Method and system for diagnosing trachomatous ocular diseases using artificial intelligence

Abstract

The present invention relates to a method and system for diagnosing trachomatous ocular diseases using artificial intelligence, the method comprising the steps of: acquiring ocular image data; preprocessing data by cropping an ocular conjunctival region or correcting the brightness and color of an image, and adjusting resolution; fine-tuning a deep learning model to analyze the image, and predicting the presence or absence of trachomatous ocular diseases; and training the deep learning model by using training data and validation data, and evaluating performance.

Description

Method and system for diagnosing trachoma eye disease using artificial intelligence

[0001] The present invention relates to a method and system for diagnosing eye diseases, and more particularly, to a method and system for diagnosing trachoma eye diseases using artificial intelligence.

[0002] Trachoma is a leading cause of blindness worldwide, posing a significant problem, particularly in low-income countries. Currently, trachoma screening is primarily performed through clinical diagnosis and microscopic examination, which are time-consuming and costly processes. These methods require highly trained medical professionals, and the accuracy of the diagnosis can vary depending on subjective judgment.

[0003] In particular, the main symptoms of trachoma eye disease primarily appear in the superior conjunctiva, but this area is difficult to expose and photograph unless trained professionals are present. In contrast, the inferior conjunctiva is easily accessible, allowing anyone to easily pull it with their fingers and photograph it. Therefore, the proposed system is designed to enable screening using the inferior conjunctiva. Trachoma eye disease can cause complications such as entropion due to conjunctival scarring, and the severity of the disease can be assessed by examining the direction of the eyebrows.

[0004] This disease is prevalent primarily in subtropical and African regions and is closely linked to public health concerns. However, the likelihood of an epidemic in Korea is increasing due to climate change caused by global warming. Therefore, preventive measures are necessary, and a reliable diagnostic method and system for trachoma eye disease are urgently needed.

[0005] The present invention is intended to overcome the above-described conventional problems, and the problem to be solved by the present invention is to provide a highly reliable trachoma eye disease diagnosis method and system using artificial intelligence.

[0006] In addition, the purpose is to provide a more convenient method and system for diagnosing trachoma eye disease by utilizing the lower conjunctival area, which is easy to photograph even for non-specialists.

[0007] According to an exemplary embodiment of the present invention, a method for diagnosing trachoma eye disease using artificial intelligence is provided, comprising: a step of acquiring eye image data; a step of preprocessing the data by cropping an eye conjunctiva area or performing brightness and color correction and resolution adjustment of the image; a step of analyzing the image by fine-tuning a deep learning model and predicting the presence or absence of trachoma eye disease; and a step of training the deep learning model using training data and verification data and evaluating its performance.

[0008] The above data preprocessing step is characterized by screening through the lower conjunctival area.

[0009] It is characterized by further including a step of evaluating the severity of the disease through an eyebrow direction image.

[0010] The above deep learning model is characterized by being a ViT (Vision Transformer).

[0011]

[0012] The present invention has the following technical advantages and differences compared to existing trachoma screening methods.

[0013] It provides high accuracy and efficiency by applying the latest AI technology, ViT, and it is excellent in accessibility and convenience as it can use images taken with various devices such as smartphones.

[0014] To overcome the difficulty of photographing the superior conjunctival area, screening through the inferior conjunctival area is possible, and the severity of the disease can be assessed through images of the eyebrow direction.

[0015]

[0016] Figure 1 is a diagram showing an eye image preprocessing technique.

[0017] Figure 2 is a structural diagram of an AI system for diagnosing trachoma eye disease according to one embodiment of the present invention.

[0018] Figure 3 is a diagram showing an actual case of reading image data.

[0019] Figure 4 is a photograph showing the eye area for screening the lower conjunctival area.

[0020] Figure 5 is a photograph showing an example of entropion symptoms due to trachoma eye disease.

[0021]

[0022] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

[0023] Figure 1 is a diagram showing an eye image preprocessing technique.

[0024] Looking at the image preprocessing process with reference to Figure 1, image preprocessing is performed by obtaining original data, performing data labeling processing, and extracting and refining only the necessary parts.

[0025] - Adjust the image format to png and proceed with the resize process (512 * 512)

[0026] - Image segmentation process

[0027] - The process of building an image segmentation model based on labeled images.

[0028] - A process of extracting and refining only the parts for learning and reading.

[0029]

[0030] FIG. 2 is a diagram of an AI structure for diagnosing trachoma eye disease according to one embodiment of the present invention, FIG. 3 is a diagram showing an actual case of reading image data, FIG. 4 is a photograph showing an eye area for screening the lower conjunctival area, and FIG. 5 is a photograph showing an example of entropion symptoms due to trachoma eye disease.

[0031] Referring to Fig. 2, a method for analyzing an eye image and diagnosing trachoma eye disease by utilizing the ViT architecture is shown.

[0032] - Learning normal / abnormal (disease) image data through the VIT model

[0033] - Detection of AI reading results for input images through learned models

[0034]

[0035] A more detailed diagnosis process is as follows:

[0036] 1) Acquire image data

[0037] The proposed system uses ocular images captured with a smartphone or other portable device to screen for trachoma. This provides an affordable and accessible solution, with the advantage of allowing patients to take their own images and receive a remote diagnosis.

[0038]

[0039] 2) Data preprocessing

[0040] Image preprocessing includes cropping the conjunctiva area, adjusting the image's brightness and color, and adjusting its resolution. In particular, preprocessing is performed regardless of the upper or lower conjunctiva area, making it easy for even the average person to capture the image.

[0041]

[0042] 3) Data augmentation

[0043] We use data augmentation techniques to learn from images captured in a variety of environments, which helps the model generalize well to changes in different conditions.

[0044]

[0045] 4) Model architecture

[0046] Using fine-tuned, state-of-the-art deep learning models like ViT (Vision Transformer), it analyzes images and predicts the presence of trachoma. This model learns based on image patches and boasts high accuracy and efficiency.

[0047]

[0048] 5) Training and validation

[0049] The model is trained using training and validation data, and methods such as hyperparameter tuning and cross-validation are used to evaluate performance. This achieves optimal model performance.

[0050]

[0051] 6) Evaluation criteria

[0052] To evaluate the performance of a model, metrics such as accuracy, precision, recall, and F1 score are used, which provide information about how well the model predicts the disease and what the error rate is.

[0053]

[0054] 7) Additional features

[0055] Future features that may be added include utilizing eyebrow orientation images to assess complications of diseases such as entropion. This feature could help more accurately assess the severity of the condition.

[0056]

[0057] The present invention can be utilized in various medical fields as follows.

[0058] - Patient screening: Screening large populations at low cost.

[0059] - Early diagnosis of disease: Early treatment can be achieved through detection of early symptoms.

[0060] - Treatment progress monitoring: Monitor the progression of the disease and evaluate the effectiveness of treatment.

[0061]

[0062] The superior conjunctiva is a relatively hard tissue, making it difficult for the average person to image. This AI software is designed to read data from the relatively soft inferior conjunctiva, making it easy for the average person to use.

[0063] Specifically, this system, implemented in the form of a smartphone application, allows patients to directly check for infection in themselves or those around them. It can also be linked to remote medical systems to aid in diagnosis and treatment planning. This can prevent blindness and, through early and appropriate treatment, halt the progression of symptoms.

[0064]

[0065] Referring to Figure 5, the severity of the disease can be additionally assessed through eyebrow direction image analysis.

[0066] As the severity of trachoma eye disease worsens, symptoms such as entropion, where the eyebrows are rolled in due to scarring of the conjunctiva, can occur. The severity of trachoma eye disease can be determined just by the direction of the eyebrows.

[0067]

[0068] The above description is merely an exemplary embodiment of the trachoma eye disease diagnosis method and system using artificial intelligence according to the present invention, and the present invention is not limited to the above-described embodiment, and as claimed in the following claims, it will be understood that the technical spirit of the present invention exists to the extent that various modifications can be implemented by anyone having ordinary skill in the art to which the present invention pertains without departing from the gist of the present invention.

Claims

1. A method for diagnosing trachoma eye disease using artificial intelligence. Step of acquiring eye image data; A step of data preprocessing by cropping the conjunctival area of ​​the eye or adjusting the brightness and color of the image and the resolution; A step of analyzing an image and predicting the presence of trachoma eye disease by fine-tuning a deep learning model; and A method for diagnosing trachoma eye disease using artificial intelligence, characterized by comprising a step of training the deep learning model using training data and verification data and evaluating its performance.

2. In paragraph 1, A method for diagnosing trachoma eye disease using artificial intelligence, characterized in that the above data preprocessing step involves screening through the lower conjunctival area.

3. In paragraph 1, A method for diagnosing trachoma eye disease using artificial intelligence, characterized in that it further includes a step of evaluating the severity of the disease through an eyebrow direction image.

4. In paragraph 1, A method for diagnosing trachoma eye disease using artificial intelligence, characterized in that the above deep learning model is ViT (Vision Transformer).

Images