Data-based personalized site-specific care advising model powered with artificial-intelligence

Abstract

Computer-implemented systems (CIS), computer-implemented methods (CIM), and / or machine learning algorithms (MLA) generate health recommendations for specific parts of local tissue structures, such as oral hygiene, based on initial images generated by mobile phones (e.g., smartphones) or other handheld devices. CIS, CIM, or MLA includes a generative AI platform operablely connected to a graphical user interface (GUI), and also includes a discriminative AI platform. The discriminative AI platform generates images of local tissue structures, including the health status of specific parts of the tissue structure, and the generative AI platform generates health recommendations for those specific parts of the tissue structure based on the images generated by the discriminative AI platform and transmits them to the GUI. Publicly available CIM, CIS, or MLA provide new platforms that leverage AI to provide automated and real-time personalized oral hygiene care.

Description

Technical Field

[0001] This invention generally relates to healthcare, and in particular to a computer-implemented system / method for analyzing data (image-based and / or text-based) on one or more tissues (e.g., from the oral cavity, such as gums), which can provide care recommendations (e.g., oral hygiene) based on the health status of specific sites of one or more tissues. Prior Technology

[0002] Gum disease is a global epidemic, affecting up to 50% of the world's population. It affects the supporting tissues of teeth, leading to alveolar fold damage, tooth loss, and difficulties eating, speaking, appearance, and many systemic diseases. The primary cause of gum disease is the accumulation of plaque along the gum line, leading to localized inflammation. Inflammatory gum disease can be prevented through self-management practices such as brushing and interdental cleaning. However, this inflammation is a localized reaction to plaque, which can only be resolved through proper physical cleaning of the affected areas. Therefore, inflammatory gum disease or conditions, such as bleeding gums, still occur extensively in adults, despite most of these adults brushing twice daily and using additional hygiene measures such as mouthwash and toothpicks. A recent survey in a specific region showed that almost all older adults brush their teeth at least once a day. However, over 90% of them had gingivitis, and over 45% had poor oral hygiene.

[0003] Inflammatory gingival disease is inherently chronic; therefore, most individuals in the population are unaware of its presence until it reaches a late or advanced stage where most preventative and / or simple treatments are ineffective. Due to time constraints and concerns about the cost of professional treatment, a significant number of people worldwide do not have regular dental checkups. This is particularly true among the elderly, whose access to dental care is further limited by increased economic and / or physical constraints. The prevalence of inflammatory gingival disease is higher and can lead to serious consequences, even tooth loss, when disease progression is subtle, oral hygiene habits are inadequate, and access to dental care is limited. The associated inflammatory mediators also affect overall health and are linked to endocrine disorders such as diabetes and cardiovascular diseases such as stroke and hypertension. To address these issues, a treatment solution that can prevent and maintain inflammatory gingival disease, and one that provides convenient, handy, and / or at-home personalized oral hygiene advice, is needed.

[0004] Therefore, one object of the present invention is to provide a computer-implemented system and / or method that can analyze data (e.g., based on images, based on text, or a combination thereof) and make health-based recommendations (e.g., oral hygiene care).

[0005] Another object of the present invention is to provide a computer-implemented system and / or method that can analyze data (e.g., based on images, based on text, or a combination thereof) and make personalized health-based recommendations (e.g., personal oral hygiene care).

[0006] Another object of the present invention is to provide a computer-implemented system and / or method comprising a generative artificial intelligence platform, which can analyze data (e.g., based on images, based on text, or a combination thereof) and make health-based recommendations (e.g., oral hygiene care).

[0007] Another object of the present invention is to provide a computer-implemented system and / or method comprising a generative artificial intelligence platform, which can analyze data (e.g., based on images, based on text, or a combination thereof) and make personalized health-based recommendations (e.g., personal oral hygiene care). Summary of the Invention

[0008] This document provides a computer-implemented system (CIS) and / or a computer-implemented method (CIM), not limited to any particular hardware or operating system. The CIS or CIM generates health care recommendations for specific sites within a local tissue structure. In a preferred form, the CIS and / or CIM generates health care recommendations for specific sites within a tissue structure, such as providing oral care recommendations based on images of the oral cavity taken directly using a mobile phone (such as a smartphone) or other handheld device.

[0009] In a preferred embodiment, the CIS includes a generative artificial intelligence (AI) platform operablely linked to a graphical user interface (GUI). Based on dental professional advice (e.g., a dentist's recommendation), the developed generative AI platform can generate health-based recommendations for specific parts of the tissue structure by analyzing intraoral images. These recommendations are preferably personalized and are transmitted to the individual user requiring the advice via the GUI. Preferably, the intraoral images include periodontal health results for specific parts of the gums, such as (i) healthy, (ii) diseased, and (iii) suspicious. The generative AI platform includes generative machine learning algorithms that, after training and / or testing, can generate health care recommendations for specific parts of the tissue structure, such as oral hygiene care. Recommendations for specific parts of the tissue structure may include brushing techniques or flossing recommendations, such as brushing gently with a soft-bristled toothbrush (e.g., when a defect is located at the gum line), interdental brushing (e.g., when a defect is located at the interdental gingiva), or a combination thereof.

[0010] Preferably, the CIS also includes a discriminative AI platform operatively connected to a generative AI platform. Preferably, the discriminative AI platform is operatively connected to a user device terminal and collects and / or transmits initial data to be analyzed by the CIS from the user device terminal. The discriminative AI platform may be located on the same or a different server as the server containing the generative AI platform. In some forms, the discriminative AI platform resides on a user device terminal (e.g., a mobile phone, such as a smartphone) that also contains a generative AI platform. The discriminative AI platform preferably performs periodontal anatomical disease detection at specific sites based on digitally pixelated photographs. Upon completion, the discriminative AI platform generates data based on intraoral images and transmits it to the generative AI platform. Based on this information and the location of the lesion, the revealed generative AI platform can provide personalized recommendations for specific sites, informing how to maintain gingival cleanliness and / or health.

[0011] A computer-implemented method (CIM) is also described herein, which uses the disclosed CIS to generate oral health care recommendations for specific sites of tissue structure. Preferably, CIM involves (i) analyzing data based on intraoral images; and (ii) displaying oral hygiene care recommendations for specific sites of tissue structure to a graphical user interface (GUI) based on the analysis in step (i). Prior to step (i), the intraoral images in step CIS generate intraoral image-based data by analyzing digitally pixelated photographs. Preferably, the intraoral image-based data is generated using the aforementioned discriminative AI platform.

[0012] The use of the disclosed CIS or CIM is also described. CIS or CIM has broad applicability and is not limited to patient data from a specific region. Preferably, CIS or CIM is used to provide information on an individual's oral health status and, when necessary, to offer suggestions for improving that health status. Users can use mobile phones (such as smartphones) and instant messaging applications (such as WhatsApp or WeChat) to take photos of their anterior teeth and upload them to the CIS hosted on a server. In some cases, if the mobile phone (such as a smartphone) has the computing power to perform CIS and / or CIM analysis, a server is not required. Users can also install mobile applications or upload photos to the CIS via a website. CIS or CIM includes a photo quality check function, prompting the user to retake the photo if necessary. Based on the photographs taken, CIS or CIM can identify the gingiva and label specific areas as (i) healthy, (ii) diseased, or (iii) suspicious. Depending on the condition and location of the area (e.g., gingiva), CIS or CIM can provide suggestions for improving the health status of one or more gingival regions. Preferably, the recommendations are based on images (such as photographs), are personalized to individual needs, and / or are supported by a generative AI platform.

[0013] The disclosed CIM or CIS provides a new platform that utilizes AI to automatically and instantly provide personalized, localized health care recommendations (such as oral care, skin care, mucosal care, etc.) for the lesion site based on one or more photos taken with a mobile phone (such as a smartphone). Simple Explanation of the Diagram

[0014] Figure 1 illustrates the proposed AI model for personalized, site-specific oral hygiene care recommendations, along with a flowchart of a non-limiting paradigm for how it works.

[0015] Figure 2 illustrates a system architecture diagram of a non-limiting example. The diagram depicts a flowchart and shows how a non-limiting, personalized, site-specific oral hygiene advice AI model operates.

[0016] Figure 3 illustrates a non-limiting example of oral hygiene recommendations for a specific site marked with a label, which is marked by the intermediate layer between the teeth and the rest of the gums. Implementation

[0017] [I. Definition]

[0018] "Discriminative artificial intelligence" or "discriminative AI" refers to a set of machine learning techniques / algorithms that typically make predictions or inferences based on the analysis of input data. Discriminative artificial intelligence platforms are also known as predictive artificial intelligence.

[0019] "Generative artificial intelligence" or "generative AI" refers to a set of machine learning techniques / algorithms that can perform creative tasks similar to those of humans with the help of machine learning models.

[0020] "Intraoral cavity" refers to the tissue structure inside the oral cavity or the tissues inside the oral cavity (such as gums and teeth).

[0021] "Operationally connected" means connecting at least two components in a CIS through technologies including, but not limited to, Ethernet, Bluetooth, near field communication, WiFi, integrated circuits, or combinations thereof.

[0022] [II. Computer-based systems and methods]

[0023] [i.] Computer Execution System

[0024] This document describes a computer-implemented system (CIS) not limited to any particular hardware or operating system, which is provided for processing and / or analyzing imaging and / or non-imaging input data. Specifically, the CIS is capable of generating health-based recommendations, preferably personalized and health-based, by analyzing data from local tissue structures of a subject. In a more preferred form, the health-based recommendations are site-specific in terms of tissue structure, including guidance on how to care for sub-sections of the relevant tissue structure. The data used can preferably be image-based, text-based, or a combination thereof. In a more preferred form, image-based data is used.

[0025] CIS includes a generative artificial intelligence (AI) platform that is operablely linked to a graphical user interface (GUI). The generative AI platform generates health-based recommendations by analyzing tissue structure data from the subject's main topic; preferably, the recommendations are personalized and health-based. The data used can preferably be image-based, text-based, or a combination thereof. In a more preferred form, image-based data is used. The tissue structure data can be from the oral cavity, mucosa, or epidermis, preferably from the oral cavity. The generative machine learning algorithm aims to produce output recommendations similar to those created by humans, such as illustrative recommendations on how to treat local tissue structures. The algorithm can analyze and learn from existing data and then generate recommendations based on the analysis. In some forms, the generative machine learning algorithm evaluates local tissue structure data (e.g., data based on intraoral images, intraoral text, or a combination thereof; data based on epidermal images, epidermal text, or a combination thereof; data based on mucosa images, mucosa text, or a combination thereof) to generate health care recommendations. Preferably, the local tissue structure data contains specific information about the health status of a specific site on the tissue structure of interest. In a preferred form, the data on local tissue structures is based on intraoral image data, text-based data, or a combination thereof, and includes periodontal health results. In a more preferred form, the image-based data, text-based data, or a combination thereof includes periodontal disease information for specific sites. In some forms, the periodontal disease information for specific sites relates to marginal gingiva, attached gingiva, interdental gingiva, oral mucosa, alveolar mucosa, or a combination thereof.

[0026] Currently, the work of CIS in healthcare settings typically involves detecting the presence or absence of disease / symptoms, or various stages of disease / symptoms. These traditional methods involve healthcare consultations with healthcare professionals in the relevant fields. However, there is currently a lack of reports on CIS providing appropriate treatment recommendations and / or suggested procedures for detected lesions / abnormalities. In particular, there is a significant lack of recommendations regarding the health status of specific sites within tissue structures. Such recommendations are especially crucial in situations of limited resources, inability to obtain immediate expert interpretation, concerns about healthcare costs, and / or limited access to care. The revealed CIS can automatically provide care recommendations for specific sites within local tissue structures (such as oral care, epidermal care, or mucosal care) after detecting inflammation. The recommendations provided contain clear and accurate instructions on how to care for specific sites of inflamed tissue structures (such as gums, epidermis, or mucosa).

[0027] Additionally, a machine learning algorithm is described that operates in conjunction with a processor, a GUI, or a combination thereof. In this paper, a processor is a hardware component of a device used to execute instructions and perform calculations to run programs and the device's operating system. Such devices typically acquire, store, analyze, process, and / or output data or other information electronically, including but not limited to smartphones, laptops, desktop computers, mainframe computers, watches, and tablets. The machine learning algorithm includes a generative AI platform configured to process data on local tissue structures (such as data based on intraoral images, intraoral text data, or a combination thereof; data based on epidermal images, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof). The output of the machine learning algorithm may be selectively sent to a GUI.

[0028] The revealed generative AI platform can be trained on various types of data, such as images, text, or combinations thereof. The platform operates by first training on a suitably sized set of relevant datasets to generate the data needed to provide recommendations. These datasets can include images, text, or a combination of both. The platform then uses this training data to build a mathematical model that represents the patterns and structures within the data. Once the model is established, the platform can use the learned patterns and structures to produce outputs similar to the original dataset, thereby generating health care recommendations.

[0029] In some forms, generative machine learning algorithms are trained to provide localized care recommendations for specific tissue structures. Preferably, these recommendations can be for oral care, epidermal care, or mucosal care, more preferably oral care. In some forms, the generative machine learning algorithm is trained and tested based on recommendations from professionals in the relevant fields. For example, in the case of epidermal care, the generative machine learning algorithm is trained and tested based on recommendations from dermatologists; in the case of oral care, it is trained and tested based on recommendations from qualified professionals, including but not limited to dental professionals, periodontologists, etc., preferably dentists. Preferably, the dentist is a periodontologist specializing in periodontal disease or an oral surgeon specializing in oral mucosal diseases.

[0030] When training a generative AI platform, both the generator and the discriminator need to be trained simultaneously. The discriminator's recommendations are provided by healthcare professionals (e.g., dentists, periodontists, dermatologists, etc.) or generated by the generator. During discriminator training, input data (e.g., images of the oral cavity, skin, mucosa) and other information are used, along with real and fake recommendations, to update its parameters. Since the input data is multimodal, it may include encoders and recurrent neural network architectures, enabling the discriminator to make predictions. During the generator's training phase, the generator updates its parameters using error feedback from the discriminator. The generator can perform cross-domain generation; for example, it can take an internal oral cavity photograph and other possible information as input and generate textual recommendations for specific locations as output. Random vectors can also be used when one or more multimodal inputs are unavailable.

[0031] When the local tissue structure is inside the oral cavity, specific care recommendations may include brushing techniques or flossing advice. For example, when there is inflammation of the marginal gingiva, use a soft-bristled toothbrush to brush gently; when there is inflammation of the adjacent gingiva, use an interdental brush and / or floss; recommend quitting smoking; recommend avoiding harmful foods such as betel nut, or a combination of the above recommendations.

[0032] In some forms, current CIS includes a generative AI platform driven by generative machine learning algorithms, which are developed through coding techniques. In some forms, generative machine learning algorithms involve supervised, unsupervised, or semi-supervised algorithms. In some forms, the machine learning module includes neural networks selected from recurrent neural networks, convolutional neural networks, and artificial neural networks. During training, CIS includes a generator and a discriminator. After training, the generator is used to generate suggestions. In a non-restrictive example, the generator can accept the following inputs: (1) an image (e.g., a photograph of the inside of the mouth); and (2) optional additional information, including random vectors. The generator can be a conditional generator that accepts an image (e.g., a photograph of the inside of the mouth) to perform conditional suggestion generation. The generator first performs inflammation segmentation and produces a pixel-level segmented image that shows the severity of inflammation. Then, the segmented image, the input image (e.g., a photograph of the inside of the mouth), and the random vectors are fed into a depth encoder to generate latent vectors. The generator can also detect mucosal changes, such as red or white mucosal lesions. Finally, the suggestions are decoded into text form using neural network architectures such as recurrent neural networks, long short-term memory networks (LSTM), converters, etc.

[0033] In some forms, the generative AI platform is located on a user device terminal, such as a mobile phone (like a smartphone). In other forms, the generative AI platform is located on a server. The server is preferably operatively connected to a GUI (Graphical User Interface), as described above. The GUI can display relevant health care suggestions generated by the generative AI platform. In some forms, the health care suggestions can be displayed on the GUI after local tissue structure data is input into the CIS, with the display time ranging from 24 hours, 18 hours, 12 hours, 6 hours, 2 hours, 1 hour, 30 minutes, 15 minutes, 5 minutes, 1 minute, 30 seconds, 20 seconds, 10 seconds, or 1 second. Preferably, the health care suggestions are displayed within 10 seconds. The GUI can be a digital display screen, such as the display screen of a smartphone, laptop, desktop computer, watch, tablet, etc.

[0034] Preferably, the CIS also includes a discriminative AI platform operatively connected to the generative AI platform. Preferably, the discriminative AI platform is operatively connected to a user device terminal to collect and / or transmit initial data for CIS analysis. In some forms, the discriminative AI platform is located on the user device terminal, such as a mobile phone (e.g., a smartphone) containing the generative AI platform. The discriminative AI platform may reside on the same or a different server as the generative AI platform. Preferably, the discriminative AI platform and the generative AI platform reside on the same server. The discriminative AI platform primarily performs disease detection of local tissue structures (e.g., periodontal tissue structures) based on digitally pixelated photographs. Upon completion, the discriminative AI platform generates data on the local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof; epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof) and transmits it to the generative AI platform. Preferably, the data analyzed by the discriminative AI platform is digitally pixelated photographs, such as those obtained using a mobile phone (like a smartphone) or other handheld device. In the disclosed CIS, the discriminative AI platform integrated therein has undergone clinical testing, achieving an accuracy rate of over 90% in periodontal disease detection, thus enabling accurate detection of periodontal disease. Based on this information and the location of the disease, a generative AI platform developed with advice from dental professionals (such as dentists) can provide personalized, site-specific recommendations to maintain gum cleanliness and / or health.

[0035] In some forms, this machine learning algorithm includes a generative AI platform supported by a generative machine learning algorithm developed through coding techniques. In some forms, the generative machine learning algorithm involves supervised, unsupervised, or semi-supervised algorithms. In some forms, the machine learning module includes neural networks selected from recurrent neural networks, convolutional neural networks, and artificial neural networks. During training, the machine learning algorithm includes a generator and a discriminator. After training, the generator is used to generate suggestions. In a non-restrictive example, the generator can accept the following inputs: (1) an image (e.g., a photograph of the inside of the mouth); and (2) optional other information, including random vectors. The generator can be a conditional generator that accepts an image (e.g., a photograph of the inside of the mouth) to perform conditional suggestion generation. The generator first performs inflammation segmentation and produces a pixel-level segmented image that shows the severity of inflammation. Then, the segmented image, the input image (e.g., a photograph of the inside of the mouth), and the random vector are fed into a depth encoder to generate latent vectors. Finally, the suggestions are decoded into text form using neural network architectures such as recurrent neural networks, long short-term memory networks (LSTM), converters, etc.

[0036] In some forms, the generative AI platform for machine learning algorithms resides on a server. Preferably, the server is operatively connected to a GUI, such as the GUI described above. The GUI can display relevant health care recommendations generated by the generative AI platform. In some forms, the health care recommendations are displayed on the GUI after inputting local tissue structure data, within a time frame that can be within 24 hours, 18 hours, 12 hours, 6 hours, 2 hours, 1 hour, 30 minutes, 15 minutes, 5 minutes, 1 minute, 30 seconds, 20 seconds, or 10 seconds. Preferably, the health care recommendations are displayed within 10 seconds. The GUI can be a digital screen, such as the display screen of a smartphone, laptop, desktop computer, watch, tablet, etc.

[0037] Preferably, the machine learning algorithm further includes a discriminative AI platform and is operatively connected to the generative AI platform. The discriminative AI platform is preferably operatively connected to a user device terminal for collecting and / or transmitting initial data for analysis by the machine learning algorithm. The discriminative AI platform may reside on the same or different servers containing the generative AI platform. Preferably, the discriminative AI platform and the generative AI platform reside on the same server or the same user device terminal (e.g., a mobile phone, such as a smartphone). The discriminative AI platform primarily performs disease detection of local tissue structures (e.g., periodontal anatomy) in specific locations based on digitally pixelated photographs. Upon completion, the discriminative AI platform generates data on the local tissue structure (e.g., intraoral image data, intraoral text data, or a combination thereof; epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof) and transmits it to the generative AI platform. Preferably, the data analyzed by the discriminative AI platform is digitally pixelated photographs, such as photographs obtained using a mobile phone (e.g., a smartphone) or other handheld device. For discriminative AI platforms that can be integrated into the disclosed machine learning algorithms, clinical testing has shown an accuracy rate exceeding 90% in detecting gum disease. Therefore, discriminative AI platforms can accurately detect gum disease. Based on this information and the location of the disease, generative AI platforms developed with advice from dental professionals (such as dentists) can provide personalized, site-specific recommendations to maintain gum cleanliness and / or health.

[0038] The content of a non-restrictive discriminative AI platform for automating multi-level gingival disease detection in specific sites is incorporated herein by reference in its entirety in GH Li et al., 15th International Symposium on Medical Information and Communication Technologies (ISMICT) 2021.

[0039] The architecture and training content used for generation:

[0040] Personalized oral hygiene recommendations for specific areas:

[0041] This paper proposes a non-limiting technical solution for a computer-implemented system with AI capabilities, which can consist of two stages to provide oral hygiene advice: 1. The first stage involves detecting mild gingival inflammation and its condition, and the output can be multi-level segmentation results, preferably at the pixel level; 2. A high-order generative suggestion network for inferring suggestions (preferably comprehensive suggestions) from the input gingivitis detection subsystem. Preferably, the generative suggestion network infers suggestions under the guidance of an object detection module and the user's medical and socio-demographic data (e.g., age, ethnicity, diabetes status, smoking status, etc.). It should be understood that the object detection module may involve using one or more neural networks to locate and classify one or more objects in an image.

[0042] a. System Design

[0043] The output of the first stage is a label map with multiple severity levels, used to display diseased, suspicious, or healthy gingival conditions, or a combination thereof, at the pixel level. Preferably, the label map is the same size as the input image. The first stage preferably includes a high-accuracy semantic segmentation module validated by clinical photographs. Semantic segmentation can be implemented using neural network models, such as DeepLab or variants thereof, visual transducers, encoder-decoder (U-Net) networks, etc.

[0044] In the second stage, the object detection module preferably first identifies the gingival margin locations at the free gingival margin (around the teeth) and interdental papillae (located between two teeth). Object detection can be achieved using techniques such as YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), RetinaNet, Faster R-CNN, Mask R-CNN, Cascade R-CNN, DETR (Detection Transformer), and Swin Transformer. The detected region of interest (ROI) proposals can be used as visual embeddings, which, along with the gingival condition data from the first stage, are fed into the proposal generation network. Medical and socio-demographic data can also be encoded as embeddings, for example, using one-hot encoding or learned embeddings to generate "demographic embeddings." These combined embeddings can be aligned to a shared space using multimodal models (such as VisualBERT and LXMERT) and used as input to a generative inference model to generate natural language inference content. This natural language inference includes one or more site-specific oral hygiene guidelines. The suggestions provided may include direction (such as vertical, horizontal, and / or mixed), duration, pattern / style, speed, and brushing pressure level. The suggestions may also include recommendations for dental tools, such as floss type, toothbrush, toothpick, toothpaste, etc. The system may also provide further explanations of the suggestions.

[0045] b. Data collection and preprocessing

[0046] Intraoral photographs with annotations of accurate diagnoses of gingival health from medical professionals (e.g., professional dentists) can be collected. Natural language descriptions or annotations corresponding to segmented and examined regions can be collected or generated. Preferably, these annotations describe the relationship between gingival health and its causes in the images, as well as oral hygiene guidelines for specific areas.

[0047] c. Image segmentation and object detection

[0048] Pre-trained models were used to perform image segmentation for gingival inflammation detection and object detection at gingival margin locations on the dataset. These models can generate region of interest (ROI) proposals and pixel-level segmentation of gingival health status, i.e., generate bounding boxes and segmentation results with location category labels.

[0049] d. Model training

[0050] In some forms, Natural Language Inference (NLI) models can be trained that take location and image embeddings as input features and predict the relationships between objects described in the text embeddings. This model can be based on architectures such as CLIP (Contrastive Language–Image Pretraining), VisualBERT, LXMERT (Learning representations of cross-modal encoders from transducers), LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit) networks, or transducer-based models. Pre-trained networks, various master architectures, and data augmentation techniques can be employed to increase the diversity of training data.

[0051] e. System Architecture Diagram

[0052] Figure 2 illustrates, without limitation, a CIS with AI capabilities. In this non-limiting example, the workflow is as follows: (e1) Population statistics input → Population statistics embedding (e2) Intraoral image → Segmentation result → Image encoder (ResNet / Visual Transformer) → Visual embedding #1 (e3) Intraoral cavity photograph → Target detection results → Visual embedding #2 (e4) Demographic embedding, visual embedding, combined embedding (e5) Output: Inference / Description

[0053] Different aspects of the system (e1 through e5, or subsets of each aspect) can be executed in series, in parallel, or a combination of both. In some forms, (e1), (e2), and (e3) are executed in parallel. In some forms, (e1), (e2), and (e3) are executed in series in an unspecified order. Preferably, (e1), (e2), and (e3) are executed before (e4) and (e5). Preferably, (e4) is executed before (e5).

[0054] The system can include various selections from (e1) to (e5), such as: (e1); (e2); (e3); (e1) and (e4); (e2) and (e4); (e3) and (e4); (e1), (e4) and (e5); (e1), (e2) and (e3); (e2), (e4) and (e5); (e3), (e4) and (e5); (e1), (e2), (e3) and (e4); (e1), (e2), (e3) and (e5); (e1), (e2), (e3), (e4) and (e5), etc.

[0055] Figure 3 illustrates a non-limiting example of personalized oral hygiene recommendations for specific sites marked at inflamed gingiva, marked by the intermediate layer between the teeth and the rest of the gums. At the interdental papilla, the oral hygiene recommendation is to clean between the teeth with a toothbrush; at the free gingival margin, the oral hygiene recommendation is to brush gently with a force of 10 N.

[0056] [ii.] [Computer Implementation Method]

[0057] A computer-implemented method (CIM) is also described herein for generating health care recommendations for specific structural tissues, covering the care of local structural tissues (such as oral hygiene care, epidermal care, mucosal care, etc.), and involving the use of any of the aforementioned CIS. Preferably, CIM involves (i) analyzing tissue structure data (such as data based on intraoral images, data based on intraoral text, or a combination of both; epidermal image data, epidermal text data, or a combination of both; mucosal image data, mucosal text data, or a combination of both); and (ii) displaying care recommendations for specific tissue structures on a graphical user interface (GUI) based on the analysis in step (i). Preferably, steps (i) and / or (ii) are performed using any of the aforementioned CIS.

[0058] In a preferred embodiment, before the CIS transmits local tissue structure data to the aforementioned generative AI platform, the CIS further involves analyzing digitally pixelated photographs and then generating image data based on the oral cavity, text data based on the oral cavity, or a combination of both; or image data based on the epidermis, text data based on the epidermis, or a combination of both; or image data based on the mucosa, text data based on the mucosa, or a combination of both. In these embodiments, the image data based on the oral cavity, text data based on the oral cavity, or a combination of both; the image data based on the epidermis, text data based on the epidermis, or a combination of both; or the image data based on the mucosa, text data based on the mucosa, or a combination of both, is generated by the distinguishing AI platform described above.

[0059] [three] [.] [How to use]

[0060] The described CIS, CIM, or machine learning algorithms can be used to analyze local tissue structure data and provide healthcare recommendations based on the analysis results. The CIS, CIM, or machine learning algorithms have broad applicability and are not limited to data from patient populations in a specific geographic area. Preferably, the data is image-based, text-based, or a combination of both, such as image data obtained using a mobile phone (e.g., a smartphone) or other handheld device.

[0061] Preferably, CIS, CIM, or machine learning algorithms can be used to provide information about an individual's oral health status and offer medical advice (if necessary) to improve that health. Users can take photos of their front teeth using a mobile phone (e.g., a smartphone) and instant messaging applications (e.g., WhatsApp or WeChat) and upload the photos to the CIS or machine learning algorithm, preferably hosted on a server. Alternatively, users can install a mobile application or access a website and send photos to the CIS or machine learning algorithm. Optionally, the CIS, CIM, or machine learning algorithm can include technical means to perform quality checks on the photos and, if necessary, guide the user to retake the photos. Based on the photos, the CIS, CIM, or machine learning algorithm can identify local tissue structures (preferably, gingiva) and then provide feedback. Feedback can include labels such as (i) healthy, (ii) diseased, and (iii) suspicious, targeting specific areas of the tissue structure, preferably specific areas of the gingiva. Based on the condition and location of the selected area (e.g., gingiva), the CIS, CIM, or machine learning algorithm can provide suggestions on how to improve one or more tissue structure areas. For example, if the gums are selected, the suggestions provided may include cleaning the area using a toothbrush, dental floss, interdental brush, etc. Preferably, the suggestions provided may be image-based (e.g., based on a photograph), personalized to meet individual needs, and / or supported by a generative AI platform.

[0062] In the non-restrictive case of gingivitis, the disease pattern of periodontitis is site-specific because certain sites are more susceptible to periodontitis than others. Therefore, oral hygiene care recommendations generated using revealed CIS, CIM, or machine learning algorithms, along with personalized approaches and / or highlighting specific diseased sites, are routine practices. Furthermore, individuals clearly understand how to proceed and can easily follow the recommendations, thus enhancing the effectiveness of self-care. Individual users simply follow the recommendations generated by CIS, CIM, or machine learning algorithms to care for inflamed gingivitis, and preferably, can reverse identified disease / conditions to a healthy state. Additionally, individual users can regularly take and upload images to check the progress of their gingival health and compare feedback with previously generated recommendations.

[0063] The disclosed CIM, CIS, or machine learning algorithms provide a new platform that leverages AI (preferably automated and real-time) and data (preferably photos) generated using mobile phones (e.g., smartphones) to offer personalized local tissue health care recommendations (e.g., oral hygiene, epidermal care, mucosal care, etc.) for diseased tissue structures. In particular, personalized oral hygiene care recommendations based on the individual user's gingival condition can provide a more ideal solution within oral health promotion strategies. Therefore, applying artificial intelligence to health advice offers an easily accessible approach and becomes an expert medical assistant in daily life.

[0064] The disclosed CIS, CIM, and machine learning algorithms can be further understood through the following paragraphs or examples. 1. A computer-implemented system (CIS) comprising a generative artificial intelligence (AI) platform operablely connected to a graphical user interface (GUI), wherein the platform is used to: (i) transmit health recommendations (e.g., oral hygiene care, epidermal care, mucosal care, etc.) for specific sites of tissue structures to the GUI based on data of local tissue structures (e.g., oral cavity image data, oral cavity text data, or combinations thereof; epidermal image data, epidermal text data, or combinations thereof; or mucosal image data, mucosal text data, or combinations thereof). 2. The CIS as described in paragraph 1, wherein the generative AI platform includes a generative machine learning algorithm. 3. The CIS as described in paragraph 2, wherein a generative machine learning algorithm is trained to generate health recommendations (e.g., oral hygiene care, epidermal care, mucosal care, etc.) for specific parts of the tissue structure. 4. The CIS according to paragraph 2 or 3, wherein the generative machine learning algorithm evaluates data on local tissue structures (e.g., intraoral image data, intraoral text data or a combination thereof; or epidermal image data, epidermal text data or a combination thereof; or mucosal image data, mucosal text data or a combination thereof). 5. The CIS according to any one of paragraphs 1 to 4, wherein the data on local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof) contains results of periodontal health. 6. The CIS according to any one of paragraphs 1 to 5, wherein the data on local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof) contains periodontal disease information for a specific site. 7. The CIS as described in paragraph 6, wherein periodontal disease information for a specific site involves marginal gingiva, attached gingiva, interdental gingiva, oral mucosa, alveolar mucosa, or a combination thereof. 8. The CIS according to any one of paragraphs 1 to 7, wherein the generative AI model is trained using an image data type, a text data type, or a combination thereof. 9. The CIS according to any one of paragraphs 1 to 8, wherein the generative AI model is trained and tested based on recommendations from dental professionals, preferably dentists. 10. The CIS according to any one of paragraphs 1 to 9, wherein the generative AI platform includes a convolutional neural network, a fully connected network, an encoder, a decoder, an LSTM, a converter, or an attention module. 11. The CIS as described in any one of paragraphs 1 through 10, wherein recommendations for specific sites of tissue structure include brushing techniques or flossing recommendations involving gentle brushing with a soft-bristled toothbrush (e.g., when inflammation is at the marginal gingiva), interdental brushing (e.g., when inflammation is at the interdental gingiva), or a combination thereof. 12. The CIS according to any one of paragraphs 1 to 11, wherein the generative AI platform is located on a server or on an end-user device. 13. The CIS according to any one of paragraphs 1 to 12 further includes a discriminative AI platform operablely connected to the generative AI platform. 14. The CIS as described in paragraph 13, wherein the discriminative AI platform preferably performs disease detection of local tissue structures (e.g., periodontal tissue structures) in a specific location based on digitally pixelated photographs. 15. The CIS as described in paragraph 14, wherein the discriminative AI platform generates data on local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof) after analyzing digitally pixelated photographs. 16. The CIS according to any one of paragraphs 13 to 15, wherein the discriminative AI platform is located on (i) a server that is the same as or different from the generative AI platform, preferably the same server, or on (ii) an end-user device that includes the generative AI platform. 17. The CIS according to any one of paragraphs 1 to 16, wherein the GUI is a digital screen, such as the screen of a smartphone, laptop, desktop computer, watch, tablet, etc. 18. A machine learning algorithm operably connected to a processor, a GUI, or a combination thereof, wherein the machine learning algorithm includes a generative AI platform configured to process data of local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof; epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof), optionally wherein the output of the machine learning algorithm can be transmitted to the GUI. 19. The machine learning algorithm according to Clause 18, wherein the generative AI platform includes a generative machine learning algorithm. 20. The machine learning algorithm described in paragraph 19, wherein the generative machine learning algorithm is trained to generate health recommendations (e.g., oral hygiene care, epidermal care, mucosal care, etc.) for specific parts of an tissue structure. 21. The machine learning algorithm according to paragraph 19 or 20, wherein the generative machine learning algorithm evaluates data of local tissue structures (e.g., intraoral image data, intraoral text data or a combination thereof; or epidermal image data, epidermal text data or a combination thereof; or mucosal image data, mucosal text data or a combination thereof). 22. The machine learning algorithm according to any one of paragraphs 18 to 21, wherein the data on local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof) contains results on periodontal health. 23. The machine learning algorithm according to any one of paragraphs 18 to 22, wherein the data on local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof) contains periodontal disease information for a specific site. 24. The machine learning algorithm described in paragraph 23, wherein periodontal disease information at a specific site involves marginal gingiva, attached gingiva, interdental gingiva, oral mucosa, alveolar mucosa, or a combination thereof. 25. The machine learning algorithm according to any one of paragraphs 18 to 24, wherein the generative AI platform is trained using an image data type, a text data type, or a combination thereof. 26. The machine learning algorithm according to any one of paragraphs 18 to 25, wherein the generative AI platform is trained and tested based on recommendations from dental professionals, preferably dentists. 27. The machine learning algorithm according to any one of paragraphs 18 to 26, wherein the generative AI platform includes a convolutional neural network, a fully connected network, an encoder, a decoder, an LSTM, a converter, or an attention module. 28. The machine learning algorithm according to any one of paragraphs 18 through 27, wherein suggestions for specific sites of tissue structure include brushing techniques or flossing recommendations involving gentle brushing with a soft-bristled toothbrush (e.g., when inflammation is at the marginal gingiva), interdental brushing (e.g., when inflammation is at the interdental gingiva), or a combination thereof. 30. The machine learning algorithm according to any one of paragraphs 18 to 29 further includes a discriminative AI platform operablely connected to the generative AI algorithm. 31. The machine learning algorithm described in paragraph 30, wherein the discriminative AI algorithm is selected based on digitally pixelated photographs to perform disease detection of local tissue structures in a specific location (e.g., periodontal tissue structures). 32. The machine learning algorithm described in paragraph 30 or 31, wherein the discriminative AI platform generates data on local tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof) after analyzing a digitally pixelated photograph. 33. The machine learning algorithm according to any one of paragraphs 30 to 32, wherein the discriminative AI platform generates local tissue structure data (e.g., intraoral image data, intraoral text data, or a combination thereof) after analyzing a digitally pixelated photograph. 34. The machine learning algorithm according to any one of paragraphs 30 to 33, wherein the discriminative AI platform is set on the same or different servers as the generative AI platform, preferably the same servers. 35. The machine learning algorithm according to any one of paragraphs 18 to 34, wherein the GUI is a digital screen, such as the screen of a smartphone, laptop, desktop computer, watch, tablet, etc. 36. A computer-implemented method (CIM) for generating health care recommendations (e.g., oral hygiene care, epidermal care, mucosal care, etc.) for specific sites of tissue structures, wherein the CIM comprises: (i) analyzing data on tissue structures (e.g., intraoral image data, intraoral text data, or a combination thereof; epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof), and (ii) displaying health care recommendations (e.g., intraoral hygiene care, epidermal care, mucosal care, etc.) for specific sites of tissue structures on a graphical user interface (GUI) based on the analysis results of step (i). Preferably, steps (i) and / or (ii) are performed by a CIS as described in any one of paragraphs 1 to 17 or a machine learning algorithm as described in any one of paragraphs 18 to 35. 37. The CIM according to paragraph 36 further includes, prior to step (i), generating intraoral image data, intraoral text data, or a combination thereof by analyzing digitally pixelated photographs; epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof. 39. The CIM as described in paragraph 36 or 37, wherein the intraoral image data, intraoral text data, or a combination thereof; the epidermal image data, epidermal text data, or a combination thereof; or the mucosal image data, mucosal text data, or a combination thereof, is generated by a discriminative AI platform.

[0065] Those skilled in the art will recognize or be able to determine many equivalents of the specific embodiments of the invention described herein using only conventional experiments. These equivalents are intended to be covered by the appended claims.

Claims

1. A computer-implemented system (CIS) comprising a generative artificial intelligence (AI) platform operablely connected to a graphical user interface (GUI), wherein the generative AI platform includes a generative machine learning algorithm, and the generative AI platform is configured to: transmit health recommendations for a specific part of a tissue structure to the GUI based on data of a local tissue structure, including at least oral cavity image data, oral cavity text data, or a combination thereof; or epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof; wherein the health recommendations transmitted to the GUI are generated by the generative machine learning algorithm, which is trained and used to decode latent vectors formed by the data of the local tissue structure into text form via a neural network architecture to generate the health recommendations for the specific part of the tissue structure, including at least oral cavity hygiene care, epidermal care, mucosal care, or a combination thereof.

2. The CIS as claimed in claim 1, wherein the generative machine learning algorithm evaluates data on local tissue structures, which includes at least intraoral image data, intraoral text data, or a combination thereof, or epidermal image data, epidermal text data, or a combination thereof, or mucosal image data, mucosal text data, or a combination thereof.

3. The CIS as described in claim 2, wherein the data on local tissue structures includes at least intraoral image data, intraoral text data, or a combination thereof, and includes results on periodontal health.

4. The CIS as described in claim 3, wherein the data on local tissue structures includes at least, for example, intraoral image data, intraoral text data, or a combination thereof, and contains periodontal disease information for a specific location.

5. The CIS as described in claim 4, wherein periodontal disease information for a specific site relates to marginal gingiva, attached gingiva, interdental gingiva, oral mucosa, alveolar mucosa, or a combination thereof.

6. The CIS as described in claim 1, wherein the generative AI platform is trained using image data types, text data types, or a combination thereof.

7. The CIS as described in claim 1, wherein the generative AI platform is trained and tested based on recommendations from dental professionals, including dentists.

8. The CIS as described in claim 1, wherein the generative AI platform comprises a convolutional neural network, a fully connected network, an encoder, a decoder, an LSTM, a converter, or an attention module.

9. The CIS as claimed in claim 1, wherein recommendations for specific sites of tissue structure include brushing techniques or flossing recommendations relating to gentle brushing with a soft-bristled toothbrush when inflammation is at the marginal gingiva, and interdental brushing when inflammation is at the interdental gingiva, or combinations thereof.

10. The CIS as described in claim 1, wherein the generative AI platform is located on a server or on an end-user device.

11. The CIS as described in any of claims 1 further includes a discriminative AI platform operablely connected to the generative AI platform.

12. The CIS as described in claim 11, wherein the discriminative AI platform performs disease detection of local tissue structures in specific locations based on digitally pixelated photographs, including at least periodontal tissue structures.

13. The CIS as described in claim 12, wherein the discriminative AI platform generates data on local tissue structures after analyzing digitally pixelated photographs, including at least intraoral image data, intraoral text data, or a combination thereof.

14. The CIS as described in claim 11, wherein the discriminative AI platform is located on (i) a server that is the same as or different from the generative AI platform, or on (ii) an end-user device that includes the generative AI platform.

15. The CIS as claimed in claim 1, wherein the GUI is a digital screen, including at least the screen of a smartphone, laptop, desktop computer, watch, or tablet computer.

16. A machine learning algorithm operably connected to a processor, a GUI, or a combination thereof, wherein the machine learning algorithm includes a generative AI platform configured to process data of a local tissue structure, including at least intraoral image data, intraoral text data, or a combination thereof; or epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof; wherein the output of the machine learning algorithm can be transmitted to the GUI; and wherein the generative AI platform includes a generative machine learning algorithm, the output transmitted to the GUI being generated by the generative machine learning algorithm, the generative machine learning algorithm being trained and used to decode latent vectors formed from the data of the local tissue structure into text form via a neural network architecture to generate health recommendations for a specific part of the tissue structure as output, including at least oral hygiene care, epidermal care, mucosal care, or a combination thereof.

17. The machine learning algorithm of claim 16, wherein the generative machine learning algorithm evaluates data on local tissue structures, which includes at least intraoral image data, intraoral text data or a combination thereof, or epidermal image data, epidermal text data or a combination thereof, or mucosal image data, mucosal text data or a combination thereof.

18. The machine learning algorithm as described in claim 17, wherein the data on local tissue structures includes at least intraoral image data, intraoral text data, or a combination thereof, and includes results on periodontal health.

19. The machine learning algorithm as described in claim 18, wherein the data on local tissue structures includes at least intraoral image data, intraoral text data, or a combination thereof, and contains periodontal disease information for a specific location.

20. The machine learning algorithm as described in claim 19, wherein periodontal disease information at a specific site relates to marginal gingiva, attached gingiva, interdental gingiva, oral mucosa, alveolar mucosa, or a combination thereof.

21. The machine learning algorithm as described in claim 16, wherein the generative AI platform is trained using image data types, text data types, or a combination thereof.

22. The machine learning algorithm as described in claim 16, wherein the generative AI platform is trained and tested based on recommendations from dental professionals, including at least dentists.

23. The machine learning algorithm as described in claim 16, wherein the generative AI platform comprises a convolutional neural network, a fully connected network, an encoder, a decoder, an LSTM, a converter, or an attention module.

24. The machine learning algorithm as described in claim 16, wherein suggestions for specific sites of tissue structure include brushing techniques or flossing recommendations, relating to using a soft-bristled toothbrush for gentle brushing when inflammation is at the marginal gingiva, using an interdental brush for cleaning when inflammation is at the interdental gingiva, or a combination thereof.

25. The machine learning algorithm as described in claim 16 further includes a discriminative AI platform operablely connected to the generative AI algorithm.

26. The machine learning algorithm of claim 25, wherein the discriminative AI algorithm performs disease detection on local tissue structures of a specific site based on digitally pixelated photographs, including at least periodontal tissue structures.

27. The machine learning algorithm as described in claim 26, wherein the discriminative AI platform generates data on local tissue structures after analyzing digitally pixelated photographs, including at least intraoral image data, intraoral text data, or a combination thereof.

28. The machine learning algorithm as described in claim 27, wherein the discriminative AI platform generates data on local tissue structures after analyzing digitally pixelated photographs, including at least intraoral image data, intraoral text data, or a combination thereof.

29. The machine learning algorithm as described in claim 25, wherein the discriminative AI platform is hosted on the same or a different server as the generative AI platform.

30. The machine learning algorithm as described in claim 16, wherein the GUI is a digital screen, including at least the screen of a smartphone, laptop, desktop computer, watch, or tablet computer.

31. A computer-implemented method for generating health care recommendations for local tissue structures, including at least oral hygiene care, epidermal care, mucosal care, or a combination thereof, for a specific location of the tissue structure. CIM), which includes: (i) Data analyzing tissue structures, including at least intraoral image data, intraoral text data or a combination thereof, or epidermal image data, epidermal text data or a combination thereof, or mucosal image data, mucosal text data or a combination thereof; (ii) Based on the analysis results of step (i), display health care recommendations for specific parts of the tissue structure on a graphical user interface (GUI), which includes at least oral hygiene care, epidermal care, mucosal care, or a combination thereof. Preferably, steps (i) and / or (ii) are performed by a CIS as described in any one of claims 1 to 15 or a machine learning algorithm as described in any one of claims 16 to 30.

32. The CIM as claimed in claim 31 further includes, prior to step (i), generating intraoral image data, intraoral text data, or a combination thereof by analyzing digitally pixelated photographs; epidermal image data, epidermal text data, or a combination thereof; or mucosal image data, mucosal text data, or a combination thereof.

33. The CIM as described in claim 32, wherein the intraoral image data, intraoral text data, or a combination thereof; the epidermal image data, epidermal text data, or a combination thereof; or the mucosal image data, mucosal text data, or a combination thereof, is generated by a discriminative AI platform.

Images