System for triage and identification of risk of complications in the lower limbs of diabetes carriers using machine learning

Description

001. The present invention relates to a process for evaluating lower limbs in diabetic patients using machine learning techniques to analyze single or sequential images of the visible light spectrum, individually or in groups, captured from at least one of the following perspectives: anterior, posterior, inferior, medial, lateral, and superior, integrated or not with textual data, to determine the risk of developing lower limb complications associated with diabetes. More specifically, the present invention relates to a system for detecting the risk of lower limb complications in diabetic patients through an automated process that combines images captured in the visible light spectrum, which may be complemented by thermal images in the infrared spectrum and three-dimensional point clouds obtained by LIDAR.The analysis of these different data sources, when combined with the patient's clinical history and trained language models (LLM) for automated report generation, results in a comprehensive assessment tool capable of detecting risk factors even before the development of ulcers and other complications. The present invention falls within the field of medical technology. Therefore, the present invention has direct application in specialized clinics, hospitals, primary care services, telemedicine systems, and public health programs, assisting in reducing the incidence of serious diabetic foot complications. FUNDAMENTALS OF THE TECHNIQUE 002. The problem of lower limb complications in diabetic patients is a critical public health issue, given its high prevalence and negative impact on patients' quality of life. The risk Petition 870250062739, dated 07 / 21 / 2025, page 8 / 32 The incidence of foot ulcers, infections, and amputations is significantly higher in diabetic individuals, with peripheral neuropathy, poor blood circulation, and compromised immune systems being key contributing factors to the development of these complications. Studies reveal that up to 50% of patients with type 2 diabetes will experience some type of foot ulcer during their lifetime, and that approximately 20% of these cases will lead to amputations, increasing the mortality and morbidity associated with the disease (ARMSTRONG, David G; BOULTON, Andrew JM; BUS, Sicco A. Diabetic foot ulcers and their recurrence. A Review N Engl J Med, v. 376, n. 24, p. 2367-2375, 15 jun. 2017. DOI: 10.1056 / NEJMra1615439). Furthermore, the difficulty in early diagnosis and the lack of continuous monitoring contribute to the progression of these complications, making them more difficult to treat when detected in more advanced stages. 003. Approximately 25% of people with diabetes will develop foot ulcers at some point in their lives, with an incidence rate of 2.5% per year. (BOULTON, Andrew JM; VILEIKYTE, Loretta; RAGNARSON-TENNVALL, Gunnel; APELQVIST, Jan. The global burden of diabetic foot disease. Lancet, v. 366, n. 9498, p. 1719-1724, 12 nov. 2005. DOI: 10.1016 / S0140-6736(05)67698-2.). Complications related to diabetic foot are responsible for more than 70% of non-traumatic lower limb amputations, which severely impacts patients' quality of life and results in high costs for healthcare systems. 004. Another important challenge is the scarcity of effective tools for early monitoring and assessment of lower limb conditions, which can lead to late interventions. Early detection of changes in the feet and legs can prevent progression to serious complications, but most diagnostic systems still rely on subjective clinical examinations and visual assessment, which can be inaccurate or fail to identify subtle changes. According to a study published in the journal Diabetes Care (2019), more than 60% of patients with type 1 and type 2 diabetes do not Petition 870250062739, dated 07 / 21 / 2025, page 9 / 32 3 / 24 undergo regular foot assessments, which increases the risk of complications. In this context, more innovative approaches, such as the use of technologies based on artificial intelligence, computer vision, and machine learning, have shown great potential to improve the accuracy and speed of diagnosis, allowing for more efficient monitoring and early intervention before complications become irreversible. 005. The term diabetic foot refers to the various alterations and complications that can occur, in isolation or in combination, in the feet and lower limbs of people with diabetes. These complications result in high human and financial costs. Prevention of this condition depends on effective disease control, combined with the implementation of simple preventive care measures, early diagnosis, and appropriate treatment in the early stages of the disease (FERNANDES, Fábia Cheyenne Gomes de Morais et al. Foot care and ulcer prevention in diabetic patients in Brazil. Cad. saúde colet., v. 28, n. 2, p. 258-268, Apr.-Jun. 2020. DOI: 10.1590 / 1414-462X202028020258.) STATE OF THE ART 006. The specialized technical literature reveals some patent documents that refer to the monitoring and diagnosis of diabetic foot. 007. Documents found in the prior art reveal various approaches to monitoring and diagnosing diabetic foot, highlighting the limitations that the present invention seeks to overcome. The Diabetic Foot Analysis and Management System Apparatus (WO2022158848A1), for example, uses machine learning to process data captured by sensors in shoes or insoles, but depends on specific hardware. However, the proposed invention adopts conventional and thermal sensors, reducing costs and increasing accessibility through a more versatile and scalable system. 008. Another significant advance is the Diabetic Foot Inspection Device (EP1518497A1), which employs a semi-transparent screen and lighting to Petition 870250062739, dated 07 / 21 / 2025, page 10 / 32 4 / 24 facilitate the visual inspection of peripheral neuropathies. However, its reliance on manual interpretation can lead to diagnostic variations. The present invention overcomes this limitation by automating image analysis through machine learning, providing objective and standardized results, and integrating clinical information for greater accuracy. Furthermore, it does not depend on specific physical devices, such as a semi-transparent screen, but rather on widely available mobile technologies. 009. Approaches such as the SMART Foot Insole (Faiz, 2024), which measures plantar pressure in real time, and the Diabetic Foot Testing Device and Method (JP2005533543A), focused on two-dimensional images, also presented for diabetic foot monitoring, have limited application to isolated aspects, such as pressure or visual inspection. The proposed technology goes beyond the combination of analysis from multiple perspectives (anterior, posterior, medial, lateral, and inferior) with artificial intelligence algorithms, allowing for the early detection of structural and dermatological changes together. 0010. Patent document CN117334344 describes a method for screening diabetic feet using photoplethysmography (PPG) signals and deep learning. The model is based on obtaining and pre-processing physiological signals to train neural networks capable of identifying the risk of diabetic foot. Although effective in settings with limited medical resources, its scope is restricted to the analysis of PPG signals, without integrating images, clinical data, or medical histories. 0011. Patent document CN116645347 proposes a method for predicting diabetic foot conditions through feature fusion from multiple views, including thermal and color images. The technology is based on feature optimization modules, such as skin texture and temperature distribution. Despite its efficiency, the approach is limited to the use of images from the same angle, without considering the acquisition of images from multiple angles and their combined use. Furthermore, it is restricted to multiview data fusion, without offering support for analysis. Petition 870250062739, dated 07 / 21 / 2025, page 11 / 32 5 / 24 based on temporal sequences or integration of textual information, such as clinical history. Finally, it also does not take into account the detection of structural changes and deformities of the foot through the use of images. 0012. Patent document CN113537300 discloses a method for classifying ischemia and infection in diabetic foot wounds using deep neural networks and advanced training strategies. Despite its accuracy in wound imaging, the application is limited to specific diagnoses of ischemia or infection, without considering structural deformities or dermatological conditions, and without generating a comprehensive risk report. 0013. Patent document CN112244803 presents a portable diabetic foot risk detection device based on PPG signals, designed to be economical and easy to operate. While useful for early warnings, the solution is limited to the analysis of PPG signals and basic physiological parameters, offering no support for broader visual or predictive diagnoses. 0014. Document IN202241048550 describes a non-invasive system for real-time detection of abnormalities in the diabetic foot. The system is designed to identify diabetic wounds in a general way, associated with an examination of the sole of the foot designed to alert the user to the need for immediate medical consultation, as well as allowing periodic self-assessment. Its scope, however, is not clearly defined beyond the comparison between thermographic images of both feet and the search for active ulcers. The proposed invention covers dermatological and structural conditions associated with the risk of ulcer development, such as calluses, hammer toe, fissures, and also integrates LLM to provide clinical reports. 0015. Finally, systems that use machine learning, such as those described in ulcer detection studies (AKEBOSHI, Wynnezel Wayne Naoto et al. WoundDAR: LiDAR and Machine Vision Based Wound Assessment. In: IEEE 14th INTERNATIONAL CONFERENCE ON HUMANOID, NANOTECHNOLOGY, INFORMATION TECHNOLOGY, COMMUNICATION AND CONTROL, ENVIRONMENT, AND MANAGEMENT, HNICEM 2022, Petition 870250062739, dated 07 / 21 / 2025, page 12 / 32 6 / 24 Boracay Island, Philippines, December 1-4, 2022. Proceedings... IEEE, 2022. p. 1-6. DOI: 10.1109 / HNICEM57413.2022.10109427. ISBN 978-166546493-2), offer advanced diagnostics, but take a comprehensive predictive approach. The present innovative invention incorporates predictive dermatological and structural analyses, identifying early signs of risk before the clinical manifestation of lesions. This multidimensional integration, coupled with personalized clinical reports, represents a holistic solution for the diagnosis and prevention of diabetic foot, promoting early and effective interventions. 0016. Thus, and based on the documents revealed, it is highlighted that the development and application of the aforementioned 'SYSTEM FOR SCREENING AND IDENTIFYING THE RISK OF COMPLICATIONS IN THE LOWER LIMBS OF DIABETIC PATIENTS USING MACHINE LEARNING' proposed in this patent application has innovative potential and, therefore, meets the criterion of novelty, since to date no scientific or technical work included in the state of the art possesses the technology for development, obtaining and application similar to that which is revealed and requested. 0017. Finally, it is important to highlight that, although the technologies known in the state of the art provide relevant contributions to the detection and monitoring of diabetic foot, the present invention differs by uniquely integrating multiple data sources, including images captured from different perspectives and clinical data, associated or not with thermographic images and three-dimensional point clouds produced by LIDAR, applied to machine learning algorithms and language models trained for the automated generation of clinical reports. This innovative approach provides a more accurate, preventive, and accessible diagnosis, optimizing the early identification of conditions associated with diabetic foot and offering effective support for both healthcare professionals and patients in its management. Petition 870250062739, dated 07 / 21 / 2025, page 13 / 32 7 / 24 ADVANTAGES OF THE INVENTION 0018. Regarding the advantages and differentiating features presented by the aforementioned 'SYSTEM FOR SCREENING AND IDENTIFYING THE RISK OF COMPLICATIONS IN THE LOWER LIMBS OF DIABETIC PATIENTS USING MACHINE LEARNING', the most relevant can be highlighted as follows: 0019. Image acquisition in varying spectra using devices that capture images in the visible light spectrum, associated or not with devices that perform infrared and / or three-dimensional point cloud image capture, providing a more detailed analysis of the conditions of the lower limbs. 0020. Transforming health data and images into a risk analysis with scalar or vector results, facilitating the prediction of amputations in diabetic patients. 0021. The process uses machine learning techniques that allow for the automated evaluation of images, in order to accurately and efficiently identify risk factors associated with diabetic foot complications. 0022. Allows for the early identification of complications by focusing on screening, monitoring, and prevention of clinical conditions of the diabetic foot, such as deformities, dermatological changes, neuropathies, ischemia, and infections, contributing to the reduction of ulcers and, consequently, amputations. 0023. The technology can be applied in clinical, hospital, and home environments, allowing healthcare professionals and patients to use accessible devices such as RGB cameras, with or without the addition of devices with infrared sensors and LIDAR technology. 0024. The process offers a precise and non-invasive analysis, ensuring a safer and more comfortable diagnosis for patients. Petition 870250062739, dated 07 / 21 / 2025, page 14 / 32 8 / 24 0025. Automated reporting and personalized recommendations optimize preventive care and facilitate remote follow-up, improving efficiency in monitoring patient health. 0026. It has direct application in specialized clinics, hospitals, primary care services, telemedicine systems, and public health programs, helping to reduce serious complications of diabetic foot. 0027. Technology facilitates more effective and accessible treatment, contributing to improved quality of care for diabetic patients. 0028. The system uniquely integrates multiple data sources, including images captured from different perspectives, associated or not with thermographic images and / or three-dimensional point clouds produced by LIDAR, machine learning algorithms, and language models trained for the automated generation of clinical reports. 0029. When available, the technology leverages information from thermographic sensors and LIDAR (Light Detection and Ranging) sensors to provide more detailed clinical information. 0030. Increased accessibility to screening methods reduces the need for specialized equipment and enables the use of technology in regions with limited infrastructure. 0031. After the advantages presented, and so that the process of the invention proposed in this patent document can be better understood and evaluated, the description of the drawings will be given below. DESCRIPTION OF THE FIGURES 0032. The structure and operation of the present invention, together with its additional advantages, can be better understood by reference to the accompanying drawings and description, which are for illustrative purposes only and are not limiting to the subject matter of the present invention. 0033. The invention will now be explained in detail, and the following drawings are presented by way of illustration: Petition 870250062739, dated 07 / 21 / 2025, p. 15 / 32 9 / 24 Figure 1. presents the system flowchart which is composed of: (1) Main Image Evaluation Environment: Visible Light Image Capture Module (1.1), Visible Light Image Processing Module (1.2); (2) Clinical and Textual Data Environment: Clinical Data Processing Module (2.1); (3) Accessory Image Evaluation Environment: Thermographic Image Capture Module (3.1), Thermographic Image Processing Module (3.2), Three-Dimensional Point Cloud Capture Module (3.3) and Three-Dimensional Point Cloud Processing Module (3.4); (4) Data Integration Environment: Data Normalization and Standardization Module, (4.1) Multimodal Unification and Representation Module (4.2) and Temporal and Evolutionary Correlation Module (4.5) and (5) Report Generation Environment: Semantic Modeling Module (5.1), Structured Report Generation Module (5.2) and Storage and Sharing Module (5.3). Figure 2 presents the flowchart for the system's operation, showing steps I, II, III, IV, V, VI, and VII. Figure 3 illustrates the different perspectives of the foot for capturing images: medial (1), lateral (2), inferior (3), superior (4), anterior (5) and posterior (6). Figure 4 shows an example of an automatically generated clinical report, including risk classification and recommendations. SUMMARY OF THE INVENTION 0034. The main object of this patent application is to describe in detail the process and product related to the 'SYSTEM FOR SCREENING AND IDENTIFYING THE RISK OF COMPLICATIONS IN THE LOWER LIMBS OF DIABETIC PATIENTS USING MACHINE LEARNING'. Thus, more clearly, the invention presents a process that integrates multiple image analysis approaches for evaluating the Petition 870250062739, dated 07 / 21 / 2025, page 16 / 32 10 / 24 risk of lower limb complications in people with diabetes. Unlike existing solutions that are limited to isolated plantar pressure analysis, thermography, or manual clinical assessment, this invention proposes an automated machine learning-based process that combines images captured in the visible light spectrum, which can be complemented by thermal images in the infrared spectrum and three-dimensional point clouds obtained by LIDAR. The analysis of these different data sources, when combined with the patient's clinical history and trained language models (LLM) for automated report generation, results in a comprehensive assessment tool capable of detecting risk factors even before the development of ulcers and other complications. 0035. This invention reflects the unique evaluation through visible spectrum imaging, as well as the unprecedented possibility of integrating different imaging modalities as available (e.g., thermographic images and three-dimensional point clouds produced by LIDAR) and the system's ability to perform structural and dermatological analyses from multiple perspectives (anterior, posterior, superior, inferior, medial, and lateral). This approach allows for the automated identification of musculoskeletal deformities and dermatological changes, dynamically correlating them with the patient's clinical data. Furthermore, the system enables the capture of individual images, in groups, or in video sequences, allowing for a dynamic analysis of the foot structure and a better interpretation of the patient's clinical evolution over time. 0036. Another aspect of the present invention is the ability to operate with widely available image capture devices, such as mobile device cameras, eliminating the need for proprietary hardware and making the solution more accessible and scalable for different clinical contexts, including remote care and home monitoring. The incorporation of machine learning for automatic thermal image segmentation, structural pattern recognition, and report generation is also a key advantage. Petition 870250062739, dated 07 / 21 / 2025, page 17 / 32 11 / 24 clinical findings differentiate the invention from conventional approaches, which often require manual interpretation of the findings. 0037. Additionally, the invention also utilizes trained language models (LLMs) to transform raw image analysis data and clinical information into automated clinical reports that are readable and understandable by healthcare professionals and patients. Based on structural and dermatological findings obtained through machine learning algorithms, the system organizes and summarizes this information in a structured format, presenting the identified risks, potential clinical impacts, and personalized recommendations. The LLM interprets the patterns detected by the analysis model and contextualizes them within best medical practices, making the report clear, accessible, and informative, without altering its content.Furthermore, the reports generated by the invention are adaptable to the user's level of knowledge, offering detailed technical descriptions for healthcare professionals and simplified explanations for patients, ensuring that the information is useful for different user profiles. This automated approach improves the communication of clinical findings, facilitates the monitoring of the patient's condition, and optimizes medical decision-making, enabling early and precise interventions in the prevention of diabetic foot complications. 0038. Finally, the present invention not only improves the accuracy and efficiency in detecting risks associated with diabetic foot, but also enables a predictive assessment model, allowing medical interventions before the progression of serious complications. By integrating image analysis with clinical history through machine learning for the automation of diagnoses and clinical recommendations, this invention stands out as a novel solution in the field of prevention and monitoring of lower limb complications associated with diabetes. 0039. Within the presented context, and so that the developed process and obtained product, objects of protection of this patent application, may be Petition 870250062739, dated 07 / 21 / 2025, page 18 / 32 12 / 24 understood and evaluated in a clearer and more objective way, its detailed description will be given below. DETAILED DESCRIPTION OF THE INVENTION 0040. This section will reveal all the details of the development process of the aforementioned "SYSTEM FOR SCREENING AND IDENTIFYING THE RISK OF COMPLICATIONS IN THE LOWER LIMBS OF DIABETIC PATIENTS USING MACHINE LEARNING," whose purpose is to describe sufficiently and clearly all the steps involved in the system for evaluating lower limbs in diabetic patients using machine learning techniques for image analysis, and thus, to substantiate the descriptive sufficiency of this invention. 0041. This assessment can also be complemented by data obtained through LIDAR by including three-dimensional point cloud, depth matrix, and reconstructed 3D image for the automated detection of risk factors associated with diabetic foot. The assessment can also be complemented by including a heat map image, captured from different perspectives, for the automated detection of risk factors associated with diabetic foot. 0042. The invention allows for the early identification of structural deformities, dermatological changes, and other risk factors for lower limb complications associated with diabetes, including amputations. 0043. The system is composed of environments and their modules: (1) Main Image Evaluation Environment: Visible Light Image Capture Module (1.1), Visible Light Image Processing Module (1.2); (2) Clinical and Textual Data Environment: Clinical Data Processing Module (2.1); (3) Accessory Image Evaluation Environment: Thermographic Image Capture Module (3.1), Thermographic Image Processing Module (3.2), Three-Dimensional Point Cloud Capture Module (3.3) and Three-Dimensional Point Cloud Processing Module (3.4); (4) Data Integration Environment: Module of Petition 870250062739, dated 07 / 21 / 2025, page 19 / 32 13 / 24 Data Normalization and Standardization (4.1), Unification and Multimodal Representation Module (4.2) and Temporal and Evolutionary Correlation Module (4.5) and (5) Report Generation Environment: Semantic Modeling Module (5.1), Structured Report Generation Module (5.2) and Storage and Sharing Module (5.3). Primary Image Capture Environment 0044. The Main Image Capture Environment (1) is responsible for obtaining and transferring images of the patient's lower limb to the processing environment. This environment is essential for the automated assessment of the risk of complications in the lower limbs of patients with diabetes, ensuring that the acquired images are suitable for subsequent analysis by machine learning techniques and is composed of the Visible Light Image Capture Module (1.1) and the Visible Light Image Processing Module (1.2). 0045. The Visible Light Image Capture Module (1.1) acquires images of the patient's foot using mobile devices such as smartphones, tablets, or specialized cameras operating in the visible light spectrum (RGB). Capture can be performed from different perspectives, including at least one anterior, posterior, medial, lateral, superior, and inferior view, allowing for a comprehensive analysis of the foot structure. Images can be obtained individually (static frame), in groups (set of images to enhance the evaluation), or sequentially (video), enabling dynamic analysis of the foot structure in different positions. After capture, the images are immediately transferred to the processing environment, where they undergo pre-processing steps to optimize quality and extract features relevant for clinical evaluation. 0046. In the Visible Light Image Processing Module (1.2), the captured images are prepared for detailed analysis, ensuring that they are in a format optimized for the identification of clinical patterns and Petition 870250062739, dated 07 / 21 / 2025, page 20 / 32 14 / 24 structural. This module performs three main steps: pre-processing, which includes adjusting image quality, removing artifacts, and normalizing lighting and contrast; tokenization, which converts images into mathematical representations for processing by machine learning algorithms; and feature extraction, which identifies relevant structural and dermatological patterns, such as musculoskeletal deformities, pre-ulcerative lesions, and skin alterations. This processing integrates the images with the patient's clinical data and combines them with other information sources, such as thermal images and three-dimensional models, when available, ensuring a more comprehensive and accurate analysis. 0047. The evaluation process for the suitability of footwear for people with diabetes is carried out through the analysis of single or sequential visible light spectrum images, individually or in groups, captured from at least one of the following perspectives: front, back, bottom, medial, side, and top of the footwear, integrated or not with partial or complete medical history data of the patient or group of patients, and its subsequent analysis by machine learning algorithms to detect suitability for use by people with diabetes. 0048. The evaluation of previously acquired images of footwear allows for a detailed analysis of the relationship between the plantar support offered by the footwear and the morphology of the patient's foot. The images are processed to identify characteristics such as shape, arch height, distribution of pressure points, and sole wear, allowing the identification of patterns that may indicate improper use, biomechanical misalignment, or negative impact on load distribution. The correlation of these images with the structural findings of the foot enables the automated detection of incompatibilities, such as excessively tight shoes that compress vulnerable joints or rigid soles that reduce impact absorption. Clinical and Textual Data Environment Petition 870250062739, dated 07 / 21 / 2025, page 21 / 32 15 / 24 0049. The Clinical and Textual Data Environment (2) is responsible for acquiring, structuring, and processing clinical information associated with the patient, ensuring that the data is compatible with the other modules of the system and can be correlated with visual findings. This environment plays an essential role in contextualizing risk assessment, allowing the model to consider the patient's clinical history when interpreting captured images. 0050. The Clinical and Textual Data Environment consists of a Clinical Data Processing Module (2.1) that organizes and prepares the patient's clinical information for integration into the automated assessment process, operating in three main stages. First, upon entering structured clinical data, the system receives relevant patient information, entered manually or automatically through integration with electronic medical records and medical databases, including history of ulcers, neuropathies, vascular diseases, blood glucose, blood pressure, medication use, and other risk factors associated with lower limb complications. Next, data processing and cleansing are performed, where normalization and validation techniques are applied to ensure the integrity and standardization of information, correcting inconsistencies, eliminating redundancies, and structuring the data for analysis.Finally, in the tokenization and feature extraction stage, clinical data are converted into a computational format suitable for interpretation by machine learning algorithms, allowing the extraction of essential clinical parameters, such as ulcer recurrence patterns, duration of the condition, and glycemic variability, favoring correlation with visual findings and improving the accuracy of the automated assessment. Image Evaluation Accessory Environment 0051. The Image Assessment Accessory Environment (3) complements the visual analysis of the lower limb, allowing the incorporation of thermographic and three-dimensional (LIDAR) images for a more detailed assessment. Petition 870250062739, dated 07 / 21 / 2025, page 22 / 32 16 / 24 This environment enhances diagnostic accuracy by identifying anomalous thermal and structural patterns associated with complications in diabetic patients. This environment consists of two capture modules (3.1) and (3.3) and their respective processing modules (3.2) and (3.4), responsible for extracting and organizing relevant information. 0052. The Thermographic Image Capture Module (3.1) receives images in the infrared spectrum and analyzes the thermal distribution of the skin to detect abnormal patterns associated with complications in the lower limbs. This approach identifies peripheral neuropathy, evidenced by areas of reduced blood circulation, inflammation and infection, characterized by localized temperature increases, and ischemia, indicating impaired blood supply in specific regions of the foot. The captured images are transferred to the processing environment, where they are analyzed in conjunction with other imaging modalities for a more precise assessment of the patient's risk factors. The Thermographic Image Processing Module (3.2) processes and analyzes the captured thermal data, ensuring its suitability for automated interpretation.The process begins with preprocessing, which involves contrast adjustment, noise removal, and thermal normalization to ensure accuracy in detecting temperature variations. Next comes tokenization, where thermal data are converted into representations compatible with machine learning algorithms, allowing integration with other information sources. Finally, the feature extraction stage identifies temperature gradients and thermal distribution patterns, correlating these findings with clinical risk factors such as neuropathy, inflammation, ischemia, and infection, enabling an in-depth analysis of the risk of lower limb complications. 0053. The Three-Dimensional Point Cloud Capture Module (3.3) uses LIDAR (Light Detection and Ranging) technology to generate a detailed three-dimensional model of the patient's foot, allowing for structural analysis. Petition 870250062739, dated 07 / 21 / 2025, page 23 / 32 Advanced 17 / 24 technology identifies musculoskeletal deformities, such as bony prominences, exemplified by exposed metatarsal heads, Charcot arthropathy, recognized by changes in bone contour, and alterations in the curvature of the plantar arch, such as flat feet or high arches. Three-dimensional capture provides an accurate representation of lower limb morphology, allowing for the automated detection of structural anomalies. The information obtained is then transferred to the processing module, where it is analyzed in conjunction with other image sources and clinical data to improve the assessment of the risk of lower limb complications. 0054. The Three-Dimensional Point Cloud Processing Module (3.4) processes and analyzes data captured via LIDAR technology, ensuring the accuracy of the structural modeling of the patient's foot. The process begins with pre-processing, which involves filtering and removing spurious points to ensure the accuracy of the generated three-dimensional model. Next, tokenization occurs, where the point cloud is converted into a computational structure compatible with machine learning algorithms, allowing its integration with other data modalities. Finally, the feature extraction stage identifies structural deviations and anatomical irregularities, correlating this information with risk factors for ulcers and amputations, making possible a detailed analysis of foot morphology and the automated detection of anomalies relevant to clinical evaluation. Data Integration Environment 0055. The Data Integration Environment (4) consolidates and structures the information captured in the previous modules, ensuring that different sources of visual and clinical data are unified and prepared for advanced analysis. This environment plays a critical role in the normalization, merging, and temporal correlation of data, allowing for a more accurate and personalized assessment of the risks of lower limb complications. Petition 870250062739, dated 07 / 21 / 2025, page 24 / 32 18 / 24 people with diabetes. This environment is composed of three main modules: Data Normalization and Standardization Module (4.1), Unification and Multimodal Representation Module (4.2), and Temporal and Evolutionary Correlation Module (4.3). 0056. The Data Normalization and Standardization Module (4.1) ensures that all images and clinical information are adjusted to a uniform format, guaranteeing their correct integration and analysis within the system workflow. This module corrects inconsistencies and harmonizes multiple input sources, allowing the fusion of data from different modalities. The process begins with the joint adjustment of image scales and formats, unifying the dimensions and parameters of different image types, such as RGB, thermal, and LIDAR, to ensure a standardized analysis model. Next, images are aligned with clinical information, directly associating the patient's structured data with the captured images, ensuring consistency in medical interpretation. Finally, the module corrects temporal inconsistencies and capture variations, eliminating differences in lighting, perspective, and image quality obtained over time.After this normalization, the data proceeds to the Unification and Multimodal Representation Module (4.2), where it is processed together with other information for in-depth analysis. 0057. The Multimodal Unification and Representation Module (4.2) is responsible for merging multiple data sources into a single integrated representation, optimizing the interpretation of clinical and structural findings. This process begins with the merging of multiple data sources, combining visible light, thermographic, and three-dimensional (LIDAR) images with the patient's clinical data, allowing for a more comprehensive analysis. Next, visual feature and embedding unification occurs, where features extracted from images and clinical texts are transformed into common mathematical representations, facilitating interpretation by machine learning models. Furthermore, the module applies Petition 870250062739, dated 07 / 21 / 2025, page 25 / 32 19 / 24 mapping of correlations by attention mechanisms, using neural networks to identify patterns and relationships between clinical and visual data, highlighting critical regions associated with the risk of complications. This approach allows information from different modalities to be analyzed in an integrated way, resulting in a more robust and accurate assessment of the patient's condition. 0058. The Temporal and Evolutionary Correlation Module (4.3) enables monitoring of the patient's clinical progression by comparing current information with previous records, allowing the identification of subtle changes that may indicate progression to more serious conditions. This module compares current findings with previous ones, longitudinally analyzing images and clinical data to detect the progression or regression of lesions, inflammation, and structural deformities. Furthermore, it applies predictive recurrence models, using neural networks and machine learning to estimate the risk of ulcer worsening or the emergence of new complications based on the patient's historical patterns. The module structures the data and vectorizes the information, organizing it into models compatible with automated processing, facilitating interpretation and visualization by healthcare professionals.In this way, this module ensures that assessments are not isolated, but rather contextualized within the patient's clinical journey, allowing for more informed decision-making and early interventions. Report Generation Environment 0059. The Reporting Environment (5) transforms processed and integrated data into understandable and organized clinical information, facilitating the interpretation of findings and assisting in medical decision-making. This environment processes the results extracted from the previous modules, correlating clinical and structural findings to provide automated, personalized reports adaptable to the end-user profile. This environment is composed of three main modules: Modeling Module Petition 870250062739, dated 07 / 21 / 2025, page 26 / 32 20 / 24 Semantics (5.1), Structured Report Generation Module (5.2) and Storage and Sharing Module (5.3).0060. The Semantic Modeling Module (5.1) analyzes previously integrated and processed clinical and visual data to extract relevant patterns, correlate risks, and categorize the severity of detected complications. This module suggests patterns, identifying trends in clinical and structural findings, allowing for predictive analysis and supporting medical decision-making. Furthermore, it performs risk correlation, associating clinical and morphological findings with known risk factors such as neuropathy, ischemia, or signs of infection, establishing essential connections for patient diagnosis and prognosis. Finally, it performs severity categorization, automatically classifying findings into different risk levels, allowing for the prioritization of critical cases and enabling an appropriate action plan.This module ensures that data interpretation is contextualized and structured, optimizing the generation of detailed reports and facilitating medical decision-making. 0061. The Structured Report Generation Module (5.2) organizes processed information into readable documents formatted for different types of users, ranging from healthcare professionals to patients themselves. Trained Language Models (LLMs) are used to transform raw image analysis data and clinical information into automated clinical reports. This module organizes data into a readable format, structuring information into clear sections that highlight clinical findings and relevant recommendations. Furthermore, it implements language customization according to the user profile, adjusting the tone and complexity level of the report to meet the needs of different audiences, such as medical specialists, nurses, or patients, ensuring that the information is understood efficiently. In this way, this module ensures that clinical findings are presented... Petition 870250062739, dated 07 / 21 / 2025, page 27 / 32 21 / 24 in an intuitive way, reducing the need for manual interpretation and increasing efficiency in diagnostic and therapeutic decision-making. 0062. The Storage and Sharing Module (5.3) manages the secure storage and distribution of clinical reports, ensuring fast and reliable access for authorized users. This module stores reports, securely recording generated documents in medical databases or electronic health record systems, guaranteeing the integrity and traceability of information. Furthermore, it enables the export of reports, making documents available in compatible formats for email delivery, facilitating access to reports for healthcare professionals and patients, ensuring continuity of clinical monitoring and informed decision-making. 0063. The process for the system to function, as shown in Figure 2, comprises the following steps: I. Utilize mobile devices, such as smartphones and specialized cameras, to capture images of the patient's foot, encompassing image types such as visible light (RGB), associated or not with thermographic (infrared) and / or three-dimensional (LIDAR) images, capturing from different perspectives, such as anterior, posterior, medial, lateral, superior and inferior, and in various formats, including static images (single frame), in groups (set of images) or sequential (videos), in order to enable a comprehensive analysis of the foot structure. II. Transfer the captured images to the processing environment (2), ensuring that they are efficiently sent to the subsequent analysis and processing steps. III. Perform image quality adjustments, remove artifacts, and normalize lighting and contrast, preparing the images for detailed and accurate analysis in subsequent steps. Petition 870250062739, dated 07 / 21 / 2025, page 28 / 32 22 / 24 IV. Perform tokenization, converting images into mathematical representations for processing by machine learning algorithms, and feature extraction, identifying relevant structural and dermatological patterns, such as deformities and lesions, to facilitate clinical analysis and evaluation. V. Collect the patient's clinical data, such as medical history, blood glucose levels, and other data, and process it to ensure its integrity, standardization, and proper preparation for integration with the captured images. VI. Fuse the images (RGB, thermal, 3D) with the patient's clinical data and perform a temporal evolution analysis, comparing current data with the patient's history to identify relevant changes and patterns over time. VII. Organize and categorize the clinical and structural findings, including clinical recommendations based on the severity of the complications detected, to provide a clear and targeted overview for the treatment and follow-up of the patient. VIII. Store the generated reports in medical databases or electronic medical record systems, enabling the export of reports and making the documents available in compatible formats for sending via email. Implementation of the Invention 0064. To validate the practical applicability and effectiveness of the present invention, a clinical usability and effectiveness test was conducted involving 6 (six) nurses trained in diabetic foot assessment and 60 (sixty) adult diabetic patients, in which the developed system was compared to the clinical assessment of a nursing professional specialized in the diagnosis of lower limb complications in diabetic patients. This study aimed to demonstrate the system's ability to capture, process, and analyze multimodal images automatically, as well as Petition 870250062739, dated 07 / 21 / 2025, page 29 / 32 23 / 24 its integration with clinical information for the generation of automated reports. 0065. During the study, each patient underwent a standardized clinical assessment, in which images of the lower limb were captured from different perspectives using the technologies available in the system. The images were analyzed by AI, which processed the data applying machine learning techniques to identify structural, dermatological, and thermal patterns associated with risk factors. Simultaneously, the nursing professional performed the conventional assessment, using traditional clinical methods such as visual inspection, palpation, and analysis of the patient's medical history. 0066. The results demonstrated an overall sensitivity rate of 82.8%, an overall accuracy rate of 81.7%, and an overall specificity rate of 80.7% when comparing the assessment performed by artificial intelligence with the assessment by the nursing professional, validating the system's effectiveness for screening purposes in identifying musculoskeletal deformities, dermatological changes, and abnormal thermal patterns. The AI ​​was able to detect bony prominences, ulcers in early stages, calluses, fissures, and signs of infection. Furthermore, the assessment included thermal gradients indicative of active infection and the detection of Charcot foot through three-dimensional point clouds using LIDAR.Furthermore, testing was conducted with multiple consecutive analyses to validate the system's ability to compare findings with previous records, demonstrating its functionality in observing patients' clinical evolution and providing predictions about the risk of worsening complications. 0067. The automated generation of clinical reports by the system was also validated, ensuring that the findings were organized in a structured and understandable manner. The generated reports presented language adapted to the user profile, allowing doctors, nurses, and patients to interpret the information intuitively and practically. A Petition 870250062739, dated 07 / 21 / 2025, pages 30 / 32 The 24 / 24 integration of clinical data with imaging findings has enabled a more contextualized analysis, improving the accuracy of diagnosis and the planning of therapeutic interventions. 0068. Based on these findings, it is concluded that the invention complements conventional approaches to diabetic foot screening and monitoring by offering an automated system capable of performing assessments quickly, accurately, and affordably. The technology can be applied in various similar clinical settings, including medical offices, primary care units, hospitals, and telemedicine services, enabling continuous monitoring and reducing the risk of serious complications such as advanced infections and amputations.

Claims

1. SYSTEM FOR SCREENING AND IDENTIFYING THE RISK OF COMPLICATIONS IN THE LOWER LIMBS OF DIABETIC PATIENTS USING MACHINE LEARNING, characterized by comprising environments and modules: (1) Main Image Evaluation Environment: Visible Light Image Capture Module (1.1), Visible Light Image Processing Module (1.2); (2) Clinical and Textual Data Environment: Clinical Data Processing Module (2.1); (3) Accessory Image Evaluation Environment: Thermographic Image Capture Module (3.1), Thermographic Image Processing Module (3.2), Three-Dimensional Point Cloud Capture Module (3.3) and Three-Dimensional Point Cloud Processing Module (3.4); (4) Data Integration Environment: Data Normalization and Standardization Module (4.1), Unification and Multimodal Representation Module (4.2) and Temporal and Evolutionary Correlation Module (4.5) and (5) Report Generation Environment: Semantic Modeling Module (5.1), Structured Report Generation Module (5.2) and Storage and Sharing Module (5.3).

2. SYSTEM FOR TRIAGE AND RISK IDENTIFICATION, according to claim 1, characterized by its operation comprising the following steps: I. Using mobile devices, such as smartphones and specialized cameras, to capture images of the patient's foot, encompassing image types such as visible light (RGB) and / or thermal (infrared) and / or three-dimensional (LIDAR); II. Collecting the patient's clinical data, such as medical history, blood glucose, among others, and processing them to ensure their integrity, standardization, and adequate preparation for integration with the captured images and transferring the captured images to the processing environment, ensuring that they are efficiently sent to the subsequent analysis and treatment steps; III.Perform tokenization, converting images into mathematical representations for processing by machine learning algorithms and feature extraction, identifying relevant structural and dermatological patterns, such as deformities and lesions, to facilitate clinical analysis and evaluation; IV. Perform image quality adjustment, remove artifacts, and normalize lighting and contrast, preparing the images for detailed and accurate analysis in subsequent steps; V. Fuse the images (RGB, thermal, 3D) with the patient's clinical data and perform temporal evolution analysis, comparing current data with the patient's history to identify relevant changes and patterns over time; VI. Organize and categorize clinical and structural findings, including clinical recommendations based on the severity of detected complications, to provide a clear and targeted view for patient treatment and follow-up; VII.Store the generated reports in medical databases or electronic medical record systems, enabling the export of reports and making the documents available in formats compatible for sending via email.

3. SYSTEM FOR TRIAGE AND RISK IDENTIFICATION, according to claim 1, characterized by capturing the image of the patient's foot from a mobile image capture device in visible light spectrum (RGB) (1.1) and / or thermal (infrared) (3.1) and / or three-dimensional (LIDAR) (3.3), single or sequential, Petition 870250062739, dated 07 / 21 / 2025, page 5 / 32 3 / 4 individually or in groups, captured from at least one of the anterior, posterior, inferior, medial, lateral and superior perspectives, integrated or not with textual data, then removing artifacts and making adjustments to brightness and contrast.

4. SYSTEM FOR SCREENING AND RISK IDENTIFICATION, according to claims 1 and 2, characterized by performing structural, thermal and dermatological analysis using machine learning algorithms to detect musculoskeletal deformities and dermatological changes, in order to develop a risk profile.

5. SYSTEM FOR TRIAGE AND RISK IDENTIFICATION, according to claim 1, characterized by a three-dimensional point cloud capture module (3.3), generating a detailed three-dimensional model of the patient's foot, using LIDAR technology, to identify musculoskeletal deformities, such as Charcot arthropathy, plantar arch deformities and other structural irregularities; 6. SYSTEM FOR TRIAGE AND RISK IDENTIFICATION, according to claims 1 and 2, characterized by capturing images of the infrared spectrum and subsequent analysis by machine learning algorithms to identify abnormal temperature patterns related to neuropathy, inflammation, infection, ischemia, or other conditions relevant to the risk profile for the development of ulcers or amputation of lower limbs.

7. SYSTEM FOR TRIAGE AND RISK IDENTIFICATION, according to claim 1, characterized by generating automated clinical reports for diabetic foot risk assessment, integrated or not with the use of trained language models (LLM - Large Language Models) that compare visual findings with the patient's clinical data, generating a risk classification where the risk of complications is classified into different levels, such as low, medium or high. Petition 870250062739, dated 07 / 21 / 2025, page 6 / 32 4 / 4 8. SYSTEM FOR SCREENING AND RISK IDENTIFICATION, according to claim 1, characterized by evaluating images of footwear for people with diabetes through the analysis of single or sequential visible light spectrum images, individually or in groups, captured from at least one of the anterior, posterior, inferior, medial, lateral and superior perspectives of the footwear, integrated or not with partial or complete medical history data of the patient or group of patients, detailing the relationship between the plantar support of the footwear and the morphology of the patient's foot, including the identification of foot shape, arch height, distribution of pressure points and sole wear, allowing the identification of patterns that may indicate inadequate use of biomechanical alignment or negative impact on load distribution.

Images