Researchers Introduce AI-Doctor - A Game-Changer in Retinal Disease Diagnosis

Researchers introduced the AI-Doctor system, an advanced artificial intelligence (AI) system meticulously designed to interpret fundus fluorescein angiography (FFA) images with precision.

This system plays a crucial role in ensuring accurate diagnoses and providing treatment recommendations for a variety of Ischemic Retinal Diseases (IRDs).

The Study

During the study, skilled technicians used the Spectralis Heidelberg Retina Angiograph + Optical Coherence Tomography (HRA+OCT) to capture FFA images from various hospitals, following a standardized image capture protocol. These images were then categorized into different phases for both clinical and research purposes.

Certified ophthalmologists with comprehensive training were responsible for classifying the images, ensuring te accuracy of phase identification and IRD diagnosis.

The segmentation task focused on developing a specialized model for Branch Retinal Vein Occlusion (BRVO) and Diabetic Retinopathy (DR) with non-perfusion areas (NPAs). This model relied heavily on expert annotations and refinements to achieve precise results.

To assess the model's adaptability, an evaluation was conducted using images of other IRDs, including Central Retinal Vein Occlusion (CRVO) and retinal vasculitis.

The AI-Doctor system comprised two models, one for classification and another for segmentation, utilizing cutting-edge convolutional neural networks, Unet, and ResNet-152.

To enhance clinical applicability, the study introduced a Clinically Applicable Ischemia Index (CAII), aiding in the identification of higher-risk patients by using FFA images from different angles and precise criteria for image selection and CAII calculations.

Initially, specialists categorized eyes into groups based on their need for laser therapy, leveraging the characteristics of FFA images, medical records, and clinical expertise.

CAII values for each eye within these groups were determined through the segmentation model, subsequently serving as thresholds for recommending laser therapy. Sensitivity and specificity were analyzed under these thresholds to determine the optimal threshold, further validated with an independent external dataset.

Evaluation of both the classification and segmentation models underwent rigorous scrutiny, including statistical tests and performance metrics such as accuracy, recall, and precision. Receiver Operating Characteristic (ROC) analysis and calculations of areas under the ROC curve were also conducted.

For segmentation assessment, metrics like the Dice similarity coefficient (DSC), Intersection over Union (IoU) value, and F1 score were employed. Empirical ROC curves were examined to establish optimal CAII thresholds for laser therapy recommendations.

The Findings

In this research, the AI-Doctor system was meticulously developed and validated using a dataset of 24,316 images. Distinct datasets were utilized for training, internal validation at the Zhongshan Ophthalmic Center (ZOC), and external tests at the Shenzhen Eye Hospital (SEH) and Foshan Second People's Hospital (FSPH).

An additional set of 1,295 images was dedicated to refining the segmentation model, with a focus on Non-Perfusion Areas (NPAs) and areas affected by branch retinal vein occlusion (BRVO).

AI-Doctor exhibited promising results, displaying exceptional precision, recall, and accuracy across multiple datasets for identifying image phases and diagnosing common Ischemic Retinal Diseases (IRDs). Its diagnostic capabilities aligned with those of expert ophthalmologists.

The system effectively employed heatmap visualization to enhance the precise identification of various conditions, such as diabetic retinopathy, emphasizing significant features in the images, especially in cases of potential misclassification

In terms of segmentation, the research compared the performance of Unet-VGG16 and Unet-Swin Transformer models, with the former demonstrating superior performance and robustness in segmenting NPAs and BRVO-affected areas, both in internal validation and external test sets.

When trained on a diverse set of images, including those featuring diabetic retinopathy and BRVO, AI-Doctor's segmentation model exhibited enhanced versatility and applicability, yielding high Dice similarity coefficients even in previously unencountered conditions.

The study introduced a clinically applicable ischemia index (CAII), derived from the segmentation results of FFA images captured with a 55°-viewing field.

Optimal CAII thresholds were established, demonstrating high sensitivity and specificity in determining the need for laser therapy. Subsequent validations reinforced the reliability of these thresholds in classifying diverse conditions, such as BRVO and diabetic retinopathy.

AI-Doctor significantly enhances clinical utility by generating comprehensive AI reports for FFA image interpretations, maintaining efficiency by completing the entire process from image phase identification to ischemic area segmentation in approximately eight seconds. This efficiency underscores AI-Doctor's potential for rapid implementation in clinical settings.

This research underscores the transformative potential of AI-Doctor in ophthalmological diagnostics, substantiating its reliability, precision, and versatility in medical image analysis.

The AI system's ability to provide detailed insights and its rapid processing time represent a pivotal advancement, with the potential to reshape diagnostic approaches in ophthalmology and underscore its significance in medical image interpretation.

AI-Doctor's detailed, rapid, and precise interpretations mark a transformative progression in medical diagnostics, particularly in ophthalmology, where it combines efficiency, reliability, and extensive applicability.

The implications of this research are profound, emphasizing the potential of advanced AI in revolutionizing medical diagnostics through detailed, efficient, and precise interpretations, paving the way for groundbreaking advancements in ophthalmology.

Reference

Xinyu Zhao,Zhenzhe Lin,Shanshan Yu,Jun Xiao, et Al. An artificial intelligence system for the whole process from diagnosis to treatment suggestion of ischemic retinal diseases. Cell Reports Medicine. 2023.