System for the analysis of wound images and its operating method

Abstract

The present invention relates to a system (1 ) for the analysis and the management of wounds, comprising a server (3) with a user interface (32). The serve comprises processor means (31 ) to execute an artificial intelligence algorithm trained on a database of wound images (I), wherein said processor means (31 ) are connected to said user interface (32), and said artificial intelligence algorithm is configured for: analyzing (130) one or more wound images (I); determining (140) morphological parameters of the wound (F); generating (150) reproducible measurements of said morphological parameters of the wound; and displaying (150), via said user interface (32), the characteristics and measurements of the wound (F). The present invention also relates to a method of (100) for the analysis of images, wherein said images are related to wounds and the like.

Description

[0001] SYSTEM FOR THE ANALYSIS OF WOUND IMAGES AND ITS OPERATING

[0002] METHOD

[0003] *****

[0004] The present invention relates to a system for analyzing wound images and a related operating method.

[0005] Field of invention

[0006] More specifically, the invention relates to a system based on artificial intelligence algorithms for the analysis of medical images, and more specifically to a system based on artificial intelligence (Al) for the analysis and management of wounds.

[0007] In the following, the description will be addressed to skin wounds, but it is clear that it should not be considered limited to this specific use.

[0008] Prior art

[0009] As is well known, chronic wounds currently affect millions of patients worldwide, placing a significant burden on healthcare systems. These wounds, which include pressure ulcers, diabetic foot ulcers, and venous leg ulcers, often require prolonged and complex treatment regimens. Accurate assessment and monitoring of wound healing progress are essential for optimal treatment.

[0010] Traditionally, wound assessment has relied on visual inspection and manual measurements by healthcare professionals. However, this approach is subject to variability and subjectivity and can be time-consuming, especially in crowded clinical settings. Furthermore, the subjective nature of visual assessment can lead to inconsistencies in wound documentation and treatment decisions.

[0011] The advent of imaging technologies Digital imaging has opened new possibilities for wound assessment. Digital photographs can provide a permanent record of wound appearance and allow for more objective comparisons over time. However, analyzing these images often relies on manual interpretation, which can be laborious and prone to human error.

[0012] Furthermore, increased interest in telemedicine and remote patient monitoring has highlighted the need for mobile, user-friendly solutions that can facilitate wound assessment outside of traditional healthcare settings. Such tools could improve access to specialized wound care expertise, particularly for patients in rural or disadvantaged areas.

[0013] Despite recent advances in wound care technology, several technical challenges persist. Conventional wound assessment methods often lack reproducibility and standardization across healthcare settings, leading to inconsistent measurements and subjective assessments. Manual wound measurements are time-consuming, prone to human error, as mentioned, and can therefore be particularly challenging for complex or irregularly shaped wounds. Furthermore, wound care imaging solutions Existing digital tools often require specialized equipment or extensive training, limiting widespread adoption.

[0014] Furthermore, current systems often fail to integrate wound assessment data with patient-specific factors to provide personalized treatment recommendations or predict healing pathways.

[0015] The prior art also includes systems for automatically estimating the area of a skin lesion, as reported in patent application CN20231156034. In this solution, the area estimate is based on a ratio between the pixels occupied by a marker of known size and those occupied by the lesion itself. However, the described technology does not include image depth mapping, which means it does not take into account the lesion's three-dimensional morphology. This makes the estimate inaccurate.

[0016] Other solutions according to the known art, such as those described in patent applications CN202220618519U, CN202310378513, PL20200435978, and CN202311277105 present device-specific methodologies that go beyond the use of a simple pixel ratio, but which present significant precision and accuracy challenges.

[0017] It is clear that these limitations are expensive in economic terms and in terms of service for the patient.

[0018] Purpose of the invention

[0019] In light of the above, it is therefore the purpose of the present invention to propose a system for the automated management of wounds.

[0020] Another aim of the invention is to propose a system that can be easily used and immediately implemented in current healthcare systems.

[0021] Object of the invention

[0022] It is specific object of the present invention a system for the analysis and the management of wounds, comprising a server with a user interface, comprising processor means to execute an artificial intelligence algorithm trained on a database of wound images, wherein said processor means are connected to said user interface, and said artificial intelligence algorithm is configured for: analyzing one or more wound images; determining morphological parameters of the wound; generating reproducible measurements of said morphological parameters of the wound; and displaying, via said user interface, the characteristics and measurements of the wound.

[0023] Always according to the invention, said system may comprise a mobile device, having a camera for acquiring images of wounds, wherein said wound images are acquired using said camera of said mobile device, and wherein said server is connectable to said mobile device through a network.

[0024] Still according to the invention, said morphological parameters may comprise one of the following: wound size, wound diameter, area, location, and depth of the wound.

[0025] Advantageously according to the invention, said wound dimensions may be determined using a standard marker.

[0026] Further according to the invention, the artificial intelligence algorithm may be configured for: determining a PWAT score (Pressure Wound Assessment Tool) for the wound based on the acquired image; and displaying the PWAT score via said user interface.

[0027] Preferably according to the invention, the artificial intelligence algorithm may be further configured for: estimating a volume of the wound based on the acquired image; and displaying the estimated wound volume via said user interface.

[0028] Always according to the invention, the artificial intelligence algorithm may be further configured for: recognizing a type of wound based on the acquired image.

[0029] Still according to the invention, the artificial intelligence algorithm may be one of the following types: Convolutional Neural Networks (CNN); Generative Adversarial Networks (GAN); Fully Convolutional Networks (FCN); Transposed Convolutional Networks.

[0030] It is further object of the present invention a method for image analysis, wherein said images are referred to wounds and the like, comprising the steps of: acquiring an image of a wound; receiving the acquired image through a server via a network; characterized in that said method further comprises the steps of: analyzing the acquired image using processor means using an artificial intelligence algorithm trained on a database of annotated wound images; determining morphological parameters of said wound; generating reproducible measurements of the wound characteristics; and displaying the characteristics and measurements of the wound on a user interface. Always according to the invention, said acquisition step may be performed using a mobile device equipped with a camera, and said receiving step of the acquired image may be performed via a network.

[0031] Still according to the invention, said morphological parameters may comprise one of the following: wound size, wound diameter, area, location, and depth of the wound.

[0032] Advantageously according to the invention, said method may further comprise the steps of: determining a PWAT score (Pressure Wound Assessment Tool) for the wound based on the acquired image; and displaying the PWAT score via the user interface.

[0033] Further according to the invention, said method may further comprise the steps of: estimating a volume of the wound based on the acquired image; and displaying the estimated wound volume via the user interface.

[0034] Preferably according to the invention, said method may further comprise the steps of: recognizing a type of wound based on the acquired image through access to a database; and estimating a healing time for the wound based on the patient’s characteristics and a series of wound images captured over time.

[0035] Always according to the invention, said method may further comprise the steps of: determining a BWAT score (Bates-Jensen Wound Assessment Tool) for the wound based on the acquired image; and displaying the BWAT score via the user interface.

[0036] Still according to the invention, said step of generating reproducible measurements of said wound features may comprise the following sub-steps: determining a two-dimensional depth map M(x y) for calculating the area of an irregular surface of the wound, in three dimensions, to have the depth of the region of interest; positioning a marker of known dimensions, such as a circle or the like, having a determined diameter D near the region of interest; fitting the longest straight segment contained by the marker using a known curve, such as a parabolic arch and the like; calculating the length of said segment; and calculating the area of the irregular surface by multiplying the maximum width (w) and the average height (h) of the wound.

[0037] Advantageously according to the invention, said method may comprise the step of determining scaling factors to adjust the measure said maximum (w) and average height (h) of the wound. It is also object of the present invention a computer program comprising instructions that, when executed by a computer, cause the computer to perform the steps of the method mentioned above.

[0038] It is further object of the present invention a computer-readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform the steps of method mentioned above.

[0039] Brief description of the figures

[0040] The present invention will be now described, for illustrative but not limitative purposes, according to its preferred embodiments, with particular reference to the figures of the enclosed drawings, wherein: figure 1 shows a functional block diagram of the wound analysis system; figure 2 shows a flowchart of an image analysis method according to the present invention; and figure 3 shows a flowchart of further steps of the image analysis method of figure 2; figure 4 shows a flowchart of further steps of the image analysis method of figure 2; figures 5(a), 5(b) and 5(c) depict elements for the validation of acquired wounds; and figure 6 shows a graph showing the degree of accuracy of the measurements taken.

[0041] Detailed description

[0042] In the various figures, similar parts will be indicated with the same numerical references.

[0043] The present invention relates to a system for analyzing wound images that implements a combination of mobile technology and artificial intelligence.

[0044] The system may generally comprise a mobile device equipped with a camera to capture or acquire images of the wounds for analysis. The acquired images may be transmitted to a server via a network. The server may include processing means to run an artificial intelligence algorithm trained on a database of wound images.

[0045] The Al algorithm can be configured to analyze wound images, determine wound morphological parameters, generate reproducible measurements of these parameters, and visualize wound characteristics and measurements via a user interface. Referring to figure 1 , a wound image analysis system 1 is shown according to one embodiment comprising a mobile device 2 equipped with a preferably high- definition camera 21. The mobile device 2 may be any type of portable electronic device, such as a smartphone or tablet, capable of acquiring images. The camera 21 may be an integral part of the mobile device 2 and may be used to acquire wound images for analysis.

[0046] The system 1 also comprises a server 3, connected to the mobile device 2 via a transmission network 4. The transmission network 4 is adapted to enable the transfer of data, such as images of the wounds to be analyzed, between the mobile device 2 and the server 3. The transmission network 4 may be any type of data transmission network, such as a cellular network, a Wi-Fi network, or the Internet.

[0047] The server 3 comprises processor means 31 , for executing an artificial intelligence (Al) algorithm 33, and a user interface 32.

[0048] The user interface 32 may be a graphical user interface (Graphic User Interface- GUI) that allows users to interact with the system 1 . The user interface 32 may be accessible via a web browser or a dedicated application on the mobile device 2 or another device.

[0049] The processor means 31 may be a computer processor or a set of computer processors configured to execute the Al algorithm 33. Alternatively, the processor means 31 may be a suitably programmed personal computer.

[0050] The Al algorithm 33 can be trained on a database of wound images and associated metadata, which may have been annotated and curated by clinical dermatology experts.

[0051] The Al algorithm 33 may be a machine learning algorithm, such as a Convolutional Neural Network (CNN), a Generative Adversarial Network (GAN), a Fully Convolutional Neural Network (FCN), or a Transposed Convolutional Neural Network (TCN).

[0052] The Al algorithm 33 is configured to analyze wound images acquired by the camera 21 of the mobile device 2.

[0053] This analysis involves first segmenting the F1 area of the wound F in the image, and then extracting the relevant features of the wound and performing various evaluations based on the extracted features.

[0054] Continuing with the description in figure 1 , the Al algorithm 33 can be configured to determine various morphological parameters of the wound. These parameters may include, but are not limited to, wound dimensions, e.g., diameter or length, area, location, and depth, as well as wound colors. In some cases, the Al algorithm 33 may analyze the captured image to identify these parameters, providing a comprehensive analysis of the physical characteristics of the acquired wound F.

[0055] The Al algorithm 33 can be further configured to generate reproducible measurements of the aforementioned morphological parameters. This allows for monitoring of wound progression over time, regardless of the operator or the context in which the image is acquired. In particular, this functionality can be particularly useful in various settings such as home visits, telemedicine, nursing clinics, and hospitals.

[0056] The Al algorithm 33 may also be configured to display wound characteristics and measurements via the user interface 32. This may include displaying the determined morphological parameters and generated reproducible measurements. The user interface 32 may present this information in an intuitive manner, allowing healthcare providers or other users to easily understand and interpret wound characteristics and progress F over time.

[0057] In some embodiments, the Al algorithm 33 may be further configured to determine a Pressure Wound Assessment Tool (PWAT) score for wound F based on the acquired image.

[0058] The PWAT score is a numerical "score" used to describe and monitor wounds over time, and is widely used in clinical practice. The Al algorithm 33 can analyze the wound F image and associated data to calculate the PWAT score, thus providing an additional level of analysis for the wound F itself.

[0059] The obtained PWAT score can then be viewed via the user interface 32, providing users with a complete overview of the wound status F.

[0060] The operation of system 1 described above is as follows.

[0061] Referring to figure 2, a flowchart of a method 100 for the analysis and management of wounds F is observed, which includes several phases.

[0062] In an initial step 110 of the operating method 100, an image I of a wound F is acquired using the camera 21 of the mobile device 2.

[0063] After acquiring image I, the method 100 comprises step 120 of receiving image file I from a server 3 via a network transmission 4.

[0064] Once the digital image I of the wound F has been received, the method 100 comprises the step 130 of analyzing the image I by means of processor means 31 , in particular by means of an artificial intelligence algorithm 33.

[0065] The processor means 31 may be a computer processor or a set of computer processors configured to execute the Al algorithm 33.

[0066] The Al algorithm 33 can be trained on a large database of wound images and associated metadata, which may have been annotated and curated by dermatology clinical experts. These training images may be stored in a database 34. Additionally, newly acquired images may be saved in the same database 34 and, following the processing, may be used for continuous training of the Al algorithm 33.

[0067] In some embodiments, the images can be provided to the Al algorithm 33 even without the aid of a mobile device 2. For example, said images can be acquired using a camera and saved in an SD memory or the like, to be read and processed by the server 3.

[0068] In step 140 of the operating method 100, morphological parameters of the wound may be determined. These parameters may include, but are not limited to, the size of the wound F, the diameter or length of the wound F, the area, location, and depth of the wound F, and the color or colors of the wound F.

[0069] In some cases, the Al algorithm 33 can analyze the acquired the image I to identify these parameters, thus providing a complete and real-time analysis of the physical and morphological characteristics of the wound F.

[0070] After determining the morphological parameters, method 100 involves step 150, where reproducible measurements of wound characteristics are generated. This can enable consistent and reliable monitoring of wound progression over time, regardless of the operator or the context in which the image is acquired.

[0071] Reproducible measurements provide a standardized method or criterion for recording wound progression, as well as some time-varying trends in morphological parameters, which can be particularly useful, as mentioned, in various settings such as home visits, telemedicine, nursing clinics, and hospitals.

[0072] Still referring to figure 2, the operating method 100 may also include the step 170, where the PWAT (Pressure Wound Assessment Tool) score for the wound F being analyzed is determined based on the acquired image.

[0073] Additionally, method 100 may include steps 160 and 180 of displaying wound characteristics and measurements on user interface 32.

[0074] In some embodiments, the Al algorithm 33 may also comprise instructions to estimate a healing time for the wound F based on patient characteristics and a series of images h , ... , In of the wound acquired at different corresponding time points ti , ... tn, thus allowing healthcare providers to optimize check-up dates and adjust treatment plans as needed.

[0075] Referring now to figure 3, the method of operation 100 of the wound F image analysis system 1 may, furthermore, comprise further steps.

[0076] In particular, the method 100 comprises the step 141 of estimating the volume of the wound F based on the acquired image I. This may involve analyzing the image to identify the “boundaries” F1 and the depth of the wound F itself, and then calculating the volume based on these parameters.

[0077] The Al algorithm 33 can use various image analysis techniques, such as edge detection F1 , segmentation, and depth estimation, to estimate the wound F volume.

[0078] After estimating the wound F volume, the method 100 provides for the visualization of the volume 181 , where the estimated wound volume is displayed on the user interface 32. In this case, the user interface 32 may provide programs for the 3D representation of the wound, to allow for a faster analysis of the same.

[0079] The operating method 100 comprises further steps, as shown in figure 4, including the recognition (step 142) of the wound F type of the acquired image I. In this case, the Al algorithm 33 performs the comparison with the images stored in the database 34 of known wound types.

[0080] Method 100 may also comprise the step 182 of displaying the estimated wound F volume using Al algorithm 33.

[0081] In some embodiments, the Al artificial intelligence algorithm 33 of system 1 may comprise an automated Al model for automatically estimating the wound F area or surface area by inserting a standard marker at the time of image acquisition F.

[0082] This may involve the Al algorithm 33 to analyze the acquired image I to identify the wound F area and the standard marker, and then calculating the wound F area based on the size of the standard marker. The standard marker may be a known object with fixed dimensions, such as a 3 cm diameter green dot, which can be used as a reference for size estimation.

[0083] In some aspects, the Al model 33 is capable of performing a 3D reconstruction of wound F from a single 2D photo. The Al algorithm 33 analyzes the acquired image I to identify the boundaries and depth of wound F, and then generates a 3D model of wound F based on these parameters. The 3D reconstruction of the wound provides a more complete and detailed view of the wound. The 3D wound model can be displayed on the user interface 32, providing users with a visual representation of the physical characteristics of the wound.

[0084] In some embodiments, the Al algorithm 33 can also determine the Bates- Jensen Wound Assessment Tool (BWAT) score. The BWAT score is a comprehensive wound assessment tool that provides a broad set of information for clinical use. The Al algorithm 33 can analyze the wound image and associated data to calculate the BWAT score, providing an additional level of analysis for the wound. The BWAT score can then be displayed via the user interface 32.

[0085] In some embodiments, the Al algorithm 33 may be configured to identify the patient's body part of wound F in image I.

[0086] Identifying the body part can provide information about the location of the wound F.

[0087] In some cases, system 1 may have the potential to suggest dressing types for wound treatment based on specific guidelines. In other words, Al algorithm 33 analyzes wound image I and associated data to identify wound F type and its characteristics, and then suggests a suitable dressing type based on these parameters and the guidelines. The suggested dressing type may also be displayed on user interface 32, providing users with a recommended treatment strategy for wound F.

[0088] In addition to the above, system 1 may indicate the primary etiological nature of the wound, such as venous, arterial, mixed, or rheumatic. In this case, the Al algorithm 33 analyzes image I of the wound F and the associated data to identify its type and characteristics, and then determine the probable etiological nature of the wound based on these parameters. The determined etiological nature of the wound may be displayed on the user interface 32.

[0089] The system 1 is designed to allow any user, including non-healthcare users, to use mobile device 2 to acquire images I of wounds F. This makes the system 1 very versatile, allowing it to be used in a wide range of settings and by a wide range of users.

[0090] For example, patients or caregivers can use the mobile device 2 to capture wound images for self-monitoring or family wound monitoring.

[0091] This is also useful for home care, where professional caregivers may not always be immediately available.

[0092] In some embodiments, system 1 may comprise a web interface for convenient inspection of data collected in the clinic or more generally for one's account. The web interface may be accessible via a web browser on mobile device 2 or another device, and may allow viewing and management of wound data.

[0093] The web interface can display wound characteristics and measurements, PWAT score, estimated wound volume, and other relevant data.

[0094] Referring now to figure 5, we can observe elements for the validation of the acquired wounds. In particular, in the two boxes (a) we can observe the automatic segmentation of the measurement acquired from the clinical data and used as ground truth of the area estimation, obtained from the ratio between the area of the lesion traced on graph paper and the area of the single square of the paper. Instead, in box (b), we observe a depth map generated through the Al artificial intelligence algorithm 33 applied to a single image I.

[0095] Finally, in box (c) we observe the automatic segmentation of the marker (circular outline) and of the wound(s) (irregular outline(s)).

[0096] As mentioned, the method 100 allows calculating the extent of the lesion in relation to a marker reference mark placed near the lesion or wound I.

[0097] Taking into account the curvature of the anatomical region and any distortion effects of the photograph, the method 100 guarantees the estimation of the lesion area with an absolute median error (EMA) less than 1 cm2(EMA of 0.93 cm2, interquartile range of 1.52 cm2), compatible with the level of manual accuracy currently achieved by clinicians, and with excellent linear correlation with actual area measurements (Pearson correlation of 0.947 with p-value < 0.0001 , see also figure 6).

[0098] The prognostic importance of the extent of superficial lesions, combined with the high (and intrinsic) human subjectivity in its manual estimation, are the main factors in which this method aims to provide an advantage. Currently, the manual estimation carried out during clinical practice involves the manual tracing of the lesion geometry by placing a material on the damaged surface and thus causing discomfort to the patient and / or preventing its use in specific cases. The ability to obtain the same information without any direct contact with the lesion (except for the placement of the marker) ensures greater speed in estimating this parameter, greater applicability for quantifying this indicator, minimizing contact with (and therefore less discomfort for) the patient.

[0099] Furthermore, the method according to the invention has currently been applied and tested on skin lesions caused by chronic wounds and ulcers.

[0100] The minimum requirements for the applicability of method 100 are in fact dictated by the availability of an image I in digital format and by the segmentation of the three main regions of interest: (i) the portion of the image whose extension one wishes to determine, (ii) the portion of the image occupied by the marker, (iii) the portion of the image occupied by the surface on which the region indicated in point (i) lies. By varying the initial hypotheses on the geometry of the base surface (which in the present method 100 according to the invention have been approximated to ellipsoidal, in order to model the majority of body surfaces) it is in fact possible to extend this methodology to other contexts as well.

[0101] The results presented in figure 6 were obtained on a cohort of 92 volunteers / patients, by processing with the proposed algorithm the photos of the lesions acquired during the standard clinical procedure, without imposing any constraints during the acquisition either in terms of brightness / exposure of the photo, nor in terms of zoom and / or orientation. Before the acquisition of each image, it was requested to place a marker circular in shape with a diameter of 1 cm (circular), provided that it is standardized and fixed for data collection.

[0102] Each image was processed using the Al algorithm 33 for automatic lesion surface identification, marker identification (ad hoc model), and depth estimation. The combination of this information was provided as input to the algorithm, in order to obtain an estimate of the lesion area with precision in the order of one square millimeter (mm2). These values were compared with the areas determined manually starting from the annotations on graph paper of the lesion tracings carried out in situ on the patient: these last paper clippings were photographed and processed using Computer Vision techniques for the estimation of the ground truth with pixel-level precision and reported in units of square millimeters (mm2) by comparing the ratio with the dimensions of the single unit square of the map. The linear regression and the quantification of the percentage errors committed by comparing the actual lesion estimate and the one automatically predicted by this algorithm are statistically significant.

[0103] Always referring to figure 6, we can see a graph showing the correlation between the area estimated at the end of the procedure and the area actually measured for each wound, the Pearson correlation index is particularly high (0.947) and the p-value of the test associated with this statistic is particularly significant (< 0.0000001 ).

[0104] The following is a procedure for calculating the area of irregular surfaces using a 2D digital image, an estimated depth map, and a reference sample (in this case, a circle of known diameter) placed near the region of interest where the shape to be measured is located.

[0105] In particular, given the marker of known dimensions, (in our case a circular marker with diameter D), the procedure aims to estimate the area A of a given irregular surface S in 3D space. A_REFis the area o S_REF, the surface area relative to the marker of known dimensions.

[0106] First, we consider that, given a 2D representation, such as a digital image, it is not possible to reconstruct the original 3D space with exact precision, and for this reason, references of known dimensions are required. Finally, as mentioned, we require an estimated 2D depth map, denoted by M(x;y), with a pixel-to-pixel correspondence to the 2D binary digital image / (x;y) of the region of interest (ROI), where each pixel has coordinates (x;y).

[0107] The elements needed to start the estimate calculation are the following:

[0108] - a 2D digital image / (x;y) of the ROI;

[0109] - a 2D depth map M(x;y);

[0110] - a marker of known dimensions, in our case specifically a circular marker of known diameter D.

[0111] The depth map needs to be converted M(x; y), measured in arbitrary units, and / (x;y), in real length measurements. This step is achieved using a marker of known dimensions, and specifically for a circular marker we can follow the procedure reported in the following paragraph.

[0112] We then need to find the conversion constants from pixels to real units of measurement. To do this, we first identify the longest, straightest horizontal segment contained within the 2D marker. This segment, if no excessive curvature is applied to the underlying surface, will pass through the center of the marker, and also identify a diameter. This segment will be a straight segment in 2D space of / (x;y), but it will become (approximately) a parabola arc in the 3D space obtained by considering the respective depth z from M(x;y). We can fit this set of points with a general equation of a parabola on the plane

[0113] (x; z): z = ArA2 + Bx + C, and impose x" e " [0; TV] with x e N without loss of generalization from adaptation.

[0114] Once identified the equation for that parabola arc, you can calculate its length using:

[0115] However, knowing the size of the marker, it is possible to impose L(x) = D and therefore the result of 2 (which numerically will not correspond to D due to the fact that it is measured in pixels and arbitrary units) can be converted to the same unit of measurement as£)(both cm and mm) using the conversion factor o_x given by: where the fixed values are being used for A, B and N since they are those obtained by adapting the parabola arc to the diameter segment.

[0116] Now it is possible to convert any straight horizontal segment of I in a curve also considering the depth to be M, fit that curve with a parabola arc, calculate its length using numerical integration of 2 and multiply that result by axto get its length in the same unit of measurement as D. The diameter fitting procedure can also be reproduced on vertical diameters to find the scale factor ay.

[0117] After obtaining the horizontal and vertical scale factors ax, ay, the areas need to be converted.

[0118] The approach chosen in this embodiment is to take the average wound F height and multiply it by the maximum wound width, both multiplied by their respective scaling factors. This procedure is a computational approximation to the trapezoidal rule for integration: and

[0119] , S" o h=~N~ with w that represents the maximum width of the wound F and h the average height of the wound F.

[0120] Further considerations can be made to better fit the estimated area to the predicted area. For example, tests were performed on a dataset of 105 real wounds collected from 92 images. These wounds had a ground truth labeled on a sheet of paper, tracing the perimeter of the wound on a sheet of graph paper. To better fit this hypothesis, it is possible to substitute the second degree polynomial of the above equation (z = AxA2 + Bx + C) with a first-degree polynomial (to mimic the flat surface of graph paper). Finally, a multiplicative conversion factor can be applied to better fit the actual measured areas, as long as the correlation between the actual and predicted areas is sufficiently good. Therefore, performance was evaluated in terms of the correlation between the actual and estimated areas and in terms of the median absolute error between the two areas.

[0121] The solution described has broad industrial applicability in the healthcare and medical technology sectors. Although the invention has been described primarily in the context of clinical wound analysis and management, it is not limited to these applications. The Al-based wound image analysis system 1 may be useful in various industries and fields requiring assessment and monitoring of wounds or similar superficial skin abnormalities.

[0122] The solution described could find applications in veterinary medicine for the analysis of animal wounds, in the cosmetics industry for the assessment of skin conditions, or in forensic science for the documentation of injuries. The mobile and cloud-based nature of the system also makes it suitable for telemedicine applications across various industries.

[0123] System 1 's data collection and analysis capabilities can be useful for medical research, pharmaceutical development, and health policy planning. Advantages

[0124] An advantage of the wound image analysis system of the present invention is the integration of an artificial intelligence algorithm trained on a large database of annotated wound images, which enables reproducible measurements of wound characteristics. An additional advantage of the present invention is that generating reproducible measurements of morphological parameters such as wound size, area, location, and depth provides healthcare professionals with standardized data that can be monitored over time.

[0125] The present invention has been described for illustrative but not limitative purposes, according to its preferred embodiments, but it is to be understood that modifications and / or changes can be introduced by those skilled in the art without departing from the relevant scope as defined in the enclosed claims.

Claims

CLAIMS1. System (1 ) for the analysis and the management of wounds, comprising a server (3) with a user interface (32), characterized in that said server (3) comprises processor means (31 ) to execute an artificial intelligence algorithm trained on a database of wound images (I), wherein said processor means (31 ) are connected to said user interface (32), and in that said artificial intelligence algorithm is configured for: analyzing (130) one or more wound images (I); determining (140) morphological parameters of the wound (F); generating (150) reproducible measurements of said morphological parameters of the wound; and displaying (150), via said user interface (32), the characteristics and measurements of the wound (F).

2. System (1 ) according to claim 1 , characterized in that it comprises a mobile device (2), having a camera (21 ) for acquiring images of wounds, wherein said wound images (I) are acquired using said camera (21 ) of said mobile device (2), and wherein said server (3) is connectable to said mobile device (2) through a network (4).

3. System (1 ) according to any one of the preceding claims, characterized in that said morphological parameters comprise one of the following: wound size, wound diameter, area, location, and depth of the wound.

4. System (1 ) according to the preceding claim, characterized in that said wound dimensions are determined using a standard marker.

5. System (1 ) according to any one of the preceding claims, characterized in that the artificial intelligence algorithm (33) is configured for: determining () a PWAT score (Pressure Wound Assessment Tool) for thewound based on the acquired image; and displaying (180) the PWAT score via said user interface (32).

6. System (1 ) according to any one of the preceding claims, characterized in that the artificial intelligence algorithm (33) is further configured for: estimating (141 ) a volume of the wound based on the acquired image; and displaying (181 ) the estimated wound volume via said user interface (32)..

7. System (1 ) according to any one of the preceding claims, characterized in that the artificial intelligence algorithm (33) is further configured for: recognizing (142) a type of wound based on the acquired image.

8. System (1 ) according to the preceding claim, characterized in that the artificial intelligence algorithm (33) is one of the following types:- Convolutional Neural Networks (CNN);- Generative Adversarial Networks (GAN);- Fully Convolutional Networks (FCN);- Transposed Convolutional Networks.

9. Method (100) for image analysis, wherein said images are referred to wounds and the like, comprising the steps of: acquiring (110) an image (I) of a wound (F); receiving (120) the acquired image (I) through a server (3) via a network (4); characterized in that said method (100) further comprises the steps of: analyzing (130) the acquired image (I) using processor means (31) using an artificial intelligence algorithm trained on a database (34) of annotated wound images; determining (140) morphological parameters of said wound; generating (150) reproducible measurements of the wound characteristics; and displaying (160) the characteristics and measurements of the wound on a user interface (32).

10. Method (100) according to the preceding claim, characterizedin that said acquisition step (110) is performed using a mobile device (2) equipped with a camera (21 ), and in that said receiving step (120) of the acquired image (I) is performed via a network (4).

11. Method (100) according to any one of claims 9 or 10, characterized in that said morphological parameters comprise one of the following: wound size, wound diameter, area, location, and depth of the wound.

12. Method (100) according to any one of the preceding claims, characterized in that it further comprises the steps of: determining (170) a PWAT score (Pressure Wound Assessment Tool) for the wound based on the acquired image; and displaying (180) the PWAT score via the user interface (32).

13. Method according to any one of claims 9-12, characterized in that it further comprising the steps of: estimating (141 ) a volume of the wound based on the acquired image; and displaying (181 ) the estimated wound volume via the user interface (32).

14. Method according to any one of claims 9-13, characterized by further comprising the steps of: recognizing (142) a type of wound based on the acquired image through access to a database; and estimating (143) a healing time for the wound based on the patient’s characteristics and a series of wound images captured over time.

15. Method according to any one of claims 9-14, characterized by further comprising the steps of: determining a BWAT score (Bates-Jensen Wound Assessment Tool) for the wound based on the acquired image; and displaying (180) the BWAT score via the user interface (32).

16. Method (100) according to any one of claims 9-15, characterized in thatsaid step of generating (150) reproducible measurements of said wound features comprises the following sub-steps: determining a two-dimensional depth map M(x;y) for calculating the area of an irregular surface of the wound, in three dimensions, to have the depth of the region of interest; positioning a marker of known dimensions, such as a circle or the like, having a determined diameter D near the region of interest; fitting the longest straight segment contained by the marker using a known curve, such as a parabolic arch and the like; calculating the length of said segment; and calculating the area of the irregular surface by multiplying the maximum width (w) and the average height (h) of the wound.

17. Method (100) according to the preceding claim, characterized in that it comprises the step of determining scaling factors to adjust the measure said maximum (w) and average height (h) of the wound.

18. Computer program comprising instructions that, when executed by a computer, cause the computer to perform the steps of the method according to any one of claims 9-17.

19. Computer-readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform the steps of method according to any one of claims 9-17.

Images