smaller registers).

[0100] When powered up, the microprocessor 51 loads and executes the instructions of the program contained in the RAM 57.

[0101] As will be understood by those skilled in the art, the presence of the graphics card 52 is not mandatory, and can be replaced by complete processing by the central processing unit and / or simpler visualization implementations.

[0102] In one embodiment, the device 1 may be implemented differently than standalone software, and a device or set of devices comprising only parts of the device 5 may be operated via an API call or a cloud interface.

Claims

1. Claims Device (1) for assisting in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one piece of information relating to the diagnosis (31), said device (1) comprising: - at least one input interface configured to receive at least one volumetric image (21) acquired with an MRI modality, said image comprising at least a portion of the pelvis of said patient, - at least one processor configured to: - segmenting said at least one volumetric image (21) by a previously trained segmentation neural network, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said volumetric image (21) based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, said volumetric images relating to the cohort of patients being acquired with at least said MRI modality, - extracting at least one first specific region of interest, from said at least one segmented image and at least one predefined rule relating to a first pathology among said pelvic pathologies, - calculating at least one piece of diagnostic information (31) linked to said specific region of interest by at least one first expert model associated with said first pathology, said at least one first expert model being a neural network configured to receive as input said at least one first region of interest and to provide as output at least one score indicative of the presence of said first pathology in said patient, said at least one neural network having been previously trained on a training data set comprising annotated volumetric images relating to a group of patients, said volumetric images being acquired with at least said MRI modality, and said group of patients comprising at least one subgroup of patients affected by said first pathology, - at least one output interface configured to provide said at least one piece of information relating to the diagnosis (31).

2. Device (1) according to claim 1, characterized in that the predefined ontology comprises at least the organs of the pelvic area and the lesions.

3. Device (1) according to any one of claims 1 or 2, characterized in that the at least one processor is configured to calculate, from the segmented image of at least one structure of interest, a second piece of information relating to the diagnosis being the diameter of said structure.

4. Device (1) according to any one of claims 1 or 3, characterized in that the at least one processor is configured to calculate a third piece of information relating to the diagnosis being the most optimal view of a given lesion.

5. Device (1) according to any one of claims 1 or 4, characterized in that the at least one processor is configured to, upstream of the segmentation step, resample the image in a predefined spatial resolution and / or normalize the intensities of the voxels of the image.

6. Device (1) according to any one of claims 1 or 5, characterized in that said at least one first expert model associated with said first pathology is configured to further receive clinical information as input.

7. Method for assisting in the diagnosis of pelvic pathologies in a female patient making it possible to obtain at least one piece of information relating to the diagnosis, said method comprising: - reception 41 of at least one volumetric image acquired with an MRI modality, said image comprising at least a portion of the pelvis of said patient, - segmentation 43 of said at least one image by a previously trained segmentation neural network, said segmentation neural network being configured to perform a multi-class semantic segmentation of each voxel of said image based on a predefined ontology of the structures of interest and having been previously trained on a training data set comprising annotated volumetric images relating to a cohort of patients, said volumetric images relating to the cohort of patients being acquired with at least said MRI modality, - extraction 44 of at least one first specific region of interest, from said at least one segmented image and at least one predefined rule relating to said first pathology among said pelvic pathologies,- calculation 45 of at least one piece of diagnostic information related to said specific region of interest by at least one first expert model associated with said first pathology, said at least one first expert model being a neural network configured to receive as input said at least one first region of interest and to provide as output at least one score indicative of the presence of said first pathology in said patient, said at least one neural network having been previously trained on a training data set comprising annotated volumetric images relating to a group of patients, said volumetric images being acquired with at least said MRI modality, and said group of patients comprising at least one subgroup of patients affected by said first pathology, - providing as output said at least one piece of diagnostic information.,

8. A non-transitory computer-readable recording medium comprising instructions which, when the program is executed by a computer, cause the computer to implement the method according to the

9. Claim 7. A computer program product comprising instructions which, when the program is executed by a computer, cause the latter to implement the method according to claim 7.

Images